JP4501948B2 - Information processing apparatus, information processing method, recording medium, and program - Google Patents

Description

  The present invention relates to an information processing apparatus, an information processing method, a recording medium, and a program. In particular, for a user who watches a program such as a television broadcast or a radio broadcast, or streaming data, for example, an automatic recording or a program. The present invention relates to an information processing device, an information processing method, a recording medium, and a program, which are suitable for use in selecting an item.

  Conventionally, for example, when a program is recommended to a user in television broadcasting or radio broadcasting, the user's information is based on program information (program metadata) such as an electronic program guide (EPG). A program matching the preference information was selected. The method of recommending a program to the user differs depending on the method of acquiring user preference data. For example, an initial interest registration method for recommending a program on the basis of information on the user's interest and registering the user based on the information. However, there have been methods such as a viewing history utilization method for recommending programs using the history of programs viewed in the past, or a collaborative filtering method for recommending programs using the viewing history of other users.

  In the initial interest registration method, at the start of use, for example, a favorite program category (for example, drama, variety, etc.), a favorite genre (inference, comedy, etc.), or a favorite talent name is registered with the user. By using this information as a keyword and matching with program metadata, the name of the program to be recommended is acquired.

  Also, in the viewing history usage method, each time a user views a program, the viewed program metadata is accumulated, and when the history metadata is accumulated to some extent, the metadata is analyzed, for example, as desired Information such as program category, favorite genre, or favorite talent name is acquired. A program name to be recommended is acquired by matching the program metadata with these information as keywords.

  For example, in a recording device using an HDD (Hard Disk Drive), user operations such as recording reservation or recording as well as viewing history are accumulated as history information and may be used for obtaining preference information. . In this case, the user is not interested in watching the program with particular interest, but the user is more interested in the program that was being watched because the television receiver or radio was on. It is possible to discriminate a part of programs that have been consciously viewed and to acquire information more reflecting the user's preferences.

  In the collaborative filtering method, the first user's viewing history is similar to the first user's viewing history by matching the viewing history of the first user with the viewing history of other users. The user is searched and the history data of viewing or operation is acquired, so that programs that the second user has watched are extracted and recommended by the first user.

  In addition, an n-dimensional attribute vector is added to the broadcast program in advance as the attribute information of the program, and the initial registration contents of the user and the average value of each attribute item of the attribute vector of the program reproduced or reserved for recording by the user are obtained. There is a technique in which a program to be recorded or a program to be played back can be selected by comparing a selection vector generated based on an attribute vector (for example, Patent Document 1).

JP 2001-160955 A

  However, when selecting a program based on the initial registration method, only a fixed interest at the time when the user performed the initial registration can be reflected, and the user's registration operation becomes complicated in order to obtain detailed information. End up. On the other hand, if the number of registered information is reduced in order to simplify the initial registration information input operation, the user can only make recommendations based on user preference information. The accuracy of selecting a program that matches the preference will be low.

  On the other hand, when a program is selected using a sum total or an average of metadata collected based on a user's viewing history or the like, such as a history usage method, the history is not accumulated to some extent, A program that correctly matches the user's preferences cannot be recommended. Furthermore, in the history usage method, the relationship between metadata is lost, and sufficient personalization cannot be achieved. Also, by accumulating histories, for example, items that are easy to overlap as histories such as genres (items that are easy to detect as user-preferred) and items that are easy to expand as histories such as performers (detected as user-preferred) In some cases, the weighting may be biased.

  Specifically, for example, since the user is a fan of the commentator A, when he likes to watch “the live broadcast of the game of the team B explained by the commentator A”, the genre is “ The information “baseball live coverage” is easy to overlap as a history (the information “baseball live coverage” is easy to detect as a user's preference), but the information “caster A” who is a performer overlaps as history. It is difficult (information “explanator A” is difficult to detect as a user's preference). For this reason, there may be a case where the live broadcast of the game of the team B explained by other commentators is recommended and the variety program in which the commentator A appears is not recommended.

  In addition, as shown in Patent Document 1, an attribute vector of a program is added to a broadcast program in advance, and the average of each attribute item of the user's initial registration content and the attribute vector of the program that the user has played or reserved for recording Since the user's operation history is also used when selecting the program to be recorded or the program to be played back by comparing the selection vector generated based on the value with the attribute vector, the same applies. In addition, there may be a bias in weighting between items that tend to overlap as histories and items that tend to spread elements as histories such as performers.

  Further, for example, in a selection vector generated for a user who likes a drama and prefers only those in which comedian A, who rarely appears in dramas, appears, and the audience ratio of variety to drama is 2: 8. Actor B who frequently appears in dramas overlaps as a history, even though he is not a favorite actor, than comedian A, who hardly appears in dramas. Therefore, a documentary in which performer B who frequently appears in dramas appears is preferentially recommended over the variety in which comedian A appears.

  In addition, even when important items for selecting a program are different depending on the user (for example, emphasis on performers or content), since all items are calculated in the same way, user-specific preferences are not reflected. was there.

  Further, since what is used in the collaborative filtering method is the preference information of other users to the last, it is difficult to extract information that represents each user's preference in detail.

  The present invention has been made in view of such a situation, and makes it possible to select a program that matches a user's preference.

An information processing apparatus according to a first aspect of the present invention is an information processing apparatus that performs a process of selecting content similar to a user's preference, and includes the attribute information of the content including a plurality of items User history information comprising receiving means for receiving from the device, and positive history information that is a history of content that the user likes and negative history information that is a history of content that the user does not like from the operation history of the user's content User history information generating means for generating the content attribute information, the positive history information and the negative history information are vectorized for each item, and the cosine distance between the content attribute information and the user history information vector The inner product with the weighting information is calculated for each item and the result of the inner product is added to all the items. And calculating means for calculating the similarity between the received attribute information of the content and the generated user history information, selecting means for selecting content based on the calculated similarity, The weighting information used by the calculation means , the cosine distance between the standard preference vector indicating the general preference of the viewer obtained by counting the attribute information for a predetermined period and the vector of the positive history information for each item And the reciprocal thereof is positive weighting information that defines the degree of positive contribution of each item when calculating the similarity between the attribute information and the positive history information , the standard preference vector, The cosine distance with the negative history information vector is calculated for each item, and the reciprocal of each item in the case of calculating the similarity between the attribute information and the negative history information. Of a weighting information generating means for generating a negative weighting information defining the degree of contribution, the calculating means, the cosine distance between the positive weighting vector information of the attribute information of the content positive history information An inner product is calculated for each item, and the inner product of the cosine distance of the content attribute information, the vector of the negative history information, and the negative weighting information is calculated from the sum of the inner product results for all the items. The similarity between the content attribute information and the user history information is calculated based on the difference obtained by calculating for each item and subtracting the sum of the inner product results for all the items .

The user history information generating unit generates the user history information for each genre, and the weighting information generating unit sets a cosine distance between the standard preference vector and the vector of the positive history information for a predetermined genre. For each item, the reciprocal number is generated as the positive weighting information, and for the predetermined genre, the cosine distance between the standard preference vector and the negative history information vector is calculated for each item. An inverse number is generated as the negative weighting information, and the calculation means calculates an inner product of the attribute information of the content and the cosine distance of the vector of the positive history information of the predetermined genre for each item, and From the sum of the inner product results for all the items, the attribute information of the content and the predetermined genre The attribute information of the content of the predetermined genre is calculated by calculating the inner product of the vector of the negative history information with the cosine distance of each item and subtracting the sum of the results of the inner products for all the items. And the user history information can be calculated .

The receiving unit also receives attribute information of the content to which information indicating a predetermined group when the content is classified into a plurality of groups is added from the other device, and the selection unit is configured to receive the content of the content. When the information indicating the group is added to the attribute information, the content corresponding to the information indicating the group is selected. When the information indicating the group is not added, the similarity calculated by the calculation unit is set. Based on this, the content can be selected .

The content may be a program, and the group may be grouped by any one of a series drama, content genre, performer, and title .

Storage means for storing weighting information corresponding to the item of the attribute information generated by the weighting information generation means can be further provided.

The information processing apparatus may further include an accumulating unit that accumulates an operation history of the user content .

An information processing method according to a first aspect of the present invention is an information processing method of an information processing apparatus that executes a process of selecting content similar to a user's preference, and the information processing apparatus includes a plurality of items. A receiving step of receiving the attribute information of the content from another device, an operation history of the user's content, a positive history information that is a history of the content that the user likes, and a history of the content that the user does not like A user history information generation step for generating user history information including certain negative history information, the attribute information of the content, the positive history information, and the negative history information are vectorized for each item, and the content attribute information The inner product of the cosine distance of the vector of the user history information and the weighting information is calculated for each item, and the inner product The sum obtained by adding all of the items results in the attribute information of the received the contents, a calculation step of calculating a similarity between the generated the user history information, based on the calculated degree of similarity A selection step for selecting content; and the weighting information used in the processing of the calculation step, the standard preference vector indicating the general preference of viewers obtained by counting the attribute information for a predetermined period; and The cosine distance with the vector of the positive history information is calculated for each item, and the reciprocal thereof defines the degree of positive contribution of each item when calculating the similarity between the attribute information and the positive history information. a positive weighting information, and the standard favorite vector, calculates a cosine distance between the vector of the negative history information for each of the items, the inverse, the genus In the case of calculating the similarity between the information and the negative history information, and a weighting information generating step of generating a negative weighting information defining the negative degree of contribution of each item, the processing of the calculation step, the The inner product of the cosine distance of the content attribute information and the vector of the positive history information and the positive weighting information is calculated for each item, and the result of the inner product is added to all the items, and the content of the content is calculated. By calculating the inner product of the cosine distance of the attribute information and the negative history information vector and the negative weighting information for each item, and subtracting the sum of the results of the inner products for all the items, The similarity between content attribute information and the user history information is calculated .

In the processing of the user history information generation step, the user history information is generated for each genre, and in the processing of the weighting information generation step, a cosine distance between the standard preference vector and the vector of the positive history information for a predetermined genre. Is calculated for each item, and the reciprocal thereof is generated as the positive weighting information, and the cosine distance between the standard preference vector and the negative history information vector is calculated for each item for the predetermined genre. The reciprocal is generated as the negative weighting information, and in the processing of the calculating step, an inner product of the attribute information of the content and the cosine distance of the vector of the positive history information of the predetermined genre is calculated for each item. The content is calculated from the sum of the inner product results for all the items. The inner product of the attribute information and the cosine distance of the vector of the negative history information of the predetermined genre is calculated for each item, and the result of the inner product is subtracted from the sum obtained by adding all the items. It is possible to calculate the similarity between the attribute information of the content of the genre and the user history information .

In the process of the receiving step, the attribute information of the content to which information indicating a predetermined group when the content is classified into a plurality of groups is added is also received from the other device. When the information indicating the group is added to the attribute information of the content, the content corresponding to the information indicating the group is selected, and when the information indicating the group is not added, the similarity calculated in the calculation step The content can be selected based on the degree .

The content may be a program, and the group may be grouped by any one of a series drama, content genre, performer, and title .

A storage step of storing weighting information corresponding to the item of the attribute information generated by the processing of the weighting information generation step can be further included.

An accumulating step of accumulating an operation history of the user content can be further included .

A program according to a first aspect of the present invention is a program for causing a computer to execute a process of selecting content similar to a user's preference, wherein the content attribute information composed of a plurality of items User history information consisting of a reception step received from the device, and positive history information that is a history of the content that the user likes and negative history information that is a history of content that the user does not like from the operation history of the user's content Generating the user history information, the content attribute information and the positive history information and the negative history information are vectorized for each item, and the cosine distance between the content attribute information and the user history information vector The inner product with the weighting information is calculated for each item, and the result of the inner product is calculated for all previous items. The sum obtained by adding the serial item, and attribute information of the received the contents, a calculation step of calculating a similarity between the generated the user history information, based on the calculated degree of similarity, select content A standard preference vector indicating general audience preference obtained by counting the attribute information for a predetermined period, and a vector of the positive history information, the weighting information used in the processing of the selection step and the calculation step the cosine distance calculated for each of the items and, its inverse, in the case of calculating the similarity between the attribute information and the positive history information, a positive weighting information defining the degree of positive contribution of each item The cosine distance between the standard preference vector and the negative history information vector is calculated for each item, and the reciprocal is calculated as the attribute information and the default In the case of calculating the similarity between the information, and a weighting information generating step of generating a negative weighting information defining the negative degree of contribution of each item, the processing of the calculation step, the attribute information of the content The inner product of the cosine distance of the vector of the positive history information and the positive weighting information is calculated for each item, and the result of the inner product is added to all the items, and the attribute information of the content and the negative information are calculated. The inner product of the cosine distance of the vector of history information and the negative weighting information is calculated for each item, and the attribute information of the content is obtained by subtracting the sum of the results of the inner product for all the items. Calculate similarity with the user history information
To perform the processing to computer.

In the first aspect of the present invention, the attribute information of the content composed of a plurality of items is received from another device, and the normal history information that is the history of the content preferred by the user from the operation history of the user's content And negative history information, which is a history of content that the user does not like, is generated, the content attribute information and the positive history information and negative history information are vectorized for each item, the content attribute information and the user history The inner product of the cosine distance of the information vector and the weighting information is calculated for each item, the attribute information of the received content, the generated user history information, and the sum of the results of the inner product for all items, The similarity is calculated, and the content is selected based on the calculated similarity. As the weighting information, the cosine distance between the standard preference vector indicating the viewer's general preference obtained by counting the attribute information of a predetermined period and the vector of the normal history information is calculated for each item, and the reciprocal thereof is the attribute When calculating the degree of similarity between information and positive history information, this is positive weighting information that defines the degree of positive contribution of each item, and the cosine distance between the standard preference vector and the negative history information vector for each item The reciprocal number is negative weighting information that defines the degree of negative contribution of each item when the similarity between attribute information and negative history information is calculated .

An information processing apparatus according to a second aspect of the present invention is an information processing apparatus that performs a process of selecting content similar to a user's preference, and stores attribute information of the content configured by a plurality of items User history information including first history information and positive history information that is a history of content that the user likes and negative history information that is a history of content that the user does not like from the operation history of the user's content. The user history information generating means for generating, the attribute information of the content, the positive history information and the negative history information are vectorized for each item, and a cosine distance and a weight of the vector of the content attribute information and the user history information An inner product with information is calculated for each item, and the result of the inner product is added to all the items, Calculating means for calculating similarity between the attribute information of the content stored by the first storage means and the generated user history information; and selecting means for selecting content based on the calculated similarity If, a the weighting information the calculation unit uses, obtained by counting the attribute information of a predetermined period, and the standard preference vector indicating the preferences of the viewer general, the cosine distance between the vector of the positive history information The standard preference vector is calculated for each item, and the reciprocal thereof is set as positive weighting information that defines the degree of positive contribution of each item when calculating the similarity between the attribute information and the positive history information. And the cosine distance between the negative history information vector and each item, and the reciprocal thereof is used to calculate the similarity between the attribute information and the negative history information. Kicking, and a weighting information generating means for generating a negative weighting information defining the negative degree of contribution of each item, the calculation unit, the cosine distance vector of the attribute information of the content positive history information An inner product with positive weighting information is calculated for each item, and the result of the inner product is added to all the items, and the cosine distance between the attribute information of the content and the vector of the negative history information and the negative An inner product with weighting information is calculated for each item, and a similarity between the attribute information of the content and the user history information is calculated based on a difference obtained by subtracting the sum of the inner product results for all the items. .

The user history information generating unit generates the user history information for each genre, and the weighting information generating unit sets a cosine distance between the standard preference vector and the vector of the positive history information for a predetermined genre. For each item, the reciprocal number is generated as the positive weighting information, and for the predetermined genre, the cosine distance between the standard preference vector and the negative history information vector is calculated for each item. An inverse number is generated as the negative weighting information, and the calculation means calculates an inner product of the attribute information of the content and the cosine distance of the vector of the positive history information of the predetermined genre for each item, and From the sum of the inner product results for all the items, the attribute information of the content and the predetermined genre The attribute information of the content of the predetermined genre is calculated by calculating the inner product of the vector of the negative history information with the cosine distance of each item and subtracting the sum of the results of the inner products for all the items. And the user history information can be calculated .

The first storage means also stores attribute information of the content to which information indicating a predetermined group when content is classified into a plurality of groups is stored, and the selection means stores the attribute information of the content When the information indicating the group is added to, the content corresponding to the information indicating the group is selected, and when the information indicating the group is not added, based on the similarity calculated by the calculation unit , The content can be selected .

The content may be a program, and the group may be grouped by any one of a series drama, content genre, performer, and title .

The storage device may further comprise second storage means for storing weighting information corresponding to the item of the attribute information generated by the weighting information generation means .

The information processing apparatus may further include an accumulating unit that accumulates an operation history of the user content .

An information processing method according to a second aspect of the present invention is an information processing method for an information processing apparatus that executes processing for selecting content similar to a user's preference, and the information processing apparatus includes a plurality of items. A first storage step for storing the attribute information of the content that is stored, and positive history information that is a history of the user's preferred content and a history of content that the user does not like from the operation history of the user's content. A user history information generating step for generating user history information comprising negative history information , the content attribute information, the positive history information and the negative history information are vectorized for each item, and the content attribute information and the content information Calculate the inner product of the cosine distance of the vector of user history information and the weighting information for each item, and the result of the inner product The calculation step of calculating the similarity between the attribute information of the content stored by the processing of the first storage step and the user history information by the sum of all the items, and the calculated similarity A standard step indicating a general preference of the viewer, which is obtained by counting the attribute information for a predetermined period , which is the weighting information used in the processing of the selection step of selecting content and the calculation step based on the degree The cosine distance between the vector and the vector of the positive history information is calculated for each item, and the reciprocal of the positive contribution of each item when calculating the similarity between the attribute information and the positive history information . a positive weighting information defining the degree, calculated with the standard favorite vector, a cosine distance between the vector of the negative history information for each of the items, The inverse of, in a case of calculating the similarity between the attribute information and the negative history information, and a weighting information generating step of generating a negative weighting information defining the negative degree of contribution of each item, the calculated In the processing of the step, the inner product of the cosine distance between the attribute information of the content and the vector of the positive history information and the positive weighting information is calculated for each item, and the result of the inner product is added to all the items. From the sum, the inner product of the cosine distance of the content attribute information and the negative history information vector and the negative weighting information is calculated for each item, and the sum of the inner product results for all the items is calculated. The similarity between the content attribute information and the user history information is calculated based on the subtracted difference .

In the processing of the user history information generation step, the user history information is generated for each genre, and in the processing of the weighting information generation step, a cosine distance between the standard preference vector and the vector of the positive history information for a predetermined genre. Is calculated for each item, and the reciprocal thereof is generated as the positive weighting information, and the cosine distance between the standard preference vector and the negative history information vector is calculated for each item for the predetermined genre. The reciprocal is generated as the negative weighting information, and in the processing of the calculating step, an inner product of the attribute information of the content and the cosine distance of the vector of the positive history information of the predetermined genre is calculated for each item. The content is calculated from the sum of the inner product results for all the items. The inner product of the attribute information and the cosine distance of the vector of the negative history information of the predetermined genre is calculated for each item, and the result of the inner product is subtracted from the sum obtained by adding all the items. It is possible to calculate the similarity between the attribute information of the content of the genre and the user history information .

In the process of the first storage step, attribute information of the content to which information indicating a predetermined group when the content is classified into a plurality of groups is also stored, and in the process of the selection step, the content of the content is stored. When the information indicating the group is added to the attribute information, the content corresponding to the information indicating the group is selected. When the information indicating the group is not added, the similarity calculated by the calculation step is set. based on, it is possible to select a content.

The content may be a program, and the group may be grouped by any one of a series drama, content genre, performer, and title .

A second storage step of storing weighting information corresponding to the item of the attribute information generated by the weighting information generation step can be further included.

An accumulating step of accumulating an operation history of the user content can be further included .

A program according to a second aspect of the present invention is a program for causing a computer to execute processing for selecting content similar to a user's preference, and stores attribute information of the content composed of a plurality of items. User history information consisting of first history step and positive history information which is a history of content preferred by the user and negative history information which is a history of content not preferred by the user from the operation history of the user's content. The user history information generation step, the content attribute information, the positive history information, and the negative history information are vectorized for each item, and the cosine distance and weight of the content attribute information and the user history information vector Calculate an inner product with information for each item, and calculate the inner product result for all the items. A calculation step of calculating a similarity between the attribute information of the content stored by the processing of the first storage step and the generated user history information by a sum obtained by adding to the calculated similarity A selection step for selecting content, and the weighting information used in the processing of the calculation step, and a standard preference vector indicating general viewer preference obtained by counting the attribute information for a predetermined period; Calculating the degree of positive contribution of each item when calculating the cosine distance to the vector of the positive history information for each item and calculating the reciprocal of the similarity between the attribute information and the positive history information. a positive weighting information defining calculates said standard preference vector, a cosine distance between the vector of the negative history information for each of the items, and vice versa The said in the case of calculating the similarity of attribute information and said negative history information, and a weighting information generating step of generating a negative weighting information defining the negative degree of contribution of each item, the calculation step The process calculates, for each item, an inner product of the cosine distance of the attribute information of the content and the vector of the positive history information and the positive weighting information, and the result of the inner product is added from all the items. The inner product of the cosine distance of the content attribute information and the negative history information vector and the negative weighting information is calculated for each item, and the result of the inner product is added to all the items. Due to the difference, the computer is caused to execute a process of calculating the similarity between the content attribute information and the user history information .

In the second aspect of the present invention, content attribute information composed of a plurality of items is stored, and from the operation history of the user's content, the correct history information that is the history of the user's preferred content and the user's preference. User history information consisting of negative history information, which is a history of no content, is generated, the attribute information of the content and the positive history information and the negative history information are vectorized for each item, and the vector of the content attribute information and the user history information The inner product of the cosine distance and the weighting information is calculated for each item, and the similarity between the attribute information of the received content and the generated user history information is obtained by adding the result of the inner product for all items. The content is selected based on the calculated similarity. As the weighting information, the cosine distance between the standard preference vector indicating the viewer's general preference obtained by counting the attribute information of a predetermined period and the vector of the normal history information is calculated for each item, and the reciprocal thereof is the attribute When calculating the degree of similarity between information and positive history information, this is positive weighting information that defines the degree of positive contribution of each item, and the cosine distance between the standard preference vector and the negative history information vector for each item The reciprocal number is negative weighting information that defines the degree of negative contribution of each item when calculating the similarity between the attribute information and the negative history information.

  As described above, according to the present invention, it is possible to calculate the similarity between the attribute information and the information acquired from the user, and in particular, the similarity between the acquired attribute information and the information acquired from the user. Can be calculated using predetermined weighting information.

  According to another aspect of the present invention, the degree of similarity between attribute information and information acquired from a user can be calculated. In particular, the similarity between stored attribute information and information acquired from a user can be calculated. The degree can be calculated using predetermined weighting information.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  A television program broadcast and stream data distribution will be described with reference to FIG.

  The broadcast station 1 transmits a terrestrial program broadcast, or transmits a satellite wave program broadcast via the satellite 2. In FIG. 1, only one broadcasting station 1 is illustrated and described, but it goes without saying that there are a plurality of broadcasting stations 1. The antenna 3 of the television receiver 4 receives a terrestrial or satellite wave program broadcast. The broadcast signal includes an EPG (Electronic Program Guide) as necessary.

  The distribution server 5 reads the streaming data from the streaming data database 6 and distributes the streaming data to the television receiver 4 via the network 8 including the Internet and other networks, and is broadcast from the broadcast station 1 from the metadata database 7. EPG, which is information related to the program to be read, or metadata including more detailed information than EPG is read out, and a program vector PP is generated for each program, and the EPG data is sent to the EPG receiver 9 via the network 8 together with the EPG data. To deliver.

  When the information amount of the EPG superimposed on a general broadcast signal is sufficient information for the processing described below, the same EPG as that superimposed on the general broadcast signal is used for the processing. Anyway. In addition, when it cannot be said that the information amount of the EPG superimposed on the general broadcast signal is sufficient information for the processing described below, the processing includes the EPG superimposed on the general broadcast signal. Alternatively, the metadata may be used alone. Hereinafter, since an EPG superimposed on a general broadcast signal does not have a sufficient amount of information for processing, in this embodiment, information including metadata is used, and this information is collectively referred to as EPG data. Shall.

  The EPG receiver 9 supplies the distributed EPG data to the television receiver 4. Further, the EPG receiving device 9 supplies the program vector PP distributed together with the EPG data to the program recommendation processing device 10.

  The television receiver 4 having a tuner receives a terrestrial wave via an antenna 3 based on a control signal indicating channel selection supplied from a television display device 11 having an operation unit or a recording / playback device 12. Alternatively, a satellite wave broadcast signal is selected and received, and distribution of streaming data is received from the distribution server 5 via the network 8. Also, the television receiver 4 receives the EPG data from the EPG receiver 9 and supplies it to the television display device 11 or the recording / reproducing device 12. When the received broadcast wave includes an EPG, the television receiver 4 separates it from the program signal and supplies it to the television display device 11 or the recording / playback device 12, respectively.

  The program recommendation processing device 10 acquires the program vector PP from the EPG receiving device 9, acquires operation logs from the television display device 11 and the recording / playback device 12, and receives this information or user operation input. Based on this, recommendation information for recommending a program that matches the user's preference is generated and supplied to the television display device 11 and the recording / playback device 12.

  The television display device 11 displays a broadcast signal supplied from the television receiver 4 or a playback signal supplied from the recording / playback device 12 based on a user operation input, or from the program recommendation processing device 10. Based on the supplied recommendation information, the channel is automatically set and recommended program information is displayed. The television display device 11 supplies an operation log, which is a user operation history, to the program recommendation processing device 10.

  The recording / playback device 12 records a broadcast signal supplied from the television receiver 4 on an attached recording medium or a built-in recording medium (for example, a hard disk) or a program based on a user operation input. Based on the recommendation information supplied from the recommendation processing device 10, the broadcast signal supplied from the television receiver 4 is automatically recorded on the attached recording medium or a built-in recording medium. In addition, the recording / reproducing apparatus 12 reproduces a program recorded on the attached recording medium or a built-in recording medium, and outputs and displays the program on the television display apparatus 11. Furthermore, the recording / playback apparatus 12 supplies an operation log, which is a user operation history, to the program recommendation processing apparatus 10.

  Here, the EPG reception device 9, the television reception device 4, the program recommendation processing device 10, the television display device 11, and the recording / playback device 12 have been described as different devices, but these devices are individually configured. It does not have to be done. For example, it goes without saying that the EPG receiving device 9, the television receiving device 4, and the television display device 11 may be integrally configured as a television receiver 15-1 having a built-in tuner function. Yes. Further, the recording / reproducing apparatus 12 may be integrally configured as a television receiver 15-2 having a recording function. Needless to say, the recording / reproducing apparatus 12 may be a so-called hard disk recorder having a large-capacity hard disk as a recording medium. Further, the program recommendation processing device 10 is incorporated in a television receiver 15-1 having a tuner function, and may be used as the television receiver 15-3 or in the television receiver 15-2 having a recording function. It may be built in and used as the television receiver 15-4.

  FIG. 2 is a block diagram showing the configuration of the distribution server 5.

  The data acquisition unit 21 acquires data from the metadata database 7 and the streaming data database 6, supplies the data to the data transmission unit 25, and supplies EPG data to the metadata extraction unit 22. In addition, the data acquisition unit 21 performs processing for grouping the EPG data registered in the metadata database 7 according to the content thereof.

  The metadata extraction unit 22 extracts data necessary for generating the program vector PP from the EPG data supplied from the data acquisition unit 21 and supplies the data to the program vector generation unit 23. The program vector generation unit 23 generates a program vector PP based on the metadata, associates the program-side effect vector EfPP stored in the data storage unit 24 with the data, and supplies it to the data transmission unit 25 as necessary. To do.

  The data storage unit 24 stores the program side effect vector EfPP, which is information necessary for generating the program vector PP, as necessary.

  Based on the viewpoint of the production side and the broadcast side on how to earn the audience rating, for example, the characteristic that the element of genre and content is strong in the educational program, and the characteristic that the element of the performer is strong in the variety program etc. In drama, etc., there is a characteristic that the elements of performers and screenwriters are strong. In order to correctly extract program characteristics, it is necessary to use these characteristics. That is, of the elements constituting the program vector PP, which element is important in the matching process executed to recommend a program that matches the user's preference differs depending on the genre of the program.

  That is, when the genre of the program is “Liberal Arts / Documents”, the important item is not the performer but the contents and title. On the other hand, when the program genre is “variety”, the important item is a performer, and when the program genre is “drama”, the important items are a performer and a screenwriter. When these are used to generate the program vector PP, a program-side effect vector that defines the degree of contribution in the matching process for each item is set for each genre of the program and stored in the data storage unit 24.

  The data transmission unit 25 receives the EPG data and streaming data supplied from the data acquisition unit 21, and the program vector PP and the program side effect vector EfPP supplied from the program vector generation unit 23 via the network 8. , And transmitted to the EPG receiver 9 or the television receiver 4.

  In addition, a drive 26 is connected to the program vector generation unit 23 as necessary. A magnetic disk 31, an optical disk 32, a magneto-optical disk 33, and a semiconductor memory 34 are mounted on the drive 26 as necessary to exchange data.

  Next, the program vector generation process 1 executed by the distribution server 5 will be described with reference to the flowchart of FIG.

  In step S <b> 1, the data acquisition unit 21 receives supply of EPG data composed of metadata from the metadata database 7.

  In step S <b> 2, the metadata extraction unit 22 receives supply of EPG data from the data acquisition unit 21, extracts metadata necessary for generating the program vector PP, and outputs the metadata to the program vector generation unit 23.

  FIG. 4 shows an example of metadata. The metadata includes, for example, “movie-Japanese movie” as the genre, “Tokaido Mitani Kaidan” as the title of the movie, date of broadcasting and distribution source data, date and time of broadcasting, name of broadcasting station, and broadcasting The time is included. Further, the metadata includes data such as a director name, a screenwriter name, a photographer (cameraman), a music representative, a performer, etc., and a description of a movie as the contents of the program.

  In step S3, the program vector generation unit 23 performs morphological analysis on titles, contents, and the like included in the metadata as necessary, and breaks them down into words. Specifically, the program vector generation unit 23 uses the title of the movie included in the metadata as a title and breaks it down into three words “Tokaido”, “Mitani”, and “Kaidan”. In addition, the program vector generation unit 23 is a masterpiece of a Japanese movie horror that describes the world of “Mitsuya Kaidan”, which is famous for its '59 Seiho style beauty, as shown in FIG. ”Is extracted, the words included in the commentary are extracted and“ Seiho ”,“ Style ”,“ Beauty ”,“ Plenty ”,“ Famous ”,“ Mitani ”,“ Kaidan ” , “World”, “Draw”, “Japanese Film”, “Horror”, “Masterpiece” are the content information.

  In step S4, the program vector generation unit 23 vectorizes each item included in the metadata, generates a program vector PP, and the process ends. The generated program vector PP is output to the data transmitter 25 and transmitted to the EPG receiver 9 via the network 8. The item vectorization may be such that all the detailed elements are arranged in one column, or may be vectorized for each major item after being separated into major items.

  When vectorization is performed for each major item, for example, as shown in FIG. 5, the major item is divided into seven items: title, genre, time zone, broadcast station, performer, screenplay / original / production, and content. The vector PP = (Tm, Gm, Pm, Am, Km) is generated. The contents of each item are title Tm = {title1, title2,...}, Genre (Genre) Gm = {drama, variety, sports, movie, music, children / education, education / documents, News / News, Other}, Hour Hm = {Morning, Noon, Evening, Golden, Late Night}, Broadcasting Station (TV Station) Sm = {NNK General, NNK Education, Asia TV, TTS, Buji, Tele-day , Higashi, NNK Satellite 1, NNK Satellite 2, WOWO}, Performer (Person) Pm = {person A, person B, ...} Screenplay / Original / Director, etc. (Author) Am = {person a, person b,...} Contents (Keyword) Km = {kw1, kw2,.

  Among the seven major items described above, the major items “genre”, “broadcasting station”, and “time zone” can be fixed in type, so that detailed elements can be represented by numerical vectors. In the case of a broadcasting station, for example, in the broadcasting station Sm = {NNK General, NNK Education, Asia TV, TTS, Buji, Tele-day, East Tokyo, NNK Satellite 1, NNK Satellite 2, WOWO} Is WOWO, the broadcast station vector can be represented by broadcast station Sm = {0, 0, 0, 0, 0, 0, 0, 0, 0, 1}. In the case of a genre, for example, in the genre Gm = {drama, variety, sport, movie, music, for children / education, education / documents, news / reports, etc.}, the corresponding program genre was education / documents. In this case, the genre vector can be represented by genre Gm = {0, 0, 0, 0, 0, 0, 1, 0, 0}.

  On the other hand, for items that cannot be represented by numeric vectors, such as the major items “title”, “performer”, “screenplay / original / production”, “content”, the item value depends on the frequency of words contained in the vector. (Word) and weight (numerical value) are associated with each other. For example, the major item “title” is represented by elements each composed of a word and a numerical value, such as title Tm = {Tokaido-1, Mitani-1, Kaidan-1}. For example, “Tokaido-1” in this example means that the frequency of the word “Tokaido” is “1”.

  Through such processing, a program vector PP is generated and transmitted to the EPG receiving device 9. Specifically, when the program vector PP is generated from the program metadata described with reference to FIG. 4, the program vector PP = {title Tm = {Tokaido-1, Mitani-1, Kaidan-1}, genre Gm = {0,0,0,1,0,0,0,0,0}, time zone Hm = {0,0,0,0,1}, broadcast station Sm = {0,0,0,0,0 , 0, 0, 0, 0, 1}, performer Pm = {Katsumi Wakasugi-1, Shigeru Amami-1, Toshihiko Emi-1, Ryujiro Nakamura-1, Noriko Nishizawa}, Screenplay / Original / Director Am = {Nobuo Nakagawa-1, Masakazu Onuki-1, Yoshihiro Ishida-1, Shoji Nishimoto-1, Sota Watanabe-1}, Content Km = {Nishiho-1, Style-1, Beauty-1, Plenty-1, Famous-1, Mitani-1, Kaidan-1, World-1, Draw-1, Japanese-style-1, Horror-1, Masterpiece-1}} are generated and transmitted to the EPG receiver 9.

  In the program vector PP generated in this way, for example, the starring is weighted out of the constituent elements of the performer Pm, or the director is weighted out of the constituent elements of the script / original / production Am, and the program vector PP is You may make it produce | generate. In such a case, the constituent elements of the program vector PP are, for example, performers Pm = {Katsumi Wakasugi-3, Shigeru Amami-2, Toshihiko Emi-1, Ryujiro Nakamura-1, Noriko Nishizawa-1}, Screenplay / Original / Production Am = {Nobuo Nakagawa-3, Shoichi Onuki-1, Yoshihiro Ishida-1, Shoji Nishimoto-1, Sota Watanabe-1}.

  In addition, as described above, which element is important for generating the program vector PP depends on the genre of the program. When the program side effect vector EfPP indicating what elements are important for each genre is stored in the data storage unit 24 as program side effect vector information, the program side effect vector EfPP is converted into the program vector PP. You may make it transmit in response to. The program side effect vector EfPP is set corresponding to the major items of the program vector PP.

  In the case of the program vector PP = {title Tm, genre Gm, time zone Hm, broadcast station Sm, performer Pm, screenplay / original / production Am, content Km}, for example, if the genre of the program is “Education / Document”, Since the important items are the contents and title (when the weighting value is 3), the program side effect vector EfPP = {3, 1, 1, 1, 1, 1, 3}. On the other hand, when the genre of the program is “variety”, the important item is the performer (when the weighting value is 5), so the program side effect vector EfPP = {1, 1, 1, 1, 5, 1,1}. Also, when the program genre is “drama”, the important items are the performer (when the weighting value is 2) and the screenwriter, especially when the screenwriter is important (when the weighting value is 3). The program side effect vector EfPP = {1, 1, 1, 1, 2, 3, 1}.

  Next, the program vector generation process 2 when the program side effect vector EfPP is transmitted in association with the program vector PP will be described with reference to the flowchart of FIG.

  In steps S11 to S14, processing similar to the processing in steps S1 to S4 described with reference to FIG. 3 is executed. That is, EPG data is supplied from the metadata database 7, metadata necessary for generating the program vector PP is extracted from the EPG data, and the title, content, etc. included in the metadata are necessary. Morphological analysis is performed according to Then, each item included in the metadata is vectorized to generate a program vector PP.

  In step S <b> 15, the program vector generation unit 23 extracts the corresponding effect vector from the effect vector information stored in the data storage unit 24 based on the genre of the supplied metadata program. For example, when the genre of the program is “variety”, an effect vector EfPP = {1, 1, 1, 1, 5, 1, 1} for weighting performers is extracted from the data storage unit 24.

  In step S16, the program vector generation unit 23 associates the effect vector EfPP extracted in step S15 with the program vector PP generated in step S14, and the process ends.

  By such processing, a program vector PP is generated, and a program-side effect vector EfPP for weighting important items is associated with the EPG receiving apparatus via the network 8 based on the genre of the program. 9 is transmitted.

  As described above, the program vector PP is generated. For example, the program vector PP can be accurately created with a small amount of calculation by grouping the programs by their attributes. .

  The program grouping can be generated by, for example, a continuous drama, a program group of the same genre in a week unit, a program group for each performer in a program organization unit (13 weeks), etc. As a specific example, the following FIG. The grouping process 1 in the case of grouping serial dramas will be described with reference to the flowchart of FIG.

  In step S31, the data acquisition unit 21 determines from the EPG data registered in the metadata database 7 that the broadcast time is the same time on weekdays or the same time on the same day of the week with the same title and the same broadcast station, for example. The programs that match the predetermined conditions (grouping conditions) are extracted.

  In step S32, the data acquisition unit 21 groups the extracted programs and attaches a group ID to the EPG data of the corresponding program.

  In step S33, the metadata extraction unit 22 extracts metadata necessary for generating a program vector of the first broadcast of a program having the same group ID, that is, a program recognized as a continuous drama.

  In step S34, the program vector generation process described with reference to FIG. 3 or FIG. 6 is executed.

  In step S35, the program vector generation unit 23 sets the program vector PP of the corresponding group ID fixed to the program vector for the first broadcast, and the process ends.

  By such processing, continuous dramas are grouped and program vectors are unified. Further, the group ID and the program vector PP may be associated with each other and stored in the data storage unit 24.

  In particular, in the EPG data of serial dramas, the initial contents are commentary for the entire program, but the contents other than the first time are often commentary for that time, representing the contents of the entire program. Not. Further, since the metadata other than the contents is almost the same every time, the number of times of the program vector generation process can be reduced only by generating the program vector PP by adopting the EPG data for the first time. Instead, it is possible to generate the program vector PP that more accurately matches the program characteristics.

  In FIG. 7, the process of adding the same ID to the EPG data of the continuous drama has been described. However, when the EPG data includes information for distinguishing the continuous drama in advance, the process of step S31 is omitted. The same ID may be added to the EPG of the continuous drama with reference to information for distinguishing the continuous drama included in the EPG data.

  Next, with reference to the flowchart of FIG. 8, the grouping process 2 which can group by elements other than a continuous drama is demonstrated. Here, it demonstrates as what performs grouping by a genre and a performer.

  In step S51, the data acquisition unit 21 refers to the EPG data registered in the metadata database 7 and adds a cluster code corresponding to the content of the metadata to the EPG data of the program.

  For example, a program that matches the conditions for a continuous drama, such as the same title and the same broadcasting station, and the broadcast time is the same time on weekdays or the same day of the week, is extracted, and is a continuous drama. In the program group recognized as, for example, according to the broadcast time of the continuous drama, code 1 if the same time on weekdays, code 2 if the same time on the same day of the week, code 1 and code 2 such as biweekly, every other day, etc. If the broadcast time is other than the first, the first cluster code of the smallest digit such as code 3 is determined. If it is not a continuous drama, the first cluster code of the minimum digit is 0.

  Next, by referring to the data related to the genre of metadata registered in the metadata database 7, for each genre, for example, a numerical value in the second digit (10th digit) from the lower order such as code 20 and code 30 A second cluster code having is determined.

  Finally, referring to the data of the performers of the metadata registered in the metadata database 7, for example, the code 2300, the code 800, etc. are represented using the second digit or more (100 or more) from the lower order. A third cluster code is determined.

  The total value of the first to third cluster codes becomes the cluster code added to the EPG of the program.

  In step S52, the data acquisition unit 21 determines whether the corresponding EPG is that of a continuous drama based on whether the first digit of the cluster code is 0 or not.

  If it is determined in step S52 that the drama is a continuous drama, the same processing as in steps S33 and S34 of FIG. 7 is executed in steps S53 and S54. That is, the EPG of the first broadcast is extracted, and the program vector generation process described with reference to FIG. 3 or FIG. 6 is executed.

  In step S55, the program vector generation unit 23 sets the program vector PP of this continuous drama fixed to the program vector PP for the first broadcast, and associates the generated cluster code with the program vector PP. The setting is completed.

  If it is determined in step S52 that the drama is not a continuous drama, the program vector generation process described with reference to FIG. 3 or FIG. 6 is executed in step S56.

  In step S57, the program vector generation unit 23 associates the generated cluster code with the program vector PP, and the process ends.

  By such a process, even if it is not a continuous drama, the program vectors PP can be grouped by genre or performer, and the cluster code indicating the group can be associated with the program vector PP.

  Since the distribution server 5 described with reference to FIG. 2 generates the program vector PP as described with reference to FIGS. 3 to 8, the program vector PP always corresponding to a new term or genre is generated. Is possible. The generated program vector PP is received together with the EPG data by the EPG receiving device 9 via the network 8 and supplied to the program recommendation processing device 10.

  In addition, the grouping process may be performed by performing morphological analysis on titles, decomposing them into words, and attaching a group ID to each word.

  Next, the title grouping process 1 will be described with reference to the flowchart of FIG.

  In step S <b> 61, the data acquisition unit 21 refers to the EPG data registered in the metadata database 7, extracts a title from the registered metadata, and supplies the title to the program vector generation unit 23.

  In step S62, the program vector generation unit 23 performs morphological analysis on the title and breaks it down into words. Specifically, if the title of the movie included in the metadata is “Tokaido Mitani Kaidan”, it will be morphologically analyzed as a title, so the three words “Tokaido”, “Mitani”, and “Kaidan” Is obtained.

  In step S63, the program vector generation unit 23 extracts either the analyzed word or a word group composed of a plurality of words, and extracts the extracted word or word from the data storage unit 24. A group ID corresponding to the group is extracted.

  Here, the word group composed of a plurality of words is a word group generated by a combination of words obtained by morphological analysis. For example, words obtained by morphological analysis are “Tokaido” and “Mitani”. , The word group is “Tokaido / Mitsuya”, “Tokaido / Kaidan”, “Mitani / Kaidan”.

  In step S <b> 64, the program vector generation unit 23 determines whether the corresponding group ID has been extracted from the data storage unit 24.

  If it is determined in step S64 that the corresponding group ID has not been extracted, no group ID has been assigned to the extracted word or a word group composed of a plurality of words. The program vector generation unit 23 associates a new group ID with the extracted word or a word group composed of a plurality of words. Further, the program vector generation unit 23 causes the data storage unit 24 to store a word or a word group composed of a plurality of words and a group ID corresponding to the word group.

  When it is determined in step S64 that the corresponding group ID has been extracted, or after the process of step S65 is completed, in step S66, the program vector generation unit 23 selects all the words constituting the title, or a plurality of words It is determined whether or not a group ID has been extracted for a word group composed of words.

  If it is determined in step S66 that no group ID has been extracted for all words constituting the title or a word group composed of a plurality of words, the process returns to step S63, and the subsequent processes are performed. Repeated.

  If it is determined in step S66 that the group ID has been extracted for all words constituting the title or a word group composed of a plurality of words, in step S67, the program vector generation unit 23 adds the program vector to the program vector. The process ends after associating the extracted group IDs.

  By such processing, the program vector is associated with the word constituting the title or the group ID corresponding to the word group, and the data transmission unit 25 transmits the television receiver 4 or the EPG via the network 8. It is transmitted to the receiving device 9.

  Also, programs with similar titles may be grouped together. For example, the title is structured so that the series “2 years A group Ginhachi teacher” and the special program “2 years A group Ginhachi teacher special” can be grouped as the same group. Based on the morphological analysis result of the word to be calculated, the word match rate is calculated for the program title of a predetermined period such as 2 weeks, 1 month, 6 months, etc., and the word match rate is equal to or higher than the predetermined value. In some cases, the same group may be used.

  Next, with reference to the flowchart of FIG. 10, the title grouping process 2 in which grouping is performed based on the matching rate of words constituting the title will be described.

  In step S401 and step S402, the same processing as step S61 and step S62 described with reference to FIG. 9 is executed. That is, the data acquisition unit 21 refers to the EPG data registered in the metadata database 7, extracts the title from the registered metadata, supplies the title to the program vector generation unit 23, and supplies the program vector generation unit 23 23 morphologically analyzes the title and breaks it down into words.

  In step S <b> 403, the program vector generation unit 23 calculates the matching degree of words between titles based on the analyzed words, that is, a matching rate indicating a ratio of matching words.

  Specifically, the title “Two Years A Group Ginpachi Teacher” and the title “Two Years A Group Ginpachi Special” are “2”, “Year”, “A”, “Group”, “Ginpachi”, “ In the case of morphological analysis of “Teacher” and “2” “Year” “A” “Group” “Ginpachi” “Teacher” “Special”, the matching rate of words constituting the titles of these two programs is 6 / 7 is 85.7%.

  In step S404, the program vector generation unit 23 determines whether or not the words match a predetermined value or more such as 70%. Needless to say, the threshold value of the coincidence rate may be any numerical value other than 70%.

  If it is determined in step S404 that the words match a predetermined value such as 70% or more, in step S405, the program vector generation unit 23 associates these programs with the same group ID. In addition, the program vector generation unit 23 causes the data storage unit 24 to store the matched word or word group and the corresponding group ID.

  When it is determined in step S404 that the matching rate is equal to or less than a predetermined value such as 70%, or after the processing in step S405 is completed, in step S406, the program vector generation unit 23 finishes the brute force of the title. Determine whether or not.

  If it is determined in step S406 that the title round robin has not ended, the process returns to step S403, and the subsequent processes are repeated.

  If it is determined in step S406 that the title round robin has been completed, the processing is terminated.

  By such processing, the program vector is associated with the group ID based on the matching rate of the words constituting the title, and the data transmission unit 25 transmits the television receiver 4 or EPG via the network 8. Since it is transmitted to the receiving device 9, for example, programs with similar titles such as a series drama and a special program can be processed as the same group.

  Furthermore, by determining the group based on the matching rate of the words that make up the title, for example, in EPG or metadata, in half-width and full-width of numbers, half-width and full-width of English letters, or capital letters Even when there is a fluctuation in the case of lowercase letters or the like, it is possible to detect programs with the same title as the same group.

  Further, in addition to the word matching rate, for example, a broadcasting station, a program genre, or a broadcast start time may be added to the grouping condition. For example, in a news program or the like, the title is composed of a small number of words including “news”. Therefore, in the process described with reference to FIG. Therefore, in addition to the word matching rate, if the broadcasting stations match, the same group may be used.

  With reference to the flowchart of FIG. 11, a title grouping process 3 for performing grouping based on the matching rate of words constituting a title, with the matching of broadcasting stations as a condition, will be described.

  In steps S421 to S424, the same processing as that in steps S401 to S404 described with reference to FIG. 10 is executed. That is, the data acquisition unit 21 refers to the EPG data registered in the metadata database 7, extracts the title from the registered metadata, supplies the title to the program vector generation unit 23, and supplies the program vector generation unit 23 23 morphologically analyzes the title and breaks it down into words. Then, the program vector generation unit 23 calculates the degree of matching between words based on the analyzed word, and determines whether or not the word matches a predetermined value such as 70% or more. .

  If it is determined in step S424 that the words match a predetermined value such as 70% or more, in step S425, the program vector generation unit 23 determines whether the broadcast stations of these programs match. to decide.

  If it is determined in step S425 that the broadcast stations of these programs match, in step S426, the program vector generation unit 23 associates the same group ID with these programs. Further, the program vector generation unit 23 stores the matched word or word group and the corresponding broadcasting station and group ID in the data storage unit 24.

  If it is determined in step S424 that the match rate is equal to or less than a predetermined value such as 70%, if it is determined in step S425 that the broadcast stations of these programs do not match, or the process of step S426 In step S427, the program vector generation unit 23 determines whether or not the title round-robin has ended.

  If it is determined in step S427 that the title round-robin has not ended, the process returns to step S423, and the subsequent processes are repeated.

  If it is determined in step S427 that the title round-robin has ended, the processing ends.

  By such processing, the program vector is associated with the group ID based on the match rate of the broadcasting station and the word constituting the title, and the data transmission unit 25 receives the television signal via the network 8. Since it is transmitted to the apparatus 4 or the EPG receiving apparatus 9, for example, when programs with similar titles are set to the same group, it is possible to prevent the news programs of other stations from being set to the same group. .

  Note that in FIG. 11, the grouping is performed on the condition that the broadcasting stations are the same, in addition to the matching rate of the words constituting the title. It goes without saying that grouping may be executed using a band, a genre, or the like as a condition other than the matching rate of words constituting the title.

  Furthermore, for example, even if the broadcast start time of a continuous drama or band program is shifted due to a sports broadcast or a special program, conditions other than the matching rate of words constituting the title so that it can be detected as the same group Grouping may be executed based on whether or not the broadcast times coincide with each other within a predetermined time range such as one hour.

  Referring to the flowchart of FIG. 12, title grouping that performs grouping based on the matching rate of words constituting the title, in addition to whether or not the broadcast time matches with a deviation within a predetermined time range Processing 4 will be described.

  In steps S441 to S444, the same processing as that in steps S401 to S404 described with reference to FIG. 10 is executed. That is, the data acquisition unit 21 refers to the EPG data registered in the metadata database 7, extracts the title from the registered metadata, supplies the title to the program vector generation unit 23, and supplies the program vector generation unit 23 23 morphologically analyzes the title and breaks it down into words. Then, the program vector generation unit 23 calculates the degree of matching between words based on the analyzed word, and determines whether or not the word matches a predetermined value such as 70% or more. .

  If it is determined in step S444 that the words match a predetermined value such as 70% or more, in step S445, the program vector generation unit 23 sets the broadcast start time of those programs to, for example, 1 hour. It is determined whether or not they coincide with each other within a predetermined range.

  If it is determined in step S445 that the broadcast start times of these programs match with a deviation within a predetermined range, in step S446, the program vector generation unit 23 assigns the same group ID to those programs. Associate. In addition, the program vector generation unit 23 causes the data storage unit 24 to store the matched word or word group, the corresponding broadcast start time range, and the group ID.

  If it is determined in step S444 that the match rate is equal to or less than a predetermined value such as 70%, if it is determined in step S445 that the broadcast start time of those programs is shifted beyond a predetermined range, or After the process of step S446 ends, in step S447, the program vector generation unit 23 determines whether or not the title round-robin has ended.

  If it is determined in step S447 that the title round-robin has not ended, the process returns to step S443, and the subsequent processes are repeated.

  If it is determined in step S447 that the title round robin has ended, the processing ends.

  By such processing, the program vector is associated with a match including a deviation within a predetermined range of the broadcast start time and a group ID based on the match rate of the words constituting the title. Since it is transmitted to the television receiving device 4 or the EPG receiving device 9 via the network 8, for example, when programs having similar titles are grouped, the broadcast time is changed by a special program or the like. In addition, it is possible to prevent programs that should be detected as the same group from being detected as the same group.

  FIG. 13 is a block diagram showing the configuration of the program recommendation processing device 10.

  The data acquisition unit 41 acquires the program vector PP transmitted from the distribution server 5 and the program side effect vector EfPP corresponding to the program vector PP. The program vector extraction unit 42 extracts a program vector PP necessary for matching processing or a program vector PP necessary for user model generation from the program vector PP acquired by the data acquisition unit 41, and if necessary, The program side effect vector EfPP corresponding to the program vector PP is supplied to the matching processing unit 43.

  The operation input unit 44 includes input devices such as a keyboard, a touch pad, and a mouse, for example. The operation input unit 44 receives initial registration information input by the user and a topic input for generating a user model, and outputs the initial registration information to the initial registration storage unit 45. To do. The initial registration storage unit 45 registers initial registration contents supplied from the operation input unit 44 and topics for user model generation, and supplies them to the operation log acquisition unit 46 or the matching processing unit 43 as necessary. To do. Further, the contents stored in the initial registration storage unit 45 are sequentially updated based on a user operation input from the operation input unit 44. Initially registered contents include information indicating programs that the user does not like, such as disliked genres, disliked keywords, disliked performers, favorite genres, favorite keywords, favorite performers, etc. There is information indicating programs that the user likes.

  The operation log acquisition unit 46 acquires operation logs from the television display device 11 or the recording / playback device 12, classifies these pieces of information into positive histories and negative histories, and stores them in the initial registration storage unit 45 as necessary. The program vector PP corresponding to the positive history and the negative history is read out from the program vector PP acquired by the data acquisition unit 41 with reference to the stored information and supplied to the positive history storage unit 47 and the negative history storage unit 48. And save it. The positive history storage unit 47 stores the supplied positive history and generates a positive history vector UP. The negative history storage unit 48 stores the supplied negative history and generates a negative history vector MUP. The generated positive history vector UP and negative history vector MUP are supplied to the matching processing unit 43.

  Here, the positive history is information used to extract a candidate of a program that the user wants to actively watch, in other words, a preferable program. For example, the user can watch the program, When recorded, or when a program proposed as a recommended program list to the user through processing described later is accepted and viewed or recorded by the user, the metadata of the program is stored as positive impression metadata as a positive history. Stored in the unit 47. The positive history storage unit 47 obtains the sum of positive histories for each detailed item or for each large item, and generates a positive history vector UP.

  On the other hand, the negative history is information used to remove programs that are reluctant to be viewed by the user, in other words, programs that are not preferable to the recommended programs. For example, they are disliked during initial registration. If the selected item, the program deleted without recording after recording, or the program proposed as a recommended program list to the user by the process described later is not accepted by the user and is not viewed or recorded, the program The metadata is stored in the negative history storage unit 48 as negative impression metadata. For example, if the user dislikes “sports” in the initial registration, the genre Gmup = {0, 0, 5, 0, 0, 0, 0, 0, 0} is negative with a weight of 5 as a negative impression. It is additionally stored in the history storage unit 48. The negative history storage unit 48 calculates the sum of negative histories for each detailed item or for each large item, and generates a negative history vector MUP.

  The matching processing unit 43 verifies matching between the program vector extracted from the program vector extracting unit 42 and the positive history vector UP or negative history vector MUP supplied from the positive history storage unit 47 or the negative history storage unit 48. To do.

  If the program vector PP, positive history vector UP, or negative history vector MUP is vectorized by arranging all the detailed elements in one column, the title, keyword, etc. are composed of a plurality of words. One word and an item such as a genre, for example, have the same weight in the vector. Therefore, the normalization processing unit 61 of the matching processing unit 43 normalizes the title and contents, which are items configured by words, by dividing the word by the number of words for each program. For example, when the title Tm = {Tokaido-1, Mitani-1, Kaidan-1}, the title Tm = {Tokaido: 0.33, Mitani: 0.33, Kaidan: 0.33} is normalized. As a result, both the title and the content have a total weight of 1 in the item, so that no inconvenience occurs in the matching process.

  The vector calculation unit 62 executes a matching process between the program vector PP and the positive history vector UP or the negative history vector MUP.

  When the program vector PP, the positive history vector UP, and the negative history vector MUP are represented by a vector in which all the elements of the detailed items are arranged in one column, the vector calculation unit 62 determines whether the positive history vector UP or the negative history vector MUP Based on the cosine distance (cosθ) between the program vector PP and the vector, the similarity SimUP between the program vector PP and the positive history vector UP is obtained as shown in the following equation (1). As described above, the similarity SimMUP between the program vector PP and the negative history vector MUP is obtained. The cosine distance is a value obtained by dividing the inner product of two vectors by the product of the absolute values of the vectors, as shown in the equations (1) and (2).

SimUP ＝ cosθu ＝ UP ・ PP / | UP | × | PP | ・ ・ ・ (1)
SimMUP = cosθm = MUP ・ PP / | MUP | × | PP | (2)

  In the equations (1) and (2), PP represents the program vector PP, UP represents the positive history vector UP, and MUP represents the negative history vector MUP. “·” Indicates an inner product, and “x” indicates element multiplication (scalar operation).

  Further, when the program vector PP, the positive history vector UP, and the negative history vector MUP are vectorized for each major item, the vector calculation unit 62 determines the similarity between the program vector PP and the positive history vector UP for each major item. Then, the similarity between the program vector PP and the negative history vector MUP is obtained, the sum of the similarities is calculated for each major item, and the similarity SimUP and the similarity SimMUP can be calculated. For example, in the large item “title” broken down into words, the positive history vector UP is the title Tup = {school−1, ghost story−1, toilet−1}, and the supplied program vector PP is the title Tm. = {Tokaido-1, Mitani-1, Kaidan-1} If the unit vector length is 1, the vector length (absolute value) is the square root of the sum of the squares of the elements. The cosine distance cos θt, which is the similarity between titles, is calculated as in the following equation (3).

    cosθt = (1 ・ 1) / (√3 × √3) = 1/3 (3)

  In Expression (3), “·” indicates an inner product, and “x” indicates a scalar operation. The similarity for each major item between the program vector PP and the negative history vector MUP is also calculated by the same calculation as in equation (3).

  For example, the positive history vector UP is set to the positive history vector UP = {title Tup, genre Gup, performer Pup, screenplay / original / production Aup, content (keyword) Kup} and negative history vector MUP, negative history vector MUP = { When title Tmup, genre Gmup, performer Pmup, screenplay / original / production Amup, and content (keyword) Kmup}, the similarity for each major item is calculated by the same calculation as in equation (3). Thus, the similarity SimUP and the similarity SimMUP are both calculated as the similarity Sim of the following equation (4).

      Sim = cosθt + cosθg + cosθp + cosθa + cosθk (4)

  Here, cosθt is the cosine distance between the program vector PP and the positive history vector UP or negative history vector MUP in the large item “title”, and cosθg is the positive value of the program vector PP in the large item “genre”. The cosine distance between the history vector UP or the negative history vector MUP, and cosθp is the cosine distance between the program vector PP and the positive history vector UP or the negative history vector MUP in the major item `` performer '', and cosθa is The cosine distance between the program vector PP and the positive history vector UP or the negative history vector MUP in the large item “screenplay / original / production”, and cosθk is the program vector PP and the positive history vector in the large item “content”. The cosine distance with the UP or negative history vector MUP.

When the similarity SimUP and the similarity SimMUP are calculated as the sum of the calculation results for each large item, there is no weight bias between items, which is synonymous with normalization. Therefore, unlike a vector in which all elements are arranged in one column, normalization processing by the normalization processing unit 61 may not be performed.

  That is, when normalization is used and a vector in which all elements are arranged in one column is used, items such as titles and contents are easily compared with items such as broadcast stations and genres, which are likely to accumulate in frequency. In, as the history increases, the number of words increases, but the frequency of each word hardly increases.

  For this reason, when all elements are used and the total is taken for each element, the influence of items such as broadcast stations and genres that tend to overlap frequently increases, so the user is a fan of commentator A, for example. So, if you like to watch “Live broadcast of the game of Team B that commentator A explains”, the information “Live broadcast of baseball” that is a genre is easy to overlap as a history. The information “Explanator A” is difficult to overlap as history. Therefore, there may be a case where the live broadcast of the game of the team B which is explained by another commentator is recommended and the variety program in which the commentator A appears is not recommended.

  On the other hand, by normalizing or calculating the sum of the calculation results for each large item as the similarity SimUP and similarity SimMUP, without being affected by the magnitude of the history frequency, The variety program in which the commentator A appears can be recommended, and the user's preference can be more accurately reflected.

  The vector calculation unit 62 generates the user initial registration information stored in the initial registration storage unit 45, the program side effect vector EfPP transmitted in association with the program vector PP, or the user information registration unit 63. The similarity SimUP and the similarity SimMUP may be calculated by weighting according to the registered user side effect vector EfUP (described later) or the user side reaction vector EfMUP (described later).

  Based on the similarity SimUP calculated by the method as described above, the vector calculation unit 62, for example, further records a history vector for a predetermined number of programs (for example, 10) having a high similarity to the positive history vector. Similarity SimMUP with MUP is calculated, SimUP−SimMUP is calculated, and the upper predetermined number (for example, 3) of the calculation results are output to the recommended information output unit 49 as recommended programs.

  Further, when the program vectors are grouped, the vector calculation unit 62 registers a recommended priority group in the user information registration unit 63 based on the recommended program information, and selects a program corresponding to the recommended priority group. , Preferentially recommend.

  Further, the vector calculation unit 62 generates a user model vector by filtering the program vector PP using the topic stored in the initial registration storage unit 45, registers it in the user information registration unit 63, and performs matching. Processing can be performed. Details of the user model will be described later.

  The user information registration unit 63 is based on the initial registration contents of the user supplied from the initial registration storage unit 45 or the positive history vector UP or negative history vector MUP supplied from the positive history storage unit 47 or the negative history storage unit 48. Then, the user side effect vector EfUP and the counter effect vector EfMUP are generated and stored. The user-side effect vector EfUP is a vector that indicates to the user which elements of the major items are important for program selection and items that are weighted for program selection, or , A vector indicating the user's preference in each item. The counter-effect vector EfMUP is a vector indicating to the user which element of the large items is an element that is not important for program selection and is an item that is not weighted for program selection, Or it is a vector which shows the item which a user does not like in each item.

  In other words, the user-side effect vector EfUP and the counter-effect vector EfMUP define which item contributes greatly in the matching between the program vector PP and the positive history vector UP or the negative history vector MUP.

  The user side effect vector EfUP and the counter effect vector EfMUP may be set by the user or may use predetermined values, but based on the initial registration contents of the user registered in the initial registration storage unit 45. May be generated.

  Specifically, when the user side effect vector EfUP is information indicating which of the major items is important for the user in selecting a program, the program vector PP = {title Tm, genre Gm , Time zone Hm, broadcast station Sm, performer Pm, screenplay / original / production Am, content Km}, if the genre is important to the user, for example, the effect vector EfUP = (1, 5, 1, 1, 1 , 1, 1). On the other hand, when the performer and the genre are important for the user, for example, the effect vector EfUP = (1, 3, 1, 1, 5, 1, 1) is set.

  When the user side effect vector EfUP is a vector indicating the user's preference in each item, the large item genre Gm = {drama, variety, sports, movie, music, children / education, education / documents, news / When the genre of the program that the user thinks preferable in the news report, etc. is education / document, the user effect vector EfUP is, for example, the genre Gm = {0, 0, 0, 0, 0, 0, 5, 0,0}.

  Further, the user side effect vector EfUP and the counter effect vector EfMUP may be generated based on the positive history vector UP or the negative history vector MUP or by counting the programs viewed by the user for a certain period. good. Furthermore, the user-side effect vector EfUP and the counter-effect vector EfMUP can be generated for each genre. A method for generating the user-side effect vector EfUP or the counter-effect vector EfMUP will be described later with reference to FIGS.

  Furthermore, the user information registration unit 63 registers the information of the recommendation priority group, the user model vector, and the like generated by the processing of the vector calculation unit 62 as necessary.

  In the matching processing unit 43 described above, from a program having a high degree of similarity to the positive history vector UP (a program that is actively viewed by the user), a program that the user does not like using the negative history (the user has For example, it is possible to determine a recommended program using only the positive history without performing a process of removing a program that is reluctant to watch.

  The recommended information output unit 49 registers the recommended program information supplied from the matching processing unit 43 in the recommended program list 50 and supplies it to the television display device 11 or the recording / reproducing device 12. The recommended program list 50 is configured to be removable from the program recommendation processing device 10, and registers recommended program information output from the recommended program output unit 49. By storing the recommended program information in the recommended program list 50, for example, even when different television receivers 4, television display devices 11, recording / reproducing devices 12, etc. are used, the history accumulated so far Using information, it is possible to execute program recommendation, automatic recording, and the like.

  The matching processing unit 43 is also connected to the drive 51 as necessary. A magnetic disk 71, an optical disk 72, a magneto-optical disk 73, and a semiconductor memory 74 are mounted on the drive 51 as necessary to exchange data.

  The positive history vector and negative history vector generation processing 1 executed by the program recommendation processing device 10 will be described with reference to the flowchart of FIG.

  In step S <b> 71, the operation log acquisition unit 46 supplies the initial registration content read from the initial registration storage unit 45 to the negative history storage unit 48. The negative history storage unit 48 refers to the supplied initial registration content and generates a negative history vector MUP.

  In step S <b> 72, the operation log acquisition unit 46 determines whether or not the initial registration content has been changed based on the registration content stored in the initial registration storage unit 45. If it is determined in step S72 that the initial registration content has been changed, the process returns to step S71, and the subsequent processes are repeated.

  If it is determined in step S72 that the initial registration content has not been changed, in step S73, the operation log acquisition unit 46 determines whether an operation log has been supplied from the television display device 11 or the recording / playback device 12. to decide. If it is determined in step S73 that no operation log is supplied, the process returns to step S72, and the subsequent processes are repeated.

  If it is determined in step S73 that an operation log has been supplied, in step S74, the operation log acquisition unit 46 determines whether or not the supplied operation log is a positive history. For example, when the operation log is a recording operation, the program vector PP of the program corresponding to the operation becomes a positive history, and when the operation log is erasure of recorded data that has not been reproduced, the program corresponding to the operation The program vector PP has a negative history.

  If it is determined in step S74 that the supplied operation log is a positive history, in step S75, the operation log acquisition unit 46 stores the program vector PP corresponding to the operation log determined to be a positive history. Extracted from the acquisition unit 41 and supplied to the positive history storage unit 47. The main history storage unit 47 additionally stores the supplied program vector PP as a main history.

  In step S76, the positive history storage unit 47 obtains the sum of the vectors of the positive history program vector PP for each detailed item or each large item, and generates a positive history vector UP. After the process of step S76 is completed, the process returns to step S72, and the subsequent processes are repeated.

  If it is determined in step S74 that the supplied operation log is not a positive history, the supplied operation log is a negative history. Therefore, in step S77, the operation log acquisition unit 46 has a negative history. The program vector PP corresponding to the determined operation log is extracted from the data acquisition unit 41 and supplied to the negative history storage unit 48. The negative history storage unit 48 additionally stores the supplied program vector PP as a negative history.

  In step S78, the negative history storage unit 48 obtains the sum of the vectors of the negative history program vector PP for each detailed item or each large item, and generates a negative history vector MUP. After the end of step S78, the process returns to step S72, and the subsequent processes are repeated.

  If the major item is the positive history vector UP = {title Tup, genre Gup, performer Pup, screenplay / original / production Aup, content (keyword) Kup}, and the detailed item is described in the major item, In the positive history vector UP, a numerical value indicating the vector sum is written after each detailed item. For example, as shown in FIG. 15, for the large item “genre”, the genre Gup = {(drama-25), (variety-34), (sports-42), (movie-37), (music-73). ), (For children / education-120), (education / document-3), (news / report-5), (others-23)}, the numerical value indicating the sum of the positive history is described for each detailed item. Is done. In addition, in a vector indicated by a word such as the large item “title”, for example, title Tup = {(title1-12), (title2-3),. A numerical value indicating the total sum of positive histories for each word is described. Similarly to the positive history vector UP, the negative history vector MUP also describes a numerical value indicating the sum following the item.

  In FIG. 15, the major items of the positive history vector UP (and negative history vector MUP) are the title, genre, performer, screenplay / original / production, content (keyword), and the program vector described with reference to FIG. Although it has been described that the number of items is smaller than that of the PP, it is needless to say that the same large items as the program vector PP may be used.

  In FIG. 14, it is described that the negative history vector MUP is generated before the operation log is input based on the contents of the initial registration. It is also possible to receive registration of information for generating the positive history vector UP before the operation log is input. Further, the positive history vector UP or the negative history vector MUP may not be generated based on the initial registration, and the positive history vector UP or the negative history vector MUP may be generated using only the operation log.

  In this way, the positive history vector UP and the negative history vector MUP are independently generated and held, whereby the matching process with the user's preference can be performed more precisely.

  At this time, the positive history and the negative history may be obtained more precisely. For example, in the process described with reference to FIG. 14, the positive history vector UP and the negative history vector MUP are generated using the sum of the program vectors PP corresponding to the positive history and the negative history in all items. For example, the total sum of the program vectors PP corresponding to the negative history may be accumulated for each genre, and the positive history vector UP and the negative history vector MUP may be generated for each genre.

  Depending on the genre of the broadcast program, the performers are often biased, so the user's preferences may not be correctly reflected. Specifically, users who like dramas and prefer only those with comedian A, who rarely appear in dramas, appear without distinguishing genres in users who have a 2: 8 audience ratio of variety versus drama. When accumulating the positive history of the performer, performer B who frequently appears in the drama is a higher point in the positive history vector UP than the comedian A, who hardly appears in the drama, even though he is not a favorite actor. May end up. In such a case, for example, a documentary in which a performer B who frequently appears in dramas appears is recommended rather than a variety in which comedian A appears. In order to prevent this, a positive history and a negative history are accumulated for each genre, and based on this, a positive history vector UP and a negative history vector MUP may be generated for each genre.

  Also, for example, since the user is a fan of commentator A, when he likes to watch “the live broadcast of the game of team B explained by commentator A”, the genre is “sports”. Information is easy to overlap as history, but the information of the commentator “Explanator A” is difficult to overlap as history, so it is recommended that live commentary of the game of Team B that other commentators are commenting on is recommended. There is a case where a variety program in which is appearing is not recommended. In order to prevent this, for example, a positive history and a negative history are accumulated for each performer, and based on this, a positive history vector UP and a negative history vector MUP may be generated for each performer.

  As described above, by storing the history for each specific element, it is possible to reflect the user's preference more precisely without easing the user's preference.

  Then, in the matching processing unit 43, the recommended program information that correctly reflects the user's preference by verifying the matching between the positive history vector UP and the negative history vector MUP generated in this way and the supplied program vector PP. Can be generated.

  Next, the positive history vector and negative history vector generation processing 2 in which the history is accumulated by genre will be described with reference to the flowchart of FIG.

  In steps S81 to S84, the same processing as in steps S71 to S74 of FIG. 14 is executed. That is, the initial registration is referred to, the negative history vector MUP is generated, and it is determined whether or not the initial registration contents have been changed. If not, the supplied operation log is a positive history. Is judged.

  If it is determined in step S84 that the supplied operation log is a positive history, in step S85, the operation log acquisition unit 46 acquires the program vector PP corresponding to the operation log determined to be a positive history. Extracted from the unit 41 and supplied to the positive history storage unit 47. The positive history storage unit 47 extracts the genre of the supplied program vector PP.

  In step S86, the main history storage unit 47 additionally stores the program vector PP extracted from the data acquisition unit 41 as a main history for each genre.

  In step S87, the correct history storage unit 47 obtains the sum of the vectors of the correct history program vector PP in the genre in which the program vectors are additionally stored for each detailed item or each large item, and for the corresponding genre. Generate positive history vector UP. After the process of step S87 ends, the process returns to step S82, and the subsequent processes are repeated.

  If it is determined in step S84 that the supplied operation log is not a positive history, the supplied operation log is a negative history. Therefore, in step S88, the operation log acquisition unit 46 is determined to be a negative history. The program vector PP corresponding to the operation log is extracted from the data acquisition unit 41 and supplied to the negative history storage unit 48. The negative history storage unit 48 extracts the genre of the supplied program vector PP.

  In step S89, the negative history storage unit 48 additionally stores the program vector PP extracted from the data acquisition unit 41 as a negative history for each genre.

  In step S90, the negative history storage unit 48 obtains the sum of the vectors of the negative history program vector PP in the genre in which the program vectors are additionally stored for each detailed item or each large item, and for the corresponding genre. A negative history vector MUP is generated. After the process of step S90 ends, the process returns to step S82, and the subsequent processes are repeated.

  By such processing, a positive history vector UP and a negative history vector MUP are generated for each genre, so that it is possible to reflect the user's preference more precisely without sloppy the user's preference. It is possible to generate recommended program information that is correctly reflected.

  With reference to the flowchart of FIG. 17, the matching process 1 in the case where the program vector PP, the positive history vector UP, and the negative history vector MUP are represented by vectors in which all elements of the detailed items are arranged in one column will be described. .

  In step S <b> 101, the program vector extraction unit 42 extracts program vectors PP of a plurality of programs (for example, programs broadcast in a predetermined time zone) from the data acquisition unit 41, and normalization processing unit 61 of the matching processing unit 43. To supply. The normalization processing unit 61 performs normalization on the title and content composed of words among the components of the supplied program vector PP and the positive history vector UP read from the positive history storage unit 47, and the vector The normalization result is supplied to the calculation unit 62.

  Specifically, when the supplied program vector PP is, for example, the title Tm = {Tokaido-1, Mitani-1, Kaidan-1}, the normalization processing unit 61 has the title Tm = {Tokaido: 0. 33, Mitani: 0.33, Kaidan: 0.33}, and normalize so that the total weight of the words in the item is 1 for each program.

  In step S102, the vector calculation unit 62 of the matching processing unit 43 calculates similarity SimUP, which is a cosine distance between the program vector PP of a plurality of programs and the positive history vector UP, using the above-described equation (1).

  In step S103, the vector calculation unit 62 compares the similarity SimUP indicating the similarity between the program vector PP and the positive history vector UP calculated in step S102. The program vector PP is extracted by the number of.

  In step S104, the vector calculation unit 62 calculates the similarity SimMUP, which is a cosine distance between the program vector PP extracted in step S103 and the negative history vector MUP read from the negative history storage unit 48, using the equation (2) described above. Calculate using.

  In step S105, the vector calculation unit 62 calculates a similarity (ie, cosine distance) SimUP-similarity (ie, cosine distance) SimMUP with the positive history vector UP, and is a higher order. A program vector or EPG data of a number of programs (for example, one) is extracted as recommended information, output to the recommended information output unit 49, registered in the recommended program list 50, and the television display device 11 and the recording / playback device. 12 to finish the process.

  By such processing, when the program vector PP, the positive history vector UP, and the negative history vector MUP are expressed by vectors in which all the elements of the detailed items are arranged in one column, the program vector PP and the positive history vector UP It is possible to determine a recommended program that matches the user's preference based on the similarity between the program vector PP and the negative history vector.

  Next, referring to the flowchart of FIG. 18, instead of normalizing the items indicated by words, the cosine distance is calculated for each large item, and the sum is calculated as similarity SimUP and similarity SimMUP. The matching process 2 for determining the recommended program will be described.

  In step S111, the program vector extraction unit 42 extracts program vectors PP of a plurality of programs (for example, programs broadcast in a predetermined time zone) from the data acquisition unit 41, and sends them to the vector calculation unit 62 of the matching processing unit 43. Supply. The vector calculation unit 62 calculates the cosine distance between the program vector PP and the positive history vector UP for each major item of the supplied program vector PP and the positive history vector UP read from the positive history storage unit 47.

  In step S112, the vector calculation unit 62 calculates the similarity SimUP by summing up the cosine distance values calculated for each item in step S111 using the above-described equation (4).

  In step S113, the vector calculation unit 62 compares the similarity SimUP between the program vector PP and the positive history vector UP calculated in step S112, and determines a predetermined number of programs such as 10 from the top of the similarity. Extract vector PP.

  In step S114, the vector calculation unit 62 determines the program vector PP and the negative history vector MUP for each major item of the program vector PP extracted by the process of step S113 and the negative history vector MUP read from the negative history storage unit 48. Cosine distance is calculated.

  In step S115, the vector calculation unit 62 calculates the similarity SimMUP by summing up the cosine distance values calculated for each item in step S114 using the above-described equation (4).

  In step S116, the vector calculation unit 62 calculates {similarity SimUP which is a cosine distance between the program vector PP and the positive history vector UP} − {similarity SimMUP which is a cosine distance between the program vector PP and the negative history vector}. Then, the program vector PP or EPG data of a predetermined number (for example, three) programs that are higher in rank are extracted as recommended information, output to the recommended information output unit 49, registered in the recommended program list 50, and television. The data is output to the display device 11 and the recording / playback device 12, and the process is terminated.

  By such processing, the sum of the calculation results for each large item is calculated as similarity SimUP and similarity SimMUP without normalizing the items indicated by words, so detailed elements belonging to different large items can be compared. Based on the similarity between the program vector PP and the positive history vector UP and the similarity between the program vector PP and the negative history vector MUP without being affected by the bias of the history overlap, a recommendation that matches the user's preference The program can be determined.

  In the matching process 1 and the matching process 2 described with reference to FIGS. 17 and 18, a program that the user seems not to like using a negative history is selected from programs having a high similarity to the user's positive history vector UP. Although it has been described that it can be removed, for example, the recommended program may be determined using only the positive history.

  In addition, some users may like programs such as news or news programs very much when selecting a program, performers are more important than genres, and content is important, although they are not particular about performers. In some cases, items that are weighted to select the item and items that are not weighted, in other words, important items and items that are not weighted are fixed.

  Therefore, the matching process may be performed using the above-described program-side effect vector EfPP, user-side effect vector EfUP, or user-side effect vector EfMUP. Further, whether to use the program side effect vector EfPP, the user side effect vector EfUP, or the user side effect vector EfMUP may be set by the user.

  Next, referring to the flowchart of FIG. 19, when the program vector PP, the positive history vector UP, and the negative history vector MUP are represented by vectors in which all the elements of the detailed items are arranged in one column, the user setting Thus, the matching process 3 for performing the matching process using the program side effect vector EfPP or the user side effect vector EfUP will be described.

  In step S121, the vector calculation unit 62 is input by the user using the operation input unit 44, and is registered in the initial registration storage unit 45. The program side effect vector EfPP and the user side effect vector EfUP or the user side effect vector EfMUP. Get the usage settings for. The effect vector usage setting content is information indicating whether or not weighting is performed using the program side effect vector EfPP, the user side effect vector EfUP, or the user side effect vector EfMUP in the matching process.

  In step S122, the vector calculation unit 62 reads the user side effect vector EfUP from the user information registration unit 63 as necessary, and uses the following equation (5) to cosine the program vector PP and the positive history vector UP. The distance is calculated and the similarity is SimUP.

  In equation (5), program vector PP = (p1, p2,...), Positive history vector UP = (u1, u2,...), Program side effect vector EfPP = (epd1, epd2,. ..) And the user side effect vector EfUP = (eud1, eud2,...). Further, in the equation (5), it is described that both the program side effect vector EfPP and the user side effect vector EfUP are used, but depending on the setting, of the program side effect vector EfPP and the user side effect vector EfUP, When any of the above is not used, the numerical value “1” is substituted for the unused vector and calculated.

  Further, the user side effect vector EfUP may be set by the user, may be set based on the initial setting of the user, or may be generated by the user information registration unit 63. good. Details of generation of the user-side effect vector EfUP will be described later with reference to FIGS.

  In step S123, the vector calculation unit 62 compares the similarity SimUP between the program vector PP and the positive history vector UP calculated in step S122, and determines a predetermined number of programs such as 10 from the top of the similarity. Extract vector PP.

  In step S124, the vector calculation unit 62 reads the user-side reaction vector EfMUP from the user information registration unit 63 as necessary, and uses the following equation (6) to extract the program vector PP extracted in step S123. And the cosine distance of the negative history vector MUP.

  In equation (6), program vector PP = (p1, p2,...), Negative history vector MUP = (m1, m2,...), Program side effect vector EfPP = (epd1, epd2,. ..., And user side reaction vector EfMUP = (emd1, emd2,...). Further, in the equation (6), it is described that both the program side effect vector EfPP and the user side reaction vector EfMUP are used. However, depending on the setting, the program side effect vector EfPP and the user side reaction vector EfMUP are used. When any one of the above is not used, the numerical value “1” is substituted for calculation instead of the unused vector.

  Further, the user side reaction vector EfMUP may be set by the user, may be set based on the initial setting of the user, or is generated by the user information registration unit 63. Also good. Details of generation of the user side reaction vector EfMUP will be described later with reference to FIG. 26 or FIG.

  In step S125, the vector calculation unit 62 calculates the similarity SimUP between the program vector PP and the positive history vector UP—the similarity SimMUP between the program vector PP and the negative history vector, and a predetermined number (for example, three) higher than that. ) Is extracted as recommendation information, output to the recommendation information output unit 49, registered in the recommended program list 50, and output to the television display device 11 and the recording / playback device 12 The process is terminated.

  Through such processing, the recommendation information is extracted using the program side effect vector EfPP, the user side effect vector EfUP, or the user side reaction vector EfMUP according to the setting, so that the user's preference is correctly reflected. The recommended program can be recommended.

  In the processing described with reference to FIG. 19, the processing in the case where the program vector PP, the positive history vector UP, and the negative history vector MUP are represented by vectors in which all elements of the detailed items are arranged in one column has been described. The program vector PP, the positive history vector UP, and the negative history vector MUP may be calculated for each major item.

  Next, referring to the flowchart of FIG. 20, the matching process 4 is performed so that the program side effect vector EfPP, the user side effect vector EfUP, or the user side effect vector EfMUP can be reflected for each major item. Will be described.

  In step S131, processing similar to that in step S121 of FIG. 19 is executed, and the usage setting content of the effect vector is acquired.

  In step S132, the vector calculation unit 62 calculates the cosine distance between the program vector PP and the positive history vector UP for each major item of the supplied program vector PP and the positive history vector UP read from the positive history storage unit 47. To do. Here, the effect vector is not used for the calculation.

  In step S133, the vector calculation unit 62 uses the following equation (7) to multiply the cosine distance calculated for each item by an effect vector as necessary, and sums the obtained values to obtain a similarity. Calculate the degree SimUP.

SimUP ＝ epdt ・ eudt ・ cosθut + epdg ・ eudg ・ cosθug + epdp ・ eudp ・ cosθup
+ epda, euda, cosθua + epdk, eudk, cosθuk (7)

  In equation (7), the cosine distance for each major item of program vector PP = (pt, pg, pp, pa, pk) and positive history vector UP = (ut, ug, up, ua, uk) , (Cosθut, cosθug, cosθup, cosθua, cosθuk), program side effect vector EfPP = (epdt, epdg, epdp, epda, epdk), user side effect vector EfUP = (eudt, eudg, eudp, euda, eudk) Suppose that Further, in the equation (7), it is described that both the program side effect vector EfPP and the user side effect vector EfUP are used, but depending on the setting, of the program side effect vector EfPP and the user side effect vector EfUP, If any one of the above is not used, the numerical value “1” is substituted for calculation instead of the unused vector.

  In step S134, the vector calculation unit 62 compares the similarity SimUP between the program vector PP and the positive history vector UP calculated in step S133, and determines a predetermined number of programs such as 10 from the top of the similarity. Extract vector PP.

  In step S135, the vector calculation unit 62 determines the program vector PP and the negative history vector MUP for each major item of the program vector PP extracted by the process of step S134 and the negative history vector MUP read from the negative history storage unit 48. Cosine distance is calculated. Here, the effect vector is not used for the calculation.

  In step S136, the vector calculation unit 62 uses the following equation (8) to multiply the cosine distance calculated for each item by an effect vector as necessary, and sums the obtained values to obtain a similarity. Calculate the degree SimMUP.

SimMUP ＝ epdt ・ emdt ・ cosθmt + epdg ・ emdg ・ cosθmg + epdp ・ emdp ・ cosθmp
+ epda ・ emda ・ cosθma + epdk ・ emdk ・ cosθmk (8)

  In equation (8), the cosine distance for each major item of program vector PP = (pt, pg, pp, pa, pk) and negative history vector UP = (mt, mg, mp, ma, mk) , (Cosθmt, cosθmg, cosθmp, cosθma, cosθmk), program-side effect vector EfPP = (epdt, epdg, epdp, epda, epdk,), user-side reaction vector EfMUP = (emdt, emdg, emdp, emda, emdk). Further, in the equation (8), it is described that both the program side effect vector EfPP and the user side reaction vector EfMUP are used, but depending on the setting, the program side effect vector EfPP and the user side reaction vector EfMUP When any one of the above is not used, the numerical value “1” is substituted for calculation instead of the unused vector.

  In step S137, the vector calculation unit 62 calculates {similarity SimUP which is a cosine distance between the program vector PP and the positive history vector UP} − {similarity SimMUP which is a cosine distance between the program vector PP and the negative history vector}. Then, the program vector PP or EPG data of a predetermined number (for example, three) programs that are higher in rank are extracted as recommended information, output to the recommended information output unit 49, registered in the recommended program list 50, and television. The data is output to the display device 11 and the recording / playback device 12, and the process is terminated.

  With such a process, the effect vector is used for each large item and weighting is performed, so that it is possible to generate recommendation information that matches the user's preference in detail.

  Next, referring to the flowchart of FIG. 21, the positive history vector UP and negative history vector MUP for each genre generated by the processing of the positive history vector and negative history vector generation processing 2 described with reference to FIG. The matching process 5 for executing the matching process using the user-side effect vector EfUP and the user-side reaction vector EfMUP for each genre will be described.

  In step S141, processing similar to that in step S121 of FIG. 19 is executed, and the effect vector usage setting content is acquired.

  In step S142, the vector calculation unit 62 extracts the genre of the supplied program vector PP. Here, for example, it is assumed that the genre of the supplied program vector PP is “drama”.

  In step S143, the vector calculation unit 62 sets the program vector PP and the correct program vector PP for each major item of the supplied program vector PP and the correct history vector UP corresponding to the genre “drama” read from the correct history storage unit 47. Calculate cosine distance of history vector UP. Here, the effect vector is not used for the calculation.

  In step S144, the vector calculation unit 62 multiplies the cosine distance calculated for each item by the user-side effect vector corresponding to the genre “drama” using the following equation (9), The obtained values are summed to calculate the similarity SimUP.

SimUP ＝ epdt ・ eudtd ・ cosθutd + epdg ・ eudgd ・ cosθugd + epdp ・ eudpd ・ cosθupd
+ epda, eudad, cosθuad + epdk, eudkd, cosθukd (9)

  In Equation (9), program vector PP = (pt, pg, pp, pa, pk) and positive history vector UP = (ut, ug, up, ua, uk) corresponding to the genre “drama”. The cosine distance for each major item is (cosθutd, cosθugd, cosθupd, cosθuad, cosθukd), the program side effect vector EfPP = (epdt, epdg, epdp, epda, epdk), and the user side corresponding to the genre “drama” It is assumed that the effectiveness vector EfUP = (eudtd, eudgd, eudpd, eudad, eudkd). Further, in the equation (9), it is described that both the program side effect vector EfPP and the user side effect vector EfUP are used, but depending on the setting, of the program side effect vector EfPP and the user side effect vector EfUP, If any one of the above is not used, the numerical value “1” is substituted for calculation instead of the unused vector.

  In step S145, the vector calculation unit 62 compares the similarity SimUP between the program vector PP and the positive history vector UP calculated in step S144, and a predetermined number of programs such as 10, for example, from the top of the similarity. Extract vector PP.

  In step S146, the vector calculation unit 62 performs the program for each major item of the program vector PP extracted by the process of step S145 and the negative history vector MUP corresponding to the genre “drama” read from the negative history storage unit 48. The cosine distance between the vector PP and the negative history vector MUP is calculated. Here, the effect vector is not used for the calculation.

  In step S147, the vector calculation unit 62 multiplies the cosine distance calculated for each item by the effect vector corresponding to the genre “drama” as necessary using the following equation (10). Similarity SimMUP is calculated by summing the values obtained.

SimMUP ＝ epdt ・ emdtd ・ cosθmtd + epdg ・ emdgd ・ cosθmgd + epdp ・ emdpd ・ cosθmpd
+ epda ・ emdad ・ cosθmad + epdk ・ emdkd ・ cosθmkd (10)

  In equation (10), the cosine distance for each major item of the program vector PP = (pt, pg, pp, pa, pk) and the negative history vector MUP = (mt, mg, mp, ma, mk). , (Cosθmtd, cosθmgd, cosθmpd, cosθmad, cosθmkd), program side effect vector EfPP = (epdt, epdg, epdp, epda, epdk), user side reaction vector EfMUP = (emdtd, emdgd, emdpd, emdad, emdkd). Further, in the equation (10), it is described that both the program side effect vector EfPP and the user side reaction vector EfMUP are used. However, depending on the setting, the program side effect vector EfPP and the user side reaction vector EfMUP are used. When any one of the above is not used, the numerical value “1” is substituted for calculation instead of the unused vector.

  In step S148, the vector calculation unit 62 calculates {similarity SimUP which is a cosine distance between the program vector PP and the positive history vector UP} − {similarity SimMUP which is a cosine distance between the program vector PP and the negative history vector}. Then, the program vector PP or EPG data of a predetermined number (for example, three) programs that are higher in rank are extracted as recommended information, output to the recommended information output unit 49, registered in the recommended program list 50, and television. The data is output to the display device 11 and the recording / playback device 12, and the process is terminated.

  By such processing, the cosine distance between the positive history vector UP and the negative history vector MUP and the program vector PP for each genre is obtained for each major item, and weighting is performed using the effect vector corresponding to the genre. Since the degree of similarity is calculated, it is possible to generate recommendation information that matches the user's preference in detail.

  Further, as described above, the user side effect vector EfUP and the counter effect vector EfMUP may be generated based on the initial registration contents of the user registered in the initial registration storage unit 45, or the positive history The user-side user-effect vector EfUP and the counter-effect vector EfMUP may be generated by counting the vector UP, the negative history vector MUP, or the programs viewed by the user during a certain period.

  With reference to the flowchart of FIG. 22, the user-side effect vector generation process 1 for generating the user-side effect vector EfUP by counting the programs viewed by the user during a certain period will be described.

  In step S151, the user information registration unit 63 of the matching processing unit 43 selects one of unprocessed large items.

  In step S152, the user information registration unit 63 refers to the main history stored in the main history storage unit 47, and for example, shows a program viewed by the user during a certain period such as one week, one month, or three months. And the program vector extraction unit 42 extracts the program vector PP corresponding to the program viewed by the user during a certain period from the data acquisition unit 41, and for each detailed item included in the large item selected in step S151. Count the number of programs.

  Specifically, for example, the large item selected in step S151 is the large item genre Gm = {drama, variety, sports, movie, music, children / education, education / documents, news / reports, etc.} For example, the user information registration unit 63 classifies and counts the constituent elements of the program vector PP of the program viewed by the user during a certain period of time into corresponding items. When the number of programs viewed by the user during a certain period is 50, for example, the count result of the number of programs is, for example, genre Gm = (10, 18, 5, 2, 8, 1, 0, 1, 5 )

  In step S153, the user information registration unit 63 causes the program vector extraction unit 42 to extract program vectors PP corresponding to all programs in the same period from the data acquisition unit 41, and includes them in the large items selected in step S151. The number of programs is counted for each detailed item.

  Specifically, for example, the large item selected in step S151 is the large item genre Gm = {drama, variety, sports, movie, music, children / education, education / documents, news / reports, etc.} For example, the user information registration unit 63 classifies the constituent elements of the program vectors PP of all programs in the same period into the corresponding items and counts them. When all the programs in the same period are 1000 programs, for example, the count result of the number of programs is, for example, genre Gm = (104, 239, 68, 25, 78, 91, 60, 254, 81). .

  In step S154, the user information registration unit 63 calculates the count number of the viewing results of the user / the count number of all the programs based on the count results of step S152 and step S153.

  Program organization is considered to reflect the taste of the masses due to the influence of audience rating competition. In other words, the calculation of the user's viewing record count / all program counts is synonymous with normalizing the user's viewing record counts with all program counts as a standard model. . The normalized vector calculated in step S154 is referred to as a normalized vector D.

For example, in the large item genre Gm = {drama, variety, sports, movie, music, children / education, education / documents, news / reports, etc.}, the count number of all programs for one week is (8, 12, 3,7,6,4,2,8,10) and the number of programs viewed by the user is (4,0,1,2,3,4,1,2,2). In this case, the normalized vector D is as follows.
D = (4/8, 0/12, 1/3, 2/7, 3/6, 4/4, 1/2, 2/8, 2/10) = (0.5, 0, 0.33 , 0.28, 0.5, 1.0, 0.5, 0.25, 0.2)
That is, when the component of the normalization vector D is 1.0, that all the programs of the corresponding item have been viewed during the set period, and when the component of the normalization vector D is 0, It means that no corresponding program was watched during the set period.

  In step S155, the user information registration unit 63 generates a corresponding large item effect vector based on the calculation result in step S154.

  In order to generate an effect vector, using one of the items of the large item genre Gm = {drama, variety, sports, movies, music, children / education, education / documents, news / reports, etc.} Set the standard value. For example, in a program of the genre: “drama” for one week, it is sufficient that a general user views about 20% of the program and sets the standard value to 0.2. Since the effect vector of the large item is calculated as a relative value, the set value may be any value from 0 to 1. The user side effect vector is a relative value between the normalized vector D calculated in step S154 and the set value.

Therefore, the effect vector E of the large item genre Gm indicating the user's interest genre is
E = (0.3, -0.2, 0.13, 0.08, 0.3, 0.8, 0.3, -0.07, 0.0)
Therefore, it can be determined that the corresponding user likes the genre of “for children / education” and does not like the genre of “variety”.

  In step S156, the user information registration unit 63 determines whether or not all large item effect vectors have been generated. If it is determined in step S156 that not all large effect vectors have been generated, the process returns to step S151, and the subsequent processes are repeated.

  If it is determined in step S156 that all large item effect vectors have been generated, in step S157, the user information registration unit 63 stores all large item effect vectors, and the process ends.

  By such processing, a difference between general preference and user-specific preference can be obtained. In addition, by recalculating the user-side effect vector EfUP every predetermined period such as three months or six months, it is possible to recommend a program that reflects the user's preferences in real time.

  Further, in the processing of FIG. 22, it has been described that the user-side effect vector EfUP is obtained based on a program viewed by the user during a certain period such as one week, one month, or three months. For example, the user side effect vector EfUP corresponding to the short term, the medium term, and the long term may be calculated, and the recommendation information may be determined using the plurality of effect vectors.

  Although the case where the user-specific preference is used as the user-side effect vector EfUP has been described here, the user-specific preference may be used as the positive history vector UP for the matching process.

  Further, instead of all programs being broadcast, all programs broadcast in a predetermined time zone (for example, so-called golden time from 18:00 to 22:00) when the viewer views the program most are counted. May be. By doing in this way, the arithmetic processing for calculating | requiring general preference can be reduced significantly.

  Next, referring to the flowchart of FIG. 23, the difference between the user-specific preference and the mass preference is used for matching by calculating the cosine distance indicating the similarity between the positive history vector UP and the mass preference. The user-side effect vector calculation process 2 for obtaining the user-side effect vector EfUP will be described.

  In step S <b> 161, the user information registration unit 63 of the matching processing unit 43 acquires the positive history vector UP stored in the positive history storage unit 47.

  In step S162, the user information registration unit 63 acquires a standard preference vector APP that indicates a general preference.

  The standard preference vector APP may be supplied from the distribution server 5, or the program organization is considered to reflect the preference of the masses due to the influence of the audience rating competition. Similarly to the user-side effect vector calculation process 1 described above, the contents of all programs broadcast for a certain period may be counted and normalized as necessary to obtain the standard preference vector APP.

  In the distribution server 5, for example, a standard preference vector APP indicating a general preference may be generated using a general audience rating survey or other methods.

  In step S163, the user information registration unit 63 calculates the cosine distance between the standard preference vector APP and the positive history vector UP for each large item. The greater the cosine distance, the higher the similarity between the standard preference vector APP and the positive history vector UP.

  In step S164, based on the cosine distance calculated in step S163, the user information registration unit 63 generates an effect vector EfUP by reciprocating the cosine distance for each major item, and the process is terminated. The greater the reciprocal of the cosine distance, the lower the similarity between the standard preference vector APP and the positive history vector UP.

  By such processing, the user-side effect vector EfUP reflecting the difference between the general preference and the corresponding user-specific preference can be obtained. When the program recommendation process is performed using the user-side effect vector EfUP, the recommended program is determined with emphasis on the difference between the user's preference and the general preference.

  Here, the program vector PP and the positive history vector UP have been described as being represented by vectors for each major item. However, the program vector PP and the positive history vector UP are all elements of the detailed items. It goes without saying that the same processing can be executed even when the vectors are represented by vectors arranged in columns.

  Further, the similarity between the standard preference vector APP and the positive history vector UP may be used not only for calculating the effectiveness vector but also directly for program recommendation as an index indicating the uniqueness of the user. . For example, when the similarity between the standard preference vector APP and the positive history vector UP is high, a new program or the like that matches the general trend may be preferentially recommended.

  As described with reference to FIGS. 22 and 23, the user side effect vector EfUP is preferably obtained by learning based on the user's operation history, but the user side effect vector is registered in advance as an initial registration. Alternatively, a preset value obtained by experience or the like may be used.

  Note that the user-side effect vector EfUP may be generated not only by generating the user-side effect vector EfUP by paying attention to the large item but also by paying attention to the components constituting the large item. For example, in the component “Performer Pm” constituting the large item, the main role and the supporting role can be distinguished from each other, and in a drama or a movie, the user who gives priority to the casting of the supporting role over the leading character The user side effect vector EfUP can be set so as to increase the weight, and in the component “screenplay / original / production Am” constituting the large item, the director, director, original author, photographer, etc. can be distinguished. A user who places more importance on the cameraman than the director or director may be able to set the user-side effect vector EfUP so as to increase the weight of the cameraman.

  Further, a user-side effect vector EfUP is generated for each genre, and is used when matching the corresponding genre positive history vector UP and program vector PP as in the matching process 5 described with reference to FIG. May be.

  Next, a user side effect vector generation process 3 for generating a user side effect vector EfUP by counting programs viewed by the user during a certain period by genre will be described with reference to the flowchart of FIG.

  In step S171, the user information registration unit 63 of the matching processing unit 43 selects any genre in order to count programs viewed by the user for a certain period by genre.

  In step S172, the user information registration unit 63 selects one of the unprocessed large items.

  In step S173, the user information registration unit 63 refers to the main history stored in the main history storage unit 47, and for example, shows a program viewed by the user during a certain period such as one week, one month, or three months. Of the selected genre is detected, and the program vector extraction unit 42 extracts, from the data acquisition unit 41, the program vector PP corresponding to the program of the selected genre among those viewed by the user during a certain period. The number of programs is counted for each detailed item included in the large item selected in step S172.

  In step S174, the user information registration unit 63 causes the program vector extraction unit 42 to extract the program vector PP corresponding to the selected genre from all the programs in the same period from the data acquisition unit 41, and in step S172. The number of programs is counted for each detailed item included in the selected large item.

  In step S175, the user information registration unit 63 calculates the count number of the user's viewing performance / the count number of all programs in the selected genre based on the count results in step S173 and step S174.

  As described above, the program organization is considered to reflect the popular preference due to the influence of audience rating competition. In other words, the calculation of the count of the user's viewing performance / the count of all the programs in the selected genre is, in other words, the count of the user's viewing performance in the corresponding genre. This is synonymous with normalization by the count of all programs. The normalized vector calculated in step S175 is referred to as a normalized vector D ′.

For example, the large program vector PP corresponding to the genre of “drama” in the large category genre Gm = {drama, variety, sports, movie, music, children / education, education / documents, news / reports, etc.} In the item time zone Tm = {morning, noon, evening, golden, midnight}, the count number of all the programs for one week is (10, 35, 7, 53, 17), and the program viewed by the user is displayed. When the count number is (5, 0, 0, 8, 4), the normalized vector D ′ is as follows.
D ′ = (5/10, 0/35, 0/7, 8/53, 4/17)
= (0.5, 0, 0, 0.28, 0.15, 0.24)
That is, when the component of the normalized vector D ′ is 1.0, it means that all programs of the corresponding item have been viewed during the set period, and the component of the normalized vector D ′ is 0. In this case, it means that no corresponding program was viewed during the set period.

  In step S176, the user information registration unit 63 generates an effect vector for the corresponding large item in the selected genre based on the calculation result in step S175.

  In order to generate an effect vector, a standard value is set using any item of the large item time zone Tm = {morning, noon, evening, golden, midnight}. For example, in a one-week program of the time zone: “Golden”, a general user may watch about 20% of the drama and set the standard value to 0.2. Since the effect vector of the large item is calculated as a relative value, the set value may be any value from 0 to 1. The user side effect vector is a relative value between the normalized vector D ′ calculated in step S175 and the set value.

Therefore, the effect vector E ′ of the large item genre Gm indicating the user's genre is
E ′ = (0.3, −0.2, −0.2, −0.05, 0.04)
Therefore, it can be determined that the corresponding user likes the drama in the morning time zone and does not like the drama in the daytime or evening time zone.

  In step S177, the user information registration unit 63 determines whether or not the effect vectors for all large items have been generated in the selected genre. If it is determined in step S177 that not all large effect vectors have been generated, the process returns to step S172, and the subsequent processes are repeated.

  When it is determined in step S177 that all large item effect vectors have been generated, in step S178, the user information registration unit 63 determines whether or not the processing of all genres has been completed. If it is determined in step S178 that all genres have not ended, the process returns to step S171, and the subsequent processes are repeated.

  If it is determined in step S178 that all genres have been completed, in step S179, the user information registration unit 63 stores the effect vectors of all large items, and the process ends.

  By such processing, the difference between the general preference and the user-specific preference can be obtained for each genre. Similarly to the case described with reference to FIG. 22, by recalculating the user side effect vector EfUP every predetermined period such as 3 months, 6 months, etc., the user's preference can be changed in real time. The reflected program can be recommended.

  Also, in the processing of FIG. 24, the description has been made assuming that the user-side effect vector EfUP is obtained based on a program viewed by the user during a certain period such as one week, one month, or three months. In the same manner as described above, for a plurality of periods, for example, the user-side effect vector EfUP corresponding to the short-term, medium-term, and long-term is calculated, and the recommendation information is determined using the plurality of effect vectors. You may do it.

  Also, in the processing of FIG. 24, instead of all the broadcasted programs, the program is broadcast in a predetermined time zone (for example, the so-called golden time from 18:00 to 22:00) when the viewer views the program most. You may make it count all the programs.

  Next, referring to the flowchart of FIG. 25, the cosine distance calculation indicating the similarity between the positive history vector UP and the public preference is executed for each genre, so that the user-specific preference and the public preference are obtained. The user side effect vector generation process 4 for obtaining the user side effect vector EfUP for using the gap for matching will be described.

  In step S191, the user information registration unit 63 of the matching processing unit 43 selects one of the genres in order to specify a genre for executing the process.

  In step S 192, the user information registration unit 63 acquires the positive history vector UP of the selected genre from the positive history vectors UP stored in the positive history storage unit 47.

  In step S193, the user information registration unit 63 acquires the standard preference vector APP of the selected genre among the standard preference vectors APP indicating the general preference.

  As described above, the standard preference vector APP may be supplied from the distribution server 5, or the program organization is considered to reflect the popular preference due to the influence of audience rating competition. Similar to the user-side effect vector calculation process 3 described with reference to FIG. 24, the contents of all programs broadcast for a certain period are counted by genre, normalized as necessary, and the genre-specific standard preference vector APP You may make it do.

  In the distribution server 5, for example, a standard preference vector APP indicating a general preference may be generated for each genre using a general audience rating survey or other methods.

  In step S194, the user information registration unit 63 uses the positive history vector UP of the selected genre and the standard preference vector APP of the selected genre, for each major item, the standard preference vector APP and the positive history vector UP. Cosine distance is calculated. The greater the cosine distance, the higher the similarity between the standard preference vector APP and the positive history vector UP.

  In step S195, the user information registration unit 63 generates the effect vector EfUP of the selected genre by reversing the cosine distance for each major item based on the cosine distance calculated in step S194. The greater the reciprocal of the cosine distance, the lower the similarity between the standard preference vector APP and the positive history vector UP.

  In step S196, the user information registration unit 63 determines whether all genres have been processed. If it is determined in step S196 that all genres have not ended, the process returns to step S191, and the subsequent processes are repeated. If it is determined in step S196 that all genres have been completed, the process ends.

  By such processing, the user-side effect vector EfUP for each genre reflecting the difference between the general preference and the corresponding user-specific preference can be obtained.

  Similarly to the processing described with reference to FIG. 23, the similarity between the standard preference vector APP and the negative history vector MUP may be obtained, and the reciprocal thereof may be calculated as the reaction vector EfMUP.

  Next, the user-side reaction vector calculation process 1 for obtaining the reaction vector EfMUP by comparing the negative history vector MUP with the public preference will be described with reference to the flowchart of FIG.

  In step S <b> 201, the user information registration unit 63 of the matching processing unit 43 acquires the negative history vector MUP stored in the negative history storage unit 48.

  In step S202, the user information registration unit 63 acquires a standard preference vector APP that indicates a general preference.

  The standard preference vector APP may be supplied from the distribution server 5. Or, since the program organization is considered to reflect the preference of the public due to the influence of audience rating competition, it is broadcasted for a certain period in the same manner as the user side effect vector calculation processing 1 described with reference to FIG. The contents of all the programs may be counted and normalized as necessary to obtain the standard preference vector APP.

  In step S203, the user information registration unit 63 calculates the cosine distance between the standard preference vector APP and the negative history vector MUP for each large item. The greater the cosine distance, the higher the similarity between the standard preference vector APP and the negative history vector MUP.

  In step S204, based on the cosine distance calculated in step S203, the user information registration unit 63 generates a reaction eye vector EfMUP by reversing the cosine distance for each major item, and the process is terminated.

  By such processing, the reaction vector EfMUP can be generated, so that programs that the user does not like can be effectively omitted from the recommended programs.

  Next, the user-side reaction vector generation process 2 executed for each genre will be described with reference to the flowchart of FIG.

  In step S <b> 211, the user information registration unit 63 of the matching processing unit 43 selects any genre in order to specify a genre for executing the process.

  In step S212, the user information registration unit 63 acquires the negative history vector MUP of the selected genre among the negative history vectors MUP stored in the negative history storage unit 48.

  In step S213, the user information registration unit 63 acquires the standard preference vector APP of the selected genre among the standard preference vectors APP indicating the general preference.

  In step S214, the user information registration unit 63 determines, based on the negative history vector MUP of the selected genre and the standard preference vector APP of the selected genre, the standard preference vector APP and the negative history vector MUP for each major item. Cosine distance is calculated. The greater the cosine distance, the higher the similarity between the standard preference vector APP and the negative history vector MUP.

  In step S215, the user information registration unit 63 generates a reaction vector EfMUP by reversing the cosine distance for each major item based on the cosine distance calculated in step S214. The greater the reciprocal of the cosine distance, the lower the similarity between the standard preference vector APP and the negative history vector MUP.

  In step S216, the user information registration unit 63 determines whether or not the processing for all genres has been completed. If it is determined in step S216 that all genres have not ended, the process returns to step S211 and the subsequent processes are repeated. If it is determined in step S216 that all genres have been completed, the process ends.

  By such processing, the reaction vector EfMUP can be generated for each genre, so that programs that the user does not like can be effectively omitted from recommended programs.

  The user side effect vector EfUP and the counter effect vector EfMUP are values obtained by multiplying the reciprocal of the cosine distance for each major item described with reference to FIG. 23 and FIGS. 25 to 27 by n, or in predetermined digits. A value obtained by rounding off may be used, or a value obtained by subtracting the reciprocal of the cosine distance from 1 or a value obtained by multiplying the value by n may be used.

  Here, the program vector PP and the negative history vector MUP are described as being represented by vectors for each major item. However, the program vector PP and the negative history vector MUP are all columns of detailed items in one column. It goes without saying that the same processing can be executed even in the case where the vectors are represented by arranged vectors.

  When the program vector extracted by the program vector extraction unit 42 is added with information indicating a group such as a group ID or a cluster code generated by the processing described with reference to FIG. 7 or FIG. There is.

  For example, if the program that the user likes to watch is a continuous drama, the calculation processing for recommendation can be reduced if the corresponding continuous drama is recommended at all times. In addition, it is possible to reduce the calculation processing for recommendation by preferentially recommending a program of the same group as a program that is highly preferred by the user.

  Next, the matching process including group recommendation will be described with reference to the flowchart of FIG.

  In step S221, the vector calculation unit 62 of the matching processing unit 43 determines whether or not information indicating a group such as a group ID or a cluster code is added to the program vector PP supplied from the program vector extraction unit 42. In addition, it is determined whether or not the corresponding program is grouped.

  When it is determined in step S221 that the corresponding programs are grouped, in step S222, the vector calculation unit 62 determines that the group indicated by the group ID or the cluster code is the recommended priority group and the user information registration unit. It is determined whether or not it is registered in 63.

  If it is determined in step S222 that it is registered as a recommended priority group, in step S223, the vector calculation unit 62 supplies the corresponding program information to the recommendation information output unit 49 as recommendation information. The recommended information output unit 49 registers recommended programs in the recommended program list 50 and outputs recommended program information to the television display device 11 or the recording / playback device 12. The process proceeds to S227.

  If it is determined in step S221 that the corresponding program is not grouped, or if it is determined in step S222 that the corresponding program is not registered as a recommended priority group, in step S224, FIGS. 17 to 21 are used. One of the matching processes 1 to 5 described above is executed.

  In step S225, the vector calculation unit 62 indicates information indicating a group such as a group ID or a cluster code in the program vector of the program recommended in any of the matching processing 1 to the matching processing 5 executed in step S224. Based on whether or not is added, it is determined whether or not the recommended program is grouped. If it is determined in step S225 that the recommended program is not grouped, the process is terminated.

  When it is determined in step S225 that the recommended program is grouped, in step S226, the vector calculation unit 62 uses the group ID or cluster code added to the program vector as a recommended priority group. The user information is registered in the user information registration unit 63 and stored.

  After the process of step S223 or step S226 is completed, in step S227, the vector calculation unit 62 refers to the negative history operation log stored in the negative history storage unit 48 and is recommended by the process of step S223. For example, a recommended program is not accepted for a program registered as a recommended priority group in the process of step S226, because the programs are grouped among the recommended programs or the programs recommended by the process of step S224. In addition, it is determined whether or not an operation input for giving a negative history such as an operation input for instructing viewing or recording of other programs or an operation input for instructing to erase the automatically recorded program before reproduction is received. If it is determined in step S227 that an operation input that is a negative history has not been received, the process ends.

  If it is determined in step S227 that an operation input having a negative history has been received, in step S228, the vector calculation unit 62 removes the corresponding group from the recommended priority group registration in the user information registration unit 63, and performs processing. Is terminated.

  Note that there may be a case where a plurality of group IDs are associated with one program vector PP by the title grouping process 1 described with reference to FIG. In particular, in such a case, in step S228, all group IDs may be removed from the recommended priority group registration of the user information registration unit 63 by an operation input as a negative history. The number of operation inputs that become the history of the user ID is accumulated, and when a predetermined number of operation inputs that become a negative history are made, the corresponding group ID is obtained from the registration of the recommended priority group of the user information registration unit 63. It may be removed. The same applies to the case where one group ID is associated with one program vector PP.

  Through such a process, it is possible to recommend a series of dramas that the user likes to watch at all times, or to recommend a program of the same group as a program that the user likes very much. As a result, calculation processing for recommendation can be reduced.

  As described above, when matching processing including group recommendation is performed, the positive history vector UP and the negative history vector MUP generated based on the user's operation history acquired from the television display device 11 or the recording / playback device 12. May be generated for each corresponding group.

  Also, based on the positive history of the user's operation history acquired from the television display device 11 or the recording / playback device 12, the number of times of viewing or recording reservations in the same continuous drama is counted, and the predetermined number or more is reached. In this case, the matching process may not be particularly performed and the recommendation may be preferentially performed.

  Instead of performing the program recommendation process using the user history, a user model may be set based on a topic designated in advance by the user, and the program recommendation process may be performed based on the user model.

  The user model is obtained by extracting a program corresponding to a topic from a predetermined program group by performing filtering using a topic designated in advance by the user. The initial registration storage unit 45 stores topics input by the user. There may be a plurality of topics to be registered, and the topics are updated as appropriate according to user operation inputs. The vector calculation unit 62 extracts the program vector including the topic stored in the initial registration storage unit 45 from the program vector of the program to be created by the user model supplied from the program vector extraction unit 42, and the extracted program A user model vector is generated by normalizing the sum of the vectors as necessary, and registered in the user information registration unit 63.

  Thus, for example, when the time zone Hm = “after 23:00” and the genre Gm = “variety” are specified as topics, filtering is performed using these topics, and the user model “late-night variety” is generated. Is done. The user model “Midnight Variety” includes a comedian who appeared in a variety program broadcast in the late-night frame after 23:00 as a component, so it can be used in programs of a genre different from variety, such as dramas. Since a program in which the comedian appears is extracted and recommended in the matching process, it is possible to apply the user's preference across genres and the like as compared with the case where the matching process is performed for each item.

  Note that the user model creation target program may be, for example, all programs for a predetermined period or a group of programs broadcast in a predetermined time zone (for example, so-called golden time).

  Also, for example, by filtering on different program sets using the same filtering condition, user model vectors corresponding to details of different program organization, for example, at different times or different time zones, are generated. be able to.

  Specifically, even under the same condition of “genre Gm = music”, there is a difference in broadcast time and performers between the current program organization and the program organization 10 years ago, so “current music fan” “10 years” A different user model can be generated, called “Previous Music Fan”. This makes it possible for a pop singer that a “music fan 10 years ago” would like to listen to recommends a drama or movie that is currently acting as an actor, or a “music fan 10 years ago”. It is possible to recommend a program that the current singer sings the music that was being heard.

  Further, for example, the same “genre Gm” is used in a time zone such as 20:00 to 24:00 that is mainly viewed by a member of society and a time zone such as 15:00 to 20:00 that is mainly viewed by a child. Since different programs can be obtained by filtering even under the condition “= music”, different user models can be generated.

  The vector calculation unit 62 calculates the similarity between the program vector PP of the program supplied from the program vector extraction unit 42 and the user model vector registered in the user information registration unit 63, and based on the result. The recommendation information is generated and supplied to the recommendation information output unit 49. The recommended information output unit 49 registers the recommended information in the recommended program list 50 and supplies the recommended information to the television display device 11 or the recording / playback device 12.

  Next, the matching process using the user model will be described with reference to the flowchart of FIG.

  In step S231, the vector calculation unit 62 of the matching processing unit 43 acquires the program vector of the program to be created by the user model extracted by the program vector extraction unit 42. The user model creation target program is, for example, a program broadcast during a predetermined period such as one month or three months, and is broadcast during a predetermined time period (for example, golden time) during the predetermined period. It may be a program that has been broadcast for a predetermined period in the past, such as 10 years ago or 20 years ago.

  In step S232, the initial registration storage unit 45 acquires the topic input by the user from the operation input unit 44 and stores it. The vector calculation unit 62 reads the topic from the initial registration storage unit 45 and sets it as a filtering condition. The filtering condition may be described as a hash table, for example.

  In step S233, the vector calculation unit 62 filters the program vector PP acquired in step S231 based on the filtering condition set in step S232, and extracts a program vector PP that matches the topic. For example, when the user model creation target program is all programs for the past three months, the vector calculation unit 62 performs “title Tm or genre Gm = soccer” for all program vectors PP for the past three months. Filtering is performed under the condition.

  In step S234, the vector calculation part 62 calculates | requires the sum total of the program vector PP extracted by filtering in step S233, and makes it a user model vector. For example, when filtering is performed under the condition “title Tm or genre Gm = soccer”, a user model vector of a user model “soccer fan” is generated.

  In step S235, the vector calculation unit 62 becomes a candidate for obtaining recommendation information extracted by the user model vector generated in step S234 and the program vector extraction unit 42 (will be broadcast later). The cosine distance with the program vector PP of the program is calculated.

  In step S236, the vector calculation unit 62 compares the similarity between the user model vector and the program vector PP of a candidate program for obtaining recommendation information based on the cosine distance calculated in step S235, and the comparison result. Based on the above, a predetermined number of programs from the top among those having high similarity are extracted as recommendation information and supplied to the recommendation information output unit 49. The recommended information output unit 49 registers the recommended information in the recommended program list 50 and supplies the recommended information to the television display device 11 or the recording / reproducing device 12, and the processing is completed.

  By such processing, even when there is no user operation history, it is possible to recommend a program that matches the topic set by the user. Conventionally, for example, when a matching process is performed by setting a condition of “title Tm or genre Gm = soccer”, it is impossible to extract a variety or drama in which a soccer player appears. On the other hand, in the process described with reference to FIG. 29, by setting a user model with the topic “title Tm or genre Gm = soccer”, items such as performers and contents of the user model vector are set. However, since many soccer player names are included, even if the program does not meet the condition of “title Tm or genre Gm = soccer”, the variety or drama in which the soccer player appears is extracted. It becomes possible to recommend.

  Here, the cosine distance between the user model vector and the program vector is calculated to calculate the similarity, but for each major item, the cosine distance is calculated individually and the sum is used. The similarity may be obtained.

  Note that the user model vector generation process may be executed in the distribution server 5. In that case, for example, the processing of steps S231 to S234 of FIG. 29 may be executed using the program vector generated by the program vector generation unit 23 described with reference to FIG.

  In addition to recommending programs that match the user's preference, choosing a low-history vector UP or negative-history vector MUP that has a low degree of similarity does not like or dislike, that is, the user has ever viewed it. There is a high possibility that a program having a characteristic that has never been performed, in other words, a program that is not eaten and hated is recommended. As described above, in order to extract the user's preference and advance the learning, it is important that the program having the characteristics that the user has never watched is evaluated.

  This not only makes it possible to give unexpected results to the recommendation results provided to the user and expands the user's interest, but is also very important for recommending programs that better match the user's preferences. History information can be acquired.

  Next, the exceptional recommended program selection process will be described with reference to the flowchart of FIG.

  In step S241, the vector calculation unit 62 of the matching processing unit 43 includes the program vector PP supplied from the program vector extraction unit 42, the positive history vector UP stored in the positive history storage unit 47, and the negative history storage unit. 48, the cosine distance between the positive history vector UP and the program vector PP and the cosine distance between the negative history vector MUP and the program vector PP for each major item using the negative history vector MUP stored in FIG. calculate.

  In step S242, the vector calculation unit 62 adds the cosine distance values calculated for each item on the positive history side and the negative history side, respectively. In other words, the above-described similarity SimUP between the positive history vector UP and the program vector PP and the similarity SimMUP between the negative history vector MUP and the program vector PP are calculated by the processing in steps S241 and S242.

  In step S243, the vector calculation unit 62 calculates an exceptional recommended value indicating a low degree of similarity between the positive history vector UP and the negative history vector MUP.

  Specifically, the exceptional recommendation value can be obtained by (1-SimUP) × (1-SimMUP) or (1 / SimUP) × (1 / SimMUP).

  In step S244, the vector calculation unit 62 obtains a program with a high exceptional recommendation value based on the calculation result in step S243, extracts it as recommendation information, and the process ends.

  By such processing, it is possible to extract and recommend programs having characteristics that have not been viewed by the user so far, so that the selection of the recommended programs to the user is surprising and the user's interest is expanded. In addition to being able to connect, it is possible to acquire very important history information in order to recommend programs more suited to the user's preference.

  Through the processing described with reference to FIGS. 13 to 30, recommendation information is generated in the program recommendation processing device 10 and supplied to the television display device 11 or the recording / playback device 12.

  The television display device 11 or the recording / playback device 12 is supplied with a satellite wave or terrestrial broadcast signal received and decoded by the television receiver 4.

  The television display device 11 displays the broadcast signal supplied from the television receiver 4 or the playback data supplied from the recording / playback device 12 or the program recommendation processing device 10 based on a user operation input. Based on the supplied recommendation information, recommended program information is displayed, and automatic channel setting is executed. Further, the television display device 11 supplies the operation log to the program recommendation processing device 10.

  Further, the recording / reproducing apparatus 12 records or reserves a broadcast signal supplied from the television receiving apparatus 4 based on a user operation input, or based on recommendation information supplied from the program recommendation processing apparatus 10. Automatically record programs. In addition, the recording / reproducing apparatus 12 reproduces a program recorded on the attached recording medium or a built-in recording medium, and outputs and displays the program on the television display apparatus 11. Further, the recording / reproducing apparatus 12 supplies the operation log to the program recommendation processing apparatus 10.

  FIG. 31 is a block diagram illustrating a configuration of the television receiver 4. The television receiver 4 will be described as being a general receiver corresponding to the standard of a digital broadcast receiver.

  The satellite wave detector 91 selects a satellite wave transmitted via the satellite 2 and received by the antenna 3 based on a signal for selecting a channel supplied from the television display device 11 or the recording / playback device 12. The control signal relating to the transmission mode is supplied to a TMCC (Transmission and Multiplexing Configuration Control) decoding unit 92 and the broadcast signal is supplied to a demodulation / decoding processing unit 93.

  The TMCC decoding unit 92 receives input of information such as a transmission mode (modulation scheme, coding rate, etc.) and a slot in the transmission multiplex control signal, decodes the information, and supplies the information to the demodulation / decoding processing unit 93.

  The demodulation / decoding processing unit 93 converts the supplied broadcast signal into, for example, a QPSK (quadri-phase shift keying: 4 phase modulation scheme, or 4) based on the information regarding the transmission mode supplied from the TMCC decoding unit 92. The signal is demodulated and decoded using a method such as a phase PSK method or an 8-phase PSK method, and supplied to the deinterleaver 94.

  The deinterleaver 94 deinterleaves the supplied signal and supplies it to the error correction processing unit 95. Further, the deinterleaver 94 may further perform frame separation and descrambling processing on the supplied signal.

  The error correction processing unit 95 performs error correction processing using, for example, a Reed-Solomon code and supplies the result to a CA (Conditional Access) descrambling unit 101.

  The terrestrial wave detection unit 96 selects and detects a terrestrial wave received by the antenna 3 based on a control signal for selecting a channel supplied from the television display device 11 or the recording / reproducing device 12, and transmits a transmission mode. The control signal is supplied to the TMCC decoding unit 97, and the broadcast signal is supplied to the demodulation / decoding processing unit 98.

  The TMCC decoding unit 97 receives input of information such as a transmission mode (modulation scheme, coding rate, etc.), slot, TS, etc. in the transmission multiplex control signal, decodes the information, and supplies it to the demodulation / decoding processing unit 98. .

  The demodulation / decoding processing unit 98 uses, for example, a method such as a QAM (quadrature amplitude modulation) method for the supplied broadcast signal based on the information regarding the transmission mode supplied from the TMCC decoding unit 97. The signal is demodulated and decoded and supplied to the deinterleaver 99.

  The deinterleaver 99 deinterleaves the supplied signal and supplies it to the TS (Transport Stream) reproducing unit 100. Further, the deinterleaver 99 may further perform frame separation and descrambling processing on the supplied signal.

  The TS reproducing unit 100 reproduces the transport stream based on the supplied signal and supplies the transport stream to the CA descrambling unit 101.

  The CA descrambling unit 101 descrambles the limited reception signal based on the signal supplied from the error correction processing unit 95 or the TS reproduction unit 100 and supplies it to the demultiplexer 102.

  The data input unit 103 receives EPG data input from the EPG receiving device 9 or receives streaming data supplied from the distribution server 5 via the network 8 and supplies the streaming data to the demultiplexer 102.

  The demultiplexer 102 demultiplexes the signal supplied from the CA descrambling unit 101 or the data input unit 103, the audio signal to the audio signal decoding unit 104, the video signal to the video signal decoding unit 105, and The control signal and data such as EPG are supplied to the data decoding unit 106, respectively.

  The audio signal decoding unit 104 decodes the supplied audio signal and supplies the decoded audio signal to the television display device 11 or the recording / playback device 12. The video signal decoding unit 105 decodes the supplied video signal and supplies the decoded video signal to the television display device 11 or the recording / playback device 12. The data decoding unit 106 decodes the supplied control signal, data such as EPG, and supplies the decoded data to the television display device 11 or the recording / playback device 12.

  The received satellite wave, terrestrial wave, or distributed streaming data is demodulated and decoded by a predetermined method by the processing of the television receiving device 4, and the television display device 11 or the recording / reproducing device 12 is decoded. To be supplied.

  FIG. 32 is a block diagram illustrating a configuration of the television display device 11.

  The operation input unit 121 receives an operation input from the user, supplies a control signal corresponding to the user's operation input to each unit of the television display device 11, and supplies the operation content of the user to the operation log list 122 for storage. Let The user operation log stored in the operation log list 122 is acquired by the operation log acquisition unit 46 of the program recommendation processing apparatus 10 described with reference to FIG. When the user operation input is a channel selection, the operation input unit 121 supplies the input user operation to the channel setting unit 123.

  The channel setting unit 123 generates a control signal indicating channel selection based on the signal indicating the user operation input supplied from the operation input unit 121 and supplies the control signal to the television receiver 4. In addition, the channel setting unit 123 generates a control signal indicating channel selection in order to automatically set a channel based on recommendation information registered in a recommended program list 128 described later, and the television receiver 4 is supplied. The television receiver 4 receives a broadcast signal of a designated channel based on the control signal.

  The data input unit 124 receives broadcast signals from the television receiver 4 and supplies them to the image processing unit 125. The image processing unit 125 performs image processing on the supplied broadcast signal based on the image display method of the output unit 126 and supplies the processed broadcast signal to the output unit 126. The output unit 126 includes, for example, a display device such as a CRT (Cathode Ray Tube) or LCD (Liquid Crystal Display) and an audio output device such as a speaker, and the supplied image signal of the broadcast signal after image processing is supplied. Is displayed on the display device, and an audio signal is output from the audio output device.

  The recommended program list acquisition unit 127 acquires recommended information from the program recommendation processing device 10 and supplies the recommended information to the recommended program list 128. The recommended program list 128 registers the supplied recommendation information. The recommended information registered in the recommended program list 128 is read by the channel setting unit 123 or the recommended program information display control unit 129.

  The recommended program information display control unit 129 outputs the recommended information read from the recommended program list 128 to the image processing unit 125 in order to present the recommended information of the program to the user. The image processing unit 125 outputs the recommended information supplied from the recommended program information display control unit 129 to the output unit 126 alone or superimposed on the image of the broadcast signal supplied from the data input unit 124. , Display on the display device.

  It should be noted that, based on the recommendation information supplied from the program recommendation processing device 10, whether the recommendation information is displayed on the output unit 126 or whether the channel is automatically set is determined by the user setting. May be.

  A recommendation information display process for presenting program recommendation information to the user based on the recommendation information supplied from the program recommendation processing apparatus 10 will be described with reference to the flowchart of FIG.

  In step S <b> 251, the recommended program list acquisition unit 127 acquires the recommendation information output from the recommendation information output unit 49 of the program recommendation processing device 10.

  In step S <b> 252, the recommended program list acquisition unit 127 registers the acquired recommendation information in the recommended program list 128.

  In step S253, the recommended program information display control unit 129 reads, from the recommended program list 128, recommended information of a program broadcast within a predetermined time from the current time, such as 3 hours or 1 day, and a recommended program. Recommendation information display data for displaying the title, content, broadcast time, broadcast channel, and the like of the image is generated and supplied to the image processing unit 125.

  In step S <b> 254, the image processing unit 125 executes image processing for causing the output unit 126 to display the supplied recommendation information display data and supplies the output unit 126 with the image processing. The recommendation information is subjected to image processing alone or superimposed on the image of the broadcast signal supplied from the data input unit 124.

  In step S255, the output unit 126 displays the recommendation information supplied from the image processing unit 125, and the process ends.

  As a result of such processing, the recommendation information is displayed on the output unit 126, so that the user can select a program to view with reference to the displayed recommendation information.

  Next, referring to the flowchart of FIG. 34, an automatic channel for automatically setting a channel based on the recommendation information supplied from the program recommendation processing device 10 and displaying a program that matches the user's preference. The setting process will be described.

  In steps S271 and S272, processing similar to that in steps S251 and S252 described with reference to FIG. 33 is executed. That is, the recommendation information output from the recommendation information output unit 49 of the program recommendation processing device 10 is acquired, and the acquired recommendation information is registered in the recommended program list 128.

  In step S273, the channel setting unit 123 acquires recommended program information corresponding to the current time from the program recommendation list 128. In step S274, the channel setting unit 123 generates channel setting information based on the recommended program information. Output to the receiver 4. The television receiver 4 receives a broadcast signal of a designated channel based on the control signal.

  In step S <b> 275, the data input unit 124 acquires a broadcast signal of a designated channel from the television receiver 4 and supplies the broadcast signal to the image processing unit 125.

  In step S <b> 276, the image processing unit 125 performs image processing for displaying the supplied broadcast signal on the output unit 126 and supplies the image signal to the output unit 126.

  In step S277, the output unit 126 displays the video of the recommended program supplied from the image processing unit 125, outputs the sound, and the process ends.

  By such processing, the channel is automatically set to the channel on which the program that matches the user's preference is broadcast. The automatic channel setting process described with reference to FIG. 34 may be executed, for example, when a user instructs. This makes it possible to automatically set a channel for a program on which an appropriate program is broadcast when the user is worried about which program to view.

  In addition, the automatic channel setting process described with reference to FIG. 34 is performed every predetermined time that can be determined to be left alone, for example, for 2 hours or the like without particularly setting the channel. It may be executed when there is no operation input from the user for a predetermined time during which it can be determined that the user is left unattended.

  In order to prevent the channel from being automatically switched while the user is consciously watching a program, a mode in which automatic channel setting processing cannot be executed is prepared and the user can instruct. You may be able to do it.

  FIG. 35 is a block diagram showing the configuration of the recording / playback apparatus 12.

  The operation input unit 141 receives an operation input from the user, supplies a control signal corresponding to the user's operation input to each unit of the recording / playback apparatus 12, and supplies the operation content of the user to the operation log list 142 for storage. . The user operation log stored in the operation log list 142 is acquired by the operation log acquisition unit 46 of the program recommendation processing apparatus 10 described with reference to FIG.

  The recording setting unit 143 starts broadcasting a program to be recorded, for example, based on a signal indicating a user's operation input supplied from the operation input unit 141 or from recommended information registered in a recommended program list 149 described later. Information necessary for recording processing, such as time and broadcast end time, and broadcast channel, is extracted. When the user operation input supplied from the operation input unit 141 is a recording reservation, the recording setting unit 143 registers information necessary for performing the recording process in the recording reservation list 144 and is supplied from the operation input unit 141. If the user's operation input is a recording process for a currently broadcast program, or if an automatic recording process is performed using recommended information registered in a recommended program list 149 described later, the recording process is performed. Information necessary for the recording is supplied to the recording control unit 145.

  The recording control unit 145 is based on the information necessary for performing the recording process supplied from the recording setting unit 143 or the recording corresponding to the current time among the recording reservation information registered in the recording reservation list 144. By extracting the reservation information, a control signal indicating the broadcast channel of the program to be recorded is generated and supplied to the television receiver 4, and a control signal for executing the recording process is generated, and a recording / playback processing unit 147. The television receiver 4 receives a broadcast signal of a designated channel based on the control signal.

  The data input unit 146 receives a broadcast signal from the television receiver 4 and supplies it to the recording / playback processing unit 147. The recording / playback processing unit 147 is configured so that a recording medium such as a magnetic tape, an optical disk, a magnetic disk, a magneto-optical disk, or a semiconductor memory can be mounted therein, for example, a hard disk, A recording medium such as a semiconductor memory is provided, and information can be recorded on the recording medium and information can be reproduced from the recording medium. Specifically, when the recording medium that can be attached to the recording / playback processing unit 147 is a magnetic tape, the recording / playback processing unit 147 has a magnetic head, and receives a broadcast signal supplied from the data input unit 146 to the magnetic tape. Information is recorded (that is, recorded) or information recorded on the magnetic tape is reproduced and supplied to the television display 11 or the like for reproduction output.

  The recommended program list acquisition unit 148 acquires recommendation information from the program recommendation processing device 10 and supplies it to the recommended program list 149. The recommended program list 149 registers the supplied recommendation information. The recommendation information registered in the recommended program list 149 is read by the recording setting unit 143, and the recording process is automatically executed.

  Next, automatic recording processing will be described with reference to the flowchart of FIG.

  In step S291, the recommended program list acquisition unit 148 acquires the recommendation information output from the recommendation information output unit 49 of the program recommendation processing device 10.

  In step S292, the recommended program list acquisition unit 148 registers the acquired recommended information in the recommended program list 149.

  In step S293, the recording setting unit 143 extracts information on a recommended program corresponding to the current time from the program recommendation list 149, and is necessary for recording processing such as broadcast start time and broadcast end time, and broadcast channel. Such information is acquired and supplied to the recording control unit 145.

  In step S294, the recording control unit 145 generates channel setting information for receiving the supply of the broadcast signal of the program to be recorded, and outputs the channel setting information to the television receiver 4. The television receiver 4 receives a broadcast signal of a designated channel based on the control signal.

  In step S295, the data input unit 146 acquires the broadcast signal of the designated channel from the television receiver 4 and supplies the broadcast signal to the recording / playback processing unit 147.

  In step S296, the recording / playback processing unit 147 records the supplied broadcast signal on a recording medium that is attached or built in, and the process ends.

  By such processing, it is possible to automatically record a program that matches the user's preference. The automatic recording process described with reference to FIG. 36 is not executed while a recording operation is already being performed, such as a recording process instructed by a user or a recording operation by a recording reservation process.

  Here, the case where the recording is automatically performed based on the recommended program corresponding to the current time has been described. However, for example, the recommended information for a predetermined time before the current time is acquired and the same processing is performed. Needless to say, the recording reservation may be automatically set.

  In the processing described with reference to FIGS. 1 to 36, the program vector PP is described as being generated in the distribution server 5, but the distribution server does not generate the program vector PP, but via the network 8. The EPG data may be supplied to the program recommendation processing device, and the program recommendation processing device may generate the program vector PP.

  The distribution server does not generate the program vector PP, but supplies the EPG data to the program recommendation processing device via the network 8, and the configuration of the distribution server 171 when the program recommendation processing device generates the program vector PP is as follows. FIG. 37 shows the configuration of the program recommendation processing device 191.

  2 corresponding to those in the distribution server 5 in FIG. 2 and the program recommendation processing apparatus 10 in FIG. 13 are denoted by the same reference numerals, and the description thereof is omitted as appropriate.

  That is, the distribution server 171 includes the data acquisition unit 21 and the data transmission unit 25 of the distribution server 5 described with reference to FIG. 2, and includes streaming data or metadata from the streaming data database 6 or the metadata database 7. The EPG data is acquired and transmitted to the EPG reception device 9 or the television reception device 4 via the network 8.

  The program recommendation processing device 191 uses FIG. 13 except that a metadata acquisition unit 22 and a program vector generation unit 23 similar to those provided in the distribution server 5 of FIG. 2 are newly provided. It has the same configuration as the program recommendation processing device 10 described, and in addition to the processing executed by the program recommendation processing device 10, the program vector generation processing 1 described with reference to FIG. 3 and FIG. The program vector generation process 2 described above, the grouping process 1 described using FIG. 7, and the grouping process 2 described using FIG. 8 are executed.

  Even when the distribution server does not generate the program vector PP but supplies EPG data to the program recommendation processing device via the network 8, and the program recommendation processing device generates the program vector PP, FIG. 1 to FIG. It is possible to achieve the same effect as described with reference to 36.

  Further, the EPG receiving device 9 collects user operation history and setting information from the television display device 11 and the recording / playback device 12 and supplies them to the distribution server via the network 8. The distribution server may not only generate the program vector PP but also execute a matching process and supply the matching result to the EPG receiver 9 via the network 8. A network configuration in this case is shown in FIG. 39, and a block diagram showing a configuration of the distribution server 201 is shown in FIG.

  In addition, the same code | symbol is attached | subjected to the part corresponding to the case in FIG.1, FIG.37, and FIG.38, The description is abbreviate | omitted suitably.

  That is, the distribution server 201 is obtained by adding the function of the program recommendation processing device 191 described with reference to FIG. 38 to the distribution server 171 described with reference to FIG. 37, and the user has the program recommendation processing device. There is no need.

  In the configuration shown in FIGS. 39 and 40, the program vector PP, the program side effect vector EfPP, the positive history vector UP, the negative history vector MUP, or the standard preference vector APP is generated, or the program vector grouping processing is performed. All processing described with reference to FIGS. 3 to 12 and FIGS. 14 to 30 such as matching processing or exceptional recommended program selection processing is executed by the distribution server 201.

  At this time, the EPG receiving device 9 collects from the television display device 11 and the recording / playback device 12 and transmits it via the network 8, for example, in the user operation history and setting information, for example, the user ID Etc. are added so that each user can be distinguished. Further, in the program recommendation processing device 191 of the distribution server 201, the initial registration storage unit 45, the positive history storage unit 47, the negative history storage unit 48, etc., each information is classified by user based on the user ID of the supplied information. save.

  In the above description, the case of recommending a program that suits the user's preference using EPG data of a television broadcast signal has been described. However, the present invention is applicable to various digital contents such as radio broadcast or streaming data. The present invention can also be applied to a case where attribute information is added and a user's preference is recommended.

  The series of processes described above can also be executed by software. The software is a computer in which the program constituting the software is incorporated in dedicated hardware, or various functions can be executed by installing various programs, for example, a general-purpose personal computer For example, it is installed from a recording medium.

  As shown in FIG. 2, FIG. 13, or FIG. 38, this recording medium is distributed to provide a program to the user separately from the computer, and is recorded on a magnetic disk 31 or 71 (flexible disk). Disk), optical disk 32 or 72 (including CD-ROM (Compact Disk-Read Only Memory), DVD (Digital Versatile Disk)), magneto-optical disk 33 or 73 (MD (Mini-Disk) (trademark) included) ) Or a package medium composed of the semiconductor memory 34 or 74.

  Further, in the present specification, the step of describing the program recorded on the recording medium is not limited to the processing performed in chronological order according to the described order, but may be performed in parallel or It also includes processes that are executed individually.

  In the present specification, the system represents the entire apparatus constituted by a plurality of apparatuses.

It is a figure explaining a television program broadcast and delivery of stream data. It is a block diagram which shows the structure of the delivery server of FIG. It is a flowchart explaining the program vector generation process 1. FIG. It is a figure explaining EPG data. It is a figure explaining a program vector. It is a flowchart explaining the program vector generation process 2. FIG. 10 is a flowchart for explaining grouping processing 1; 10 is a flowchart for explaining grouping processing 2; It is a flowchart explaining the title grouping process 1. FIG. It is a flowchart explaining the title grouping process 2. FIG. 10 is a flowchart for explaining title grouping processing 3; 10 is a flowchart for explaining title grouping processing 4; It is a block diagram which shows the structure of the program recommendation processing apparatus of FIG. 10 is a flowchart for explaining positive history vector and negative history vector generation processing 1; It is a figure explaining a positive history vector. 10 is a flowchart for explaining positive history vector and negative history vector generation processing 2; It is a flowchart explaining the matching process 1. FIG. It is a flowchart explaining the matching process 2. FIG. It is a flowchart explaining the matching process 3. FIG. It is a flowchart explaining the matching process 4. FIG. It is a flowchart explaining the matching process 5. FIG. It is a flowchart explaining the user side effect vector generation process 1. FIG. It is a flowchart explaining the user side effect vector generation process 2. FIG. It is a flowchart explaining the user side effect vector generation process 3. FIG. It is a flowchart explaining the user side effect vector generation process 4. FIG. It is a flowchart explaining the user side reaction vector generation process 1. FIG. It is a flowchart explaining the user side reaction vector generation process 2. FIG. It is a flowchart explaining the matching process including group recommendation. It is a flowchart explaining the matching process using a user model. It is a flowchart explaining an exception recommendation process. It is a block diagram which shows the structure of the television receiver of FIG. It is a block diagram which shows the structure of the television display apparatus of FIG. It is a flowchart explaining a recommendation information display process. It is a flowchart explaining an automatic channel setting process. It is a block diagram which shows the structure of the video recording / reproducing apparatus of FIG. It is a flowchart explaining an automatic video recording process. It is a block diagram which shows the example of a different structure of a delivery server. It is a block diagram which shows the example of a different structure of a program recommendation processing apparatus. It is a figure explaining the example from which the network for a television program broadcast and the delivery of stream data differs. It is a block diagram which shows the example of a different structure of a delivery server.

Explanation of symbols

  5 distribution server, 10 program recommendation processing device, 21 data acquisition unit, 22 metadata extraction unit, 23 program vector generation unit, 24 data storage unit, 25 data transmission unit, 41 data acquisition unit, 42 program vector extraction unit, 43 matching Processing unit, 44 operation input unit, 45 initial registration storage unit, 46 operation log acquisition unit, 47 positive history storage unit, 48 negative history storage unit, 49 recommended information output unit, 50 recommended program list, 61 normalization processing unit, 62 Vector calculation unit, 63 user information registration unit, 171 distribution server, 191 program recommendation processing device, 201 distribution server

Claims (28)

An information processing apparatus that executes processing for selecting content similar to a user's preference,
Receiving means for receiving attribute information of the content composed of a plurality of items from another device;
User history information generating means for generating user history information consisting of positive history information that is a history of content preferred by the user and negative history information that is a history of content not preferred by the user from an operation history of the user's content When,
The attribute information of the content, the positive history information, and the negative history information are vectorized for each item, and the inner product of the cosine distance of the content attribute information and the vector of the user history information and the weighting information is calculated for each item. Calculating means for calculating the similarity between the attribute information of the received content and the generated user history information by calculating and summing the result of the inner product for all the items ;
Selection means for selecting content based on the calculated similarity;
The weighting information used by the calculation means, and the cosine distance between a standard preference vector indicating general viewer preference obtained by counting the attribute information for a predetermined period and the vector of the positive history information is the item And the reciprocal number thereof is positive weighting information that defines the degree of positive contribution of each item when calculating the similarity between the attribute information and the positive history information, and the standard preference vector, The cosine distance with the negative history information vector is calculated for each item, and the reciprocal thereof defines the degree of negative contribution of each item when calculating the similarity between the attribute information and the negative history information. Weighting information generating means for generating negative weighting information
With
The calculation means calculates the inner product of the cosine distance of the content attribute information and the vector of the positive history information and the positive weighting information for each item, and adds the result of the inner product for all the items. From the sum, the inner product of the cosine distance of the content attribute information and the negative history information vector and the negative weighting information is calculated for each item, and the sum of the inner product results for all the items is calculated. The similarity between the content attribute information and the user history information is calculated based on the subtracted difference.
Information processing device . The user history information generating means generates the user history information for each genre,
The weighting information generation means calculates a cosine distance between the standard preference vector and the vector of the positive history information for each item for a predetermined genre, and generates the reciprocal thereof as the positive weighting information, For a predetermined genre, a cosine distance between the standard preference vector and the negative history information vector is calculated for each item, and the reciprocal thereof is generated as the negative weighting information.
The calculation means calculates the inner product of the attribute information of the content and the cosine distance of the vector of the positive history information of the predetermined genre for each item, and sums the results of the inner products for all the items. From the above, the inner product of the attribute information of the content and the cosine distance of the vector of the negative history information of the predetermined genre is calculated for each item, and the sum of the results of the inner product for all the items is subtracted. the difference, the information processing apparatus according to claim 1 for calculating the degree of similarity between the attribute information and the user history information of the content of the predetermined genres. The receiving means also receives attribute information of the content to which information indicating a predetermined group when content is classified into a plurality of groups is added from the other device,
The selection means selects the content corresponding to the information indicating the group when the information indicating the group is added to the attribute information of the content, and calculates the calculation when the information indicating the group is not added. The information processing apparatus according to claim 1, wherein content is selected based on the similarity calculated by the means. The information processing apparatus according to claim 3, wherein the content is a program, and the group is grouped by any one of a series drama, a content genre, a performer, and a title. The information processing apparatus according to claim 1, further comprising a storage unit that stores the weighting information corresponding to the item of the attribute information generated by the weighting information generation unit. The information processing apparatus according to claim 1, further comprising an accumulation unit that accumulates an operation history of the user's content. An information processing method for an information processing apparatus that executes processing for selecting content similar to a user's preference, wherein the information processing apparatus includes:
A receiving step of receiving attribute information of the content composed of a plurality of items from another device;
A user history information generation step for generating user history information including positive history information that is a history of content that the user likes and negative history information that is a history of content that the user does not like from the operation history of the user's content. When,
The attribute information of the content, the positive history information, and the negative history information are vectorized for each item, and the inner product of the cosine distance of the content attribute information and the vector of the user history information and the weighting information is calculated for each item. A calculation step of calculating the similarity between the attribute information of the received content and the generated user history information by calculating and summing the result of the inner product for all the items ;
A selection step of selecting content based on the calculated similarity;
The cosine distance between the standard preference vector indicating the general preference of the viewer , which is the weighting information used in the processing of the calculation step, and obtained by counting the attribute information for a predetermined period, and the vector of the positive history information Is calculated for each item, and the reciprocal thereof is positive weighting information that defines the degree of positive contribution of each item when calculating the similarity between the attribute information and the positive history information, and the standard preference The cosine distance between the vector and the vector of the negative history information is calculated for each item, and the reciprocal of the negative contribution of each item in calculating the similarity between the attribute information and the negative history information. A weighting information generating step for generating as negative weighting information for prescribing the degree ;
Including
The processing of the calculating step calculates an inner product of the cosine distance of the content attribute information and the vector of the positive history information and the positive weighting information for each item, and the result of the inner product is calculated for all the items. From the sum, the inner product of the cosine distance of the content attribute information and the vector of the negative history information and the negative weighting information is calculated for each item, and the result of the inner product is added for all the items. The similarity between the content attribute information and the user history information is calculated based on the difference obtained by subtracting the sum.
Information processing method . The processing of the user history information generation step generates the user history information for each genre,
The processing of the weighting information generation step calculates a cosine distance between the standard preference vector and the vector of the positive history information for each item for a predetermined genre, and generates the reciprocal thereof as the positive weighting information. For each of the predetermined genres, a cosine distance between the standard preference vector and the negative history information vector is calculated for each item, and the reciprocal thereof is generated as the negative weighting information.
In the calculation step, the inner product of the attribute information of the content and the cosine distance of the vector of the positive history information of the predetermined genre is calculated for each item, and the result of the inner product is added to all the items. The inner product of the attribute information of the content and the cosine distance of the vector of the negative history information of the predetermined genre is calculated for each item, and the sum of the results of the inner products for all the items is calculated. The information processing method according to claim 7, wherein the similarity between the attribute information of the content of the predetermined genre and the user history information is calculated based on the subtracted difference . The process of the receiving step also receives attribute information of the content to which information indicating a predetermined group when content is classified into a plurality of groups is added from the other device,
In the process of the selection step, when information indicating the group is added to the attribute information of the content, the content corresponding to the information indicating the group is selected, and when the information indicating the group is not added, The information processing method according to claim 7, wherein content is selected based on the similarity calculated by the calculation step. The information processing method according to claim 9, wherein the content is a program, and the group is grouped by any one of a series drama, a content genre, a performer, and a title. The information processing method according to claim 7, further comprising a storage step of storing the weighting information corresponding to the item of the attribute information generated by the processing of the weighting information generation step. The information processing method according to claim 7, further comprising an accumulation step of accumulating an operation history for the user's content. A program for causing a computer to execute processing for selecting content similar to a user's preference,
A receiving step of receiving attribute information of the content composed of a plurality of items from another device;
A user history information generation step for generating user history information including positive history information that is a history of content that the user likes and negative history information that is a history of content that the user does not like from the operation history of the user's content. When,
The attribute information of the content, the positive history information, and the negative history information are vectorized for each item, and the inner product of the cosine distance of the content attribute information and the vector of the user history information and the weighting information is calculated for each item. A calculation step of calculating the similarity between the attribute information of the received content and the generated user history information by calculating and summing the result of the inner product for all the items ;
A selection step of selecting content based on the calculated similarity;
The cosine distance between the standard preference vector indicating the general preference of the viewer , which is the weighting information used in the processing of the calculation step, and obtained by counting the attribute information for a predetermined period, and the vector of the positive history information Is calculated for each item, and the reciprocal thereof is positive weighting information that defines the degree of positive contribution of each item when calculating the similarity between the attribute information and the positive history information, and the standard preference The cosine distance between the vector and the vector of the negative history information is calculated for each item, and the reciprocal of the negative contribution of each item in calculating the similarity between the attribute information and the negative history information. A weighting information generating step for generating as negative weighting information for prescribing the degree ;
Including
The processing of the calculating step calculates an inner product of the cosine distance of the content attribute information and the vector of the positive history information and the positive weighting information for each item, and the result of the inner product is calculated for all the items. From the sum, the inner product of the cosine distance of the content attribute information and the vector of the negative history information and the negative weighting information is calculated for each item, and the result of the inner product is added for all the items. The similarity between the content attribute information and the user history information is calculated based on the difference obtained by subtracting the sum.
A program that causes a computer to execute processing . A computer-readable recording medium on which the program according to claim 13 is recorded. An information processing apparatus that executes processing for selecting content similar to a user's preference,
First storage means for storing attribute information of the content composed of a plurality of items;
User history information generating means for generating user history information consisting of positive history information that is a history of content preferred by the user and negative history information that is a history of content not preferred by the user from an operation history of the user's content When,
The attribute information of the content, the positive history information, and the negative history information are vectorized for each item, and the inner product of the cosine distance of the content attribute information and the vector of the user history information and the weighting information is calculated for each item. calculated and, by a sum obtained by adding all of the items the result of the inner product, and calculates the similarity between the attribute information of the content stored by the first storing means, the user history information generated calculated Means,
Selection means for selecting content based on the calculated similarity;
The weighting information used by the calculation means, and the cosine distance between a standard preference vector indicating general viewer preference obtained by counting the attribute information for a predetermined period and the vector of the positive history information is the item And the reciprocal number thereof is positive weighting information that defines the degree of positive contribution of each item when calculating the similarity between the attribute information and the positive history information, and the standard preference vector, The cosine distance with the negative history information vector is calculated for each item, and the reciprocal thereof defines the degree of negative contribution of each item when calculating the similarity between the attribute information and the negative history information. Weighting information generating means for generating negative weighting information
With
The calculation means calculates the inner product of the cosine distance of the content attribute information and the vector of the positive history information and the positive weighting information for each item, and adds the result of the inner product for all the items. From the sum, the inner product of the cosine distance of the content attribute information and the negative history information vector and the negative weighting information is calculated for each item, and the sum of the inner product results for all the items is calculated. The similarity between the content attribute information and the user history information is calculated based on the subtracted difference.
Information processing device . The user history information generating means generates the user history information for each genre,
The weighting information generation means calculates a cosine distance between the standard preference vector and the vector of the positive history information for each item for a predetermined genre, and generates the reciprocal thereof as the positive weighting information, For a predetermined genre, a cosine distance between the standard preference vector and the negative history information vector is calculated for each item, and the reciprocal thereof is generated as the negative weighting information.
The calculation means calculates the inner product of the attribute information of the content and the cosine distance of the vector of the positive history information of the predetermined genre for each item, and sums the results of the inner products for all the items. From the above, the inner product of the attribute information of the content and the cosine distance of the vector of the negative history information of the predetermined genre is calculated for each item, and the sum of the results of the inner product for all the items is subtracted. The information processing apparatus according to claim 15, wherein a similarity between the attribute information of the content of the predetermined genre and the user history information is calculated based on the difference . The first storage means also stores attribute information of the content to which information indicating a predetermined group when content is classified into a plurality of groups is added,
The selection means selects the content corresponding to the information indicating the group when the information indicating the group is added to the attribute information of the content, and calculates the calculation when the information indicating the group is not added. The information processing apparatus according to claim 15, wherein content is selected based on the similarity calculated by a means. The information processing apparatus according to claim 17, wherein the content is a program, and the group is grouped by any one of a series drama, a content genre, a performer, and a title. The information processing apparatus according to claim 15, further comprising a second storage unit that stores weighting information corresponding to the item of the attribute information generated by the weighting information generation unit. The information processing apparatus according to claim 15, further comprising storage means for storing an operation history of the user's content. An information processing method for an information processing apparatus that executes processing for selecting content similar to a user's preference, wherein the information processing apparatus includes:
A first storage step of storing attribute information of the content composed of a plurality of items;
A user history information generation step for generating user history information including positive history information that is a history of content that the user likes and negative history information that is a history of content that the user does not like from the operation history of the user's content. When,
The attribute information of the content, the positive history information, and the negative history information are vectorized for each item, and the inner product of the cosine distance of the content attribute information and the vector of the user history information and the weighting information is calculated for each item. calculated and, by a sum obtained by adding all of the items the result of the inner product is calculated and the attribute information of the content stored by the processing of said first storage step, a similarity between the user history information calculated Steps,
A selection step of selecting content based on the calculated similarity;
The cosine distance between the standard preference vector indicating the general preference of the viewer , which is the weighting information used in the processing of the calculation step, and obtained by counting the attribute information for a predetermined period, and the vector of the positive history information Is calculated for each item, and the reciprocal thereof is positive weighting information that defines the degree of positive contribution of each item when calculating the similarity between the attribute information and the positive history information, and the standard preference The cosine distance between the vector and the vector of the negative history information is calculated for each item, and the reciprocal of the negative contribution of each item in calculating the similarity between the attribute information and the negative history information. A weighting information generating step for generating as negative weighting information for prescribing the degree ;
Including
The processing of the calculating step calculates an inner product of the cosine distance of the content attribute information and the vector of the positive history information and the positive weighting information for each item, and the result of the inner product is calculated for all the items. From the sum, the inner product of the cosine distance of the content attribute information and the vector of the negative history information and the negative weighting information is calculated for each item, and the result of the inner product is added for all the items. The similarity between the content attribute information and the user history information is calculated based on the difference obtained by subtracting the sum.
Information processing method . The processing of the user history information generation step generates the user history information for each genre,
The processing of the weighting information generation step calculates a cosine distance between the standard preference vector and the vector of the positive history information for each item for a predetermined genre, and generates the reciprocal thereof as the positive weighting information. For each of the predetermined genres, a cosine distance between the standard preference vector and the negative history information vector is calculated for each item, and the reciprocal thereof is generated as the negative weighting information.
In the calculation step, the inner product of the attribute information of the content and the cosine distance of the vector of the positive history information of the predetermined genre is calculated for each item, and the result of the inner product is added to all the items. The inner product of the attribute information of the content and the cosine distance of the vector of the negative history information of the predetermined genre is calculated for each item, and the sum of the results of the inner products for all the items is calculated. The information processing method according to claim 21, wherein a similarity between the attribute information of the content of the predetermined genre and the user history information is calculated based on the difference drawn . The processing of the first storing step also stores attribute information of the content to which information indicating a predetermined group when content is classified into a plurality of groups is added,
In the process of the selection step, when information indicating the group is added to the attribute information of the content, the content corresponding to the information indicating the group is selected, and when the information indicating the group is not added, The information processing method according to claim 21, wherein content is selected based on the similarity calculated by the calculation step. The content is a program, the group is a drama series, the genre of the content, performer information processing method according to claim 23 which are grouped in one title. The information processing method according to claim 21, further comprising a second storage step of storing weighting information corresponding to the item of the attribute information generated by the weighting information generation step. The information processing method according to claim 21, further comprising an accumulation step of accumulating an operation history of the user's content. A program for causing a computer to execute processing for selecting content similar to a user's preference,
A first storage step of storing attribute information of the content composed of a plurality of items;
A user history information generation step for generating user history information including positive history information that is a history of content that the user likes and negative history information that is a history of content that the user does not like from the operation history of the user's content. When,
The attribute information of the content, the positive history information, and the negative history information are vectorized for each item, and the inner product of the cosine distance of the content attribute information and the vector of the user history information and the weighting information is calculated for each item. calculated and, by a sum obtained by adding all of the items the result of the inner product, and the first attribute information of the content stored by the processing of the storage step, the similarity between the generated the user history information A calculating step for calculating;
A selection step of selecting content based on the calculated similarity;
The cosine distance between the standard preference vector indicating the general preference of the viewer , which is the weighting information used in the processing of the calculation step, and obtained by counting the attribute information for a predetermined period, and the vector of the positive history information Is calculated for each item, and the reciprocal thereof is positive weighting information that defines the degree of positive contribution of each item when calculating the similarity between the attribute information and the positive history information, and the standard preference The cosine distance between the vector and the vector of the negative history information is calculated for each item, and the reciprocal of the negative contribution of each item in calculating the similarity between the attribute information and the negative history information. A weighting information generating step for generating as negative weighting information for prescribing the degree ;
Including
The processing of the calculating step calculates an inner product of the cosine distance of the content attribute information and the vector of the positive history information and the positive weighting information for each item, and the result of the inner product is calculated for all the items. From the sum, the inner product of the cosine distance of the content attribute information and the vector of the negative history information and the negative weighting information is calculated for each item, and the result of the inner product is added for all the items. The similarity between the content attribute information and the user history information is calculated based on the difference obtained by subtracting the sum.
A program that causes a computer to execute processing . A computer-readable recording medium on which the program according to claim 27 is recorded.

Images