JP2008048001A - Information processor and processing method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To increase variation of content recommendation. <P>SOLUTION: The information processor extracts a specific person whom a user can judge as acceptable from user taste information (S21), determines a content in which the specific person appears as a recommended candidate (S22), executes a processing for extracting a similar person of the specific person, i.e. a similar person extraction processing (S3), determines a content in which the similar person appears as a recommended candidate (S24), selects a recommended content among recommended candidates (S25), and outputs a list of recommended contents (recommendation information) (S26). The invention is applicable to an apparatus for recommending the content of television broadcast program, and the like. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an information processing apparatus and method, and a program, and more particularly to an information processing apparatus capable of recommending content with variations within a range that does not deviate from the user's preference as well as content recommendation based on a specific person preferred by the user The present invention relates to a method and a program.

Conventionally, a recommendation device that recommends content using metadata attached to the content has been used (for example, Patent Document 1). For example, in a conventional recommendation device that targets a television broadcast program (hereinafter referred to as a program as appropriate) as content, the user's preference is obtained by using metadata for each program including program performers and program titles. It is possible to recognize a program in which a specific person appears and present it to the user as a recommended program.
JP 2004-194108 A

  However, for the user, the stronger the preference for the specific person, the more often the program in which the specific person appears has already been viewed or already reserved for recording. Therefore, in such a case, the program recommended by the conventional recommendation device including Patent Document 1 is a program known to the user, and the recommendation of such a program does not make sense. This applies not only to programs but also to general contents distributed to users.

  In other words, there is a demand for the user to recommend content with variations within a range that does not deviate from the user's preference. However, the conventional recommendation device including Patent Document 1 does not sufficiently meet such a demand.

  The present invention has been made in view of such a situation, and enables not only content recommendation based on a specific person preferred by the user but also content recommendation with variations within a range that does not deviate from the user preference. Is.

  An information processing apparatus according to an aspect of the present invention is an information processing apparatus that recommends content to a user, and includes preference information holding means that holds preference information including personal information about a specific person that the user likes, and the specific person. Recognizing the specific person based on the preference information held in the preference information holding means, and person similarity information holding means for holding person similarity information indicating each similarity based on a predetermined similarity criterion among a plurality of persons including Then, based on the person similarity information held in the person similarity information holding means, a similar person that can be determined to be similar to the specific person is recognized, and the content in which the specific person appears and the content in which the similar person appears Recommending means for determining at least one of them as a candidate of recommended content to be recommended to the user.

  The similarity based on the predetermined similarity criterion is a similarity based on facial features of a person.

  The similarity based on the predetermined similarity criterion is a similarity based on the characteristics of a voice uttered by a person.

  The similarity based on the predetermined similarity criterion is a similarity based on characteristics of a language used by a person.

  The person similarity information includes information indicating each degree of similarity based on each of a plurality of similarity criteria, and the recommendation unit determines whether the similar person candidate and the specific person are in each of the plurality of similarity criteria. Are obtained from the person similarity information, and based on the obtained similarities for each of the plurality of similarity criteria and the importance levels of the plurality of similarity criteria that change in accordance with the content content The overall similarity of the similar person candidates is calculated, and it is determined whether to recognize the similar person candidate as the similar person based on the total similarity.

  The recommendation unit recognizes a plurality of persons as the specific person, acquires each similarity between the similar person candidate and each of the plurality of specific persons from the person similarity information, and acquires each similar Based on the degree, a total similarity of the similar person candidates is calculated, and based on the total similarity, it is determined whether to recognize the similar person candidate as the similar person.

  Similarity calculating means for calculating each similarity based on the predetermined similarity criterion between the plurality of persons including the specific person and holding the calculation result as the person similarity information in the person similarity information holding means; Further prepare.

  An information processing method according to one aspect of the present invention shows preference information holding means for holding preference information related to a specific person preferred by a user, and each similarity based on a predetermined similarity criterion between a plurality of persons including the specific person An information processing method of an information processing apparatus that recommends content to the user, the identification information being based on the preference information held in the preference information holding means Recognizing a person, recognizing a similar person that can be judged to be similar to the specific person based on the person similarity information held in the person similarity information holding means, and the content in which the specific person appears and the similar person appear Determining at least one of the content to be recommended as a recommended content candidate to be recommended to the user.

  A program according to one aspect of the present invention is a program corresponding to the information processing method according to one aspect of the present invention described above.

  In the information processing apparatus, method, and program according to one aspect of the present invention, the preference information holding unit that holds preference information related to the specific person that the user likes, and a predetermined similarity criterion between the plurality of persons including the specific person are used. And a person similarity information holding unit that holds person similarity information indicating each similarity based on the information processing apparatus or the recommendation apparatus that recommends content to the user, and the following processing is executed. That is, it can be determined that the specific person is recognized based on the preference information held in the preference information holding unit and is similar to the specific person based on the person similarity information held in the person similarity information holding unit. A similar person is recognized, and at least one of content in which the specific person appears and content in which the similar person appears is determined as a recommended content candidate to be recommended to the user.

  As described above, according to the present invention, it is possible to recommend content to the user. In particular, not only content recommendation based on a specific person preferred by a user but also content recommendation based on a similar person can be performed. That is, it is possible to recommend content with variations within a range that does not deviate from the user's preference.

  Embodiments of the present invention will be described below. Correspondences between the configuration requirements of the present invention and the embodiments described in the detailed description of the present invention are exemplified as follows. This description is to confirm that the embodiments supporting the present invention are described in the detailed description of the invention. Accordingly, although there are embodiments that are described in the detailed description of the invention but are not described here as embodiments corresponding to the constituent elements of the present invention, It does not mean that the embodiment does not correspond to the configuration requirements. Conversely, even if an embodiment is described here as corresponding to a configuration requirement, that means that the embodiment does not correspond to a configuration requirement other than the configuration requirement. It's not something to do.

  Further, this description does not mean that all the inventions corresponding to the specific examples described in the embodiments of the invention are described in the claims. In other words, this description is an invention corresponding to the specific example described in the embodiment of the invention, and the existence of an invention not described in the claims of this application, that is, in the future, a divisional application will be made. Nor does it deny the existence of an invention added by amendment.

An information processing apparatus according to one aspect of the present invention (for example, the content recommendation processing apparatus 10 in FIG. 2)
An information processing device that recommends content to a user,
Preference information holding means (for example, the user preference information holding unit 29 in FIG. 2) that holds preference information (for example, user preference information including the performer information Pup in FIG. 4) related to the specific person that the user likes;
Person similarity information holding means (for example, person similarity in FIG. 2) that holds person similarity information (for example, person similarity information in FIG. 6) indicating each similarity based on a predetermined similarity criterion among a plurality of persons including the specific person. An information holding unit 31);
A similar person that recognizes the specific person based on the preference information held in the preference information holding means and can determine that the specific person is similar to the specific person based on the person similarity information held in the person similarity information holding means And recommending means for determining at least one of the content in which the specific person appears and the content in which the similar person appears as a recommended content candidate to be recommended to the user (for example, the recommended content in FIG. 2) And an extraction unit 32).

The person similarity information includes information indicating each similarity based on each of a plurality of similarity criteria (for example, the person similarity information in the example of FIG. 6 includes a similarity table of “face” as an example of a similarity criterion, A similarity table of “voice” as an example of similarity criteria, and a similarity table of “language” as an example of similarity criteria),
The recommendation means is:
For each of the plurality of similarity criteria, each similarity between the candidate for the similar person and the specific person is acquired from the person similarity information (for example, the process of step S44 of FIG. 8 is executed)
Each of the similarities for each of the plurality of similar criteria acquired, and the importance of the plurality of similar criteria changing according to the content content (for example, each changing according to the genre (content) of the content in FIG. 11) Based on the weighting coefficients α, β, γ), the total similarity of the similar person candidates is calculated (for example, the total similarity of a plurality of use similarity criteria in step S50 of FIG. 8 (details are FIG. 10)). Execute arithmetic processing)
Whether or not the similar person candidate is recognized as the similar person is determined based on the total similarity (for example, the process of step S51 in FIG. 8 is executed).

The recommendation means is:
Recognizing a plurality of persons as the specific person,
Each similarity between the similar person candidate and each of the plurality of specific persons is acquired from the person similarity information (for example, the loop process of steps S43 to S45 in FIG. 8 is repeatedly executed)
Based on the obtained similarities, the overall similarity of the similar person candidates is calculated (for example, the total similarity calculation processing of a plurality of specific persons in step S47 in FIG. 8 (details are shown in FIG. 9)). And)
Whether or not the similar person candidate is recognized as the similar person is determined based on the total similarity (for example, the process of step S51 in FIG. 8 is executed).

  A similarity calculation unit that calculates each similarity based on the predetermined similarity criterion between the plurality of persons including the specific person and holds the calculation result as the person similarity information in the person similarity information holding unit ( For example, it further includes the person similarity information generation unit 30) of FIG.

An information processing method according to one aspect of the present invention (for example, a method corresponding to the content recommendation process in FIG. 7)
Preference information holding means (for example, the user preference information holding unit 29 in FIG. 2) that holds preference information (for example, user preference information including the performer information Pup in FIG. 4) relating to a specific person that the user likes;
Person similarity information holding means (for example, person similarity in FIG. 2) that holds person similarity information (for example, person similarity information in FIG. 6) indicating each similarity based on a predetermined similarity criterion among a plurality of persons including the specific person. An information holding unit 31),
An information processing method of an information processing device (for example, the content recommendation processing device 10 in FIG. 2) that recommends content to the user,
The specific person is recognized based on the preference information held in the preference information holding means (for example, step S21 in FIG. 7), and the specific person is determined based on the person similarity information held in the person similarity information holding means. A similar person that can be determined to be similar to a person is recognized (for example, step S23 in FIG. 7, ie, similar person extraction processing in FIG. 8), and at least of the content in which the specific person appears and the content in which the similar person appears One of them is determined as a candidate of recommended content to be recommended to the user (for example, steps S22 and S24 in FIG. 7).

  The program according to one aspect of the present invention is a program corresponding to the information processing method according to one aspect of the present invention described above, and is executed by, for example, the computer of FIG.

  Embodiments of the present invention will be described below with reference to the drawings.

  First, a television program broadcast and streaming data distribution will be described with reference to FIG.

  That is, in this embodiment, a television broadcast program (also abbreviated as a program here as appropriate) and distribution streaming data are adopted as examples of content.

  The broadcast station 1 transmits a terrestrial program broadcast, or transmits a satellite wave program broadcast via the satellite 2. In the example of FIG. 1, only one broadcasting station 1 is illustrated, but it is needless to say that there may be 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 EPG (Electronic Program Guide) information as necessary.

  The distribution server 5 reads out streaming data constituting content such as video and audio from the streaming data database 6 and distributes it to the television receiver 4 via the network 8 including the Internet and other networks. In addition, the distribution server 5 reads information including metadata regarding the streaming data from the metadata database 7 and distributes the information to the EPG reception device 9 via the network 8. Furthermore, the distribution server 5 reads EPG information related to the program broadcast from the broadcasting station 1 from the metadata database 7 and distributes it to the EPG receiver 9 via the network 8.

  When the amount of information of EPG information superimposed on a general broadcast signal is sufficient for the processing described below, the same EPG information superimposed on a general broadcast signal is used for the processing. You may make it do. In addition, if the amount of information of EPG information superimposed on a general broadcast signal is not sufficient information for the processing described below, the processing includes EPG information superimposed on a general broadcast signal. In addition to this, the metadata may be used alone. Hereinafter, since EPG information superimposed on a general broadcast signal does not have a sufficient amount of information for processing, information including metadata is used in the present embodiment. A combination of this information and information including metadata related to streaming data is collectively referred to as EPG data.

  The EPG receiver 9 provides the distributed EPG data to the television receiver 4.

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

  The content recommendation processing device 10 includes streaming data distributed or scheduled to be distributed from the distribution server 5 and television broadcast programs broadcast or scheduled to be broadcast from the broadcasting station 1 (hereinafter collectively referred to as contents). The recommendation information for recommending the content that matches the user's preference is generated and provided to the television display device 11 and the recording / playback device 12. Note that details of the content recommendation processing device 10 will be described later with reference to FIG.

  The television display device 11 displays and outputs a video / audio signal provided from the television receiver 4 or a video / audio signal reproduced from the recording / reproducing device 12 based on a user operation input. Or automatically setting channels based on the recommendation information provided from the content recommendation processing device 10 or displaying recommended program information. The television display device 11 provides an operation log, which is a user operation history, to the content recommendation processing device 10.

  The recording / reproducing apparatus 12 records the video / audio signal provided from the television receiving apparatus 4 on an attached recording medium or a built-in recording medium (for example, a hard disk) based on a user operation input. Based on the recommendation information provided from the content recommendation processing device 10, the video / audio signal provided from the television receiver 4 is automatically recorded on the attached recording medium or a built-in recording medium. The recording / playback apparatus 12 plays back the video / audio signals recorded on the attached recording medium or the built-in recording medium, and provides them to the television display apparatus 11. Furthermore, the recording / playback apparatus 12 provides an operation log, which is a user operation history, to the content recommendation processing apparatus 10.

  Here, the EPG receiving device 9, the television receiving device 4, the content 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 not necessarily individually provided. There is no need to configure. 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. Nor. Furthermore, 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. In addition, the content 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 shows a functional configuration example of the content recommendation processing apparatus 10.

  In the example of FIG. 2, the content recommendation processing device 10 is configured to include a data acquisition unit 21 to a recommendation information holding unit 34.

  The data acquisition unit 21 acquires data distributed from the distribution server 5 via the network 8 and data of a television broadcast program broadcast from the broadcast station 1 and received by the television receiver 4 (see FIG. 1). ). When the acquired data is streaming data or television broadcast program data (hereinafter collectively referred to as content data), the data acquisition unit 21 causes the content information holding unit 24 to hold the content data. In addition, when the acquired data is EPG data, the data acquisition unit 21 provides the EPG data to the metadata extraction unit 22.

  The metadata extraction unit 22 extracts various metadata necessary for generating content information, which will be described later using FIG. 3, from the EPG data provided from the data acquisition unit 21, and provides the metadata to the content information generation unit 23. .

  The content information generation unit 23 generates various information groups (hereinafter referred to as content information) related to the predetermined content based on the metadata related to the predetermined content provided from the metadata extraction unit 22, and the content information holding unit 24.

  That is, in the content information holding unit 24, content data that is actual data and content information that is an aggregate of various pieces of information are associated and held for each of a plurality of contents.

  An example of such content information is shown in FIG. In the example of FIG. 3, the content information is expressed by a vector PP. That is, the content information generation unit 23 of FIG. 2 has a vector having information of N items (N is an integer value of 1 or more) items included in various metadata extracted from the metadata extraction unit 22 as components. A PP is generated as content information and is held in the content information holding unit 24.

  The N items here are not limited to items defined by EPG data, but items that can be arbitrarily set by a designer or the like. For example, in the example of FIG. 3, N = 7 major items T, G, H, S, P, A, and K are employed. Each piece of information regarding the contents of each of these seven major items is Tm, Gm, Hm, Sm, Pm, Am, Km. These seven item information Tm, Gm, Hm, Sm, Pm, Am, and Km are substituted for each component of the vector PP as content information. That is, the vector PP = (Tm, Gm, Hm, Sm, Pm, Am, Km) is generated as the content information.

  However, when content information is used for matching processing with the user UP information vector UP in FIG. 4 to be described later, PP ′ = (Tm, Gm, Pm, Am, Km) is used instead of the vector PP itself. . This is in order to match the dimension with the vector UP of the user preference information of the comparison partner.

  Here, the large item T means “title”, and title information Tm = {Title} is substituted into the corresponding component of the vector PP.

  Major item G means “genre”, and genre information Gm = {drama, variety, sports, movie, music, children / education, education / documents, news / reports, etc.} is a vector PP Assigned to the corresponding component.

  The large item H means “Hour”, and the time zone information Hm = {morning, noon, evening, golden, midnight} is substituted into the corresponding component of the vector PP.

  Major item S means “TV Station”, and broadcast station information Sm = {NNK General, NNK Education, Asia TV, TTS, Buji, Tele-Nippon, Tokyo, NNK Satellite No. 1, NNK Satellite No. 2 , WOWO} is assigned to the corresponding component of the vector PP. Although not shown in the example of FIG. 3, the distribution server 5 may be regarded as one broadcasting station, and the name of the distribution server 5 may be included in the broadcasting station information Sm.

  The large item P means “Person”, and the performer information Pm = {person A, person B,...} Is substituted into the corresponding component of the vector PP.

  The large item A means “screenplay / original / production etc. (Author)”, and screenplay / original / production etc. information Am = {person a, person b,...} Is the corresponding component of the vector PP. Assigned.

  The large item K means “content (keyword)”, and the content information Km = {kw1, kw2,...} Is substituted into the corresponding component of the vector PP.

  Among the seven major items described above, the major items “G: Genre”, “S: Broadcasting Station”, and “H: Time Zone” can be fixed, so that each item information Gm, Sm, Hm is Can be represented by a numeric vector.

  Specifically, for example, in the genre information Gm = {drama, variety, sports, movie, music, children / education, education / documents, news / reports, etc.}, the genre of the content for which the vector PP is generated Can be expressed by genre information Gm = {0, 0, 0, 0, 0, 0, 1, 0, 0}.

  Also, for example, the content that is the target of the generation of the vector PP at the broadcasting station Sm = {NNK General, NNK Education, Asia TV, TTS, Buji, Tele-Nichi, Toto, NNK Satellite 1, NNK Satellite 2, WOWO} When the broadcast station that broadcasts (here, a television broadcast program) is WOWO, the broadcast station information Sm = {0, 0, 0, 0, 0, 0, 0, 0, 0, 1} may be used. it can.

  Although the specific example is not described, the time zone information Hm can also be expressed by a numerical vector, like the genre information Gm and the broadcast station information Sm described above.

  On the other hand, each item information Tm, Pm, Am, Km such as “T: Title”, “P: Performer”, “A: Screenplay / Original / Director”, “K: Contents” is simply a numeric vector. Cannot be represented by Therefore, the item value (word) and the weight (numerical value) are generated in association with each other depending on the frequency of the word included in the content metadata or the like. For example, the title information Tm = {Tokaido-1, Mitani-1, Kaidan-1} can be represented by a vector having each item value-numerical item information. In this example, for example, “Tokaido-1” means that the frequency of the word “Tokaido” is “1”. Although a specific example is not described, the performer information Pm, the script / original / production information Am, and the content information Km can also be represented by vectors having item information composed of word-numerical values as elements.

  Returning to FIG. 2, the operation input unit 25 is configured by input devices such as a keyboard, a touch pad, and a mouse, for example, and receives initial registration information input by the user and topic input for generating a user model. The information is held in the initial registration information holding unit 26. That is, the initial registration information holding unit 26 holds initial registration contents provided from the operation input unit 25 and topics for user model generation, and a user preference information generation unit 28 and a recommended content extraction unit as necessary. 32. Further, the contents held in the initial registration information holding unit 26 are sequentially updated based on a user operation input from the operation input unit 25. The initial registration includes, for example, information indicating content that the user does not like, such as disliked genres, disliked keywords, disliked performers, favorite genres, favorite keywords, favorite performers, Information indicating content that the user likes can be employed.

  The operation log acquisition unit 27 acquires an operation log from the television display device 11 or the recording / playback device 12 (FIG. 1) and provides the operation log to the user preference information generation unit 28.

  The user preference information generation unit 28, based on the operation log, various information held in the initial registration information holding unit 26, and the like, the content recommendation processing device 10 itself, the television display device 11 in FIG. 12 (hereinafter referred to as user preference information) is generated and stored in the user preference information holding unit 29. Further, when the user preference information is already held in the user preference information holding unit 29, the user preference information generating unit 28 operates the operation log newly provided from the operation log acquisition unit 27 or the initial registration information holding unit 29. The content of the user preference information is updated as appropriate based on the updated content and the like.

  As the user preference information, for example, both positive history information and negative history information can be adopted. However, in the present embodiment, only the positive history information is used as the user preference information for the sake of simplicity.

  Here, the negative history information refers to information used to exclude content that is reluctant to use (viewing, recording, etc.) by the user, in other words, content that is considered undesirable, from the recommended content. However, as described above, the negative history information is not used in the present embodiment, and thus detailed description thereof is omitted.

  On the other hand, the positive history information is used to extract content that a user will actively use (viewing, recording, etc.), in other words, content that would be preferable, as recommended content candidates. Information used. For example, when the user uses the content (viewing or recording), or the recommended content presented as recommended information to the user by the process described later is accepted and used (viewing or recording) by the user In this case, the metadata of the recommended content is used as part of the positive history information as good impression metadata. Here, a part of the description is that the positive history information can be expressed as a vector like the content information vector PP described above, and the component information is generated as the information itself substituted into one component of the vector. This is because it is used as information to be performed.

  As described above, in the present embodiment, the user preference information is generated in the form of a vector by the user preference information generation unit 28 and held in the user preference information holding unit 29. Specifically, for example, a vector UP as shown in FIG. 4 is generated as user preference information by the user preference information generation unit 28 and held in the user preference information holding unit 29.

  That is, in the example of FIG. 4, “title”, “genre”, “performer”, “screenplay / original / production”, “content (keyword)” are adopted as the major items. A vector UP composed of each component of the user's positive history information Tup, Gup, Pup, Up, and Kup, that is, UP = (Tm, Gm, Pm, Am, Km) is generated as the user preference information. ing.

  In this case, each of the positive history information Tup, Gup, Pup, Up, Kup is generated as information in which item values (words) and weights (numerical values) are associated with each other. As this weight (numerical value), for example, a value (including an operation frequency value) indicating the user's preference level (the degree of preference) is adopted. For example, in the example of FIG. 4, the title information composed of each item value-numeric value can be represented by a vector such as title correct history information Tup = {Title1-12, Title2-3}. . In this example, for example, “Title1-12” means that the user's preference degree for “Title1” is “12”. On the other hand, “Title2-3” means that the user's preference degree for “Title2” is “3”. In other words, if the preference value increases as the user's preference increases, the user can determine that the user prefers “Title1” over “Title2”.

  Returning to FIG. 2, the human similarity information generation unit 30 generates information on the similarity between each person (hereinafter, person similarity information) for each person who appears in each of the one or more contents. The person similarity information holding unit 31 holds it. Details of the person similarity information will be described later with reference to FIGS.

  The recommended content extraction unit 32 recommends content that can be determined to match the user's preference based on the user preference information held in the user preference information holding unit 29 (hereinafter referred to as recommended content). Is extracted from the content information holding unit 24. The extraction here means not extraction of the content data itself of the recommended content, but extraction of at least part of information that can identify the recommended content, that is, content information related to the recommended content.

  The extraction method in this case is not particularly limited. For example, each vector PP ′ = (Tm, Gm, Pm, Am, Km) for each of a plurality of contents held in the content information holding unit 24 and ( 3), the user preference information vector UP held in the user preference information holding unit 29 = (Tm, Gm, Pm, Am, Km) (see FIG. 4) is matched, and the vector UP A method of extracting content having a matched vector PP as recommended content may be employed.

  However, in the present embodiment, for simplicity of explanation, attention is given to the characters of the content, and in the user preference information held in the user preference information holding unit 29, a person having a high user preference level is extracted as a specific person, Assume that a method of extracting content in which the specific person appears as recommended content is employed.

  In other words, in the present embodiment, the recommended content extraction unit 32 extracts a person (person) having a weight value (preference degree) equal to or greater than a predetermined threshold from the performer information Pup in FIG. 4 as a specific person. For example, when the threshold is 30, personB having the weight value “43” is extracted as the specific person. The recommended content extraction unit 32 includes a specific person in the performer information Pm in FIG. 3 of the content information related to each of the plurality of contents held in the content information holding unit 24 (vector correspondence) The content information including the specific person or a part thereof is extracted as the content information of the recommended content or a part thereof. That is, recommended content is extracted.

  Further, the recommended content extraction unit 32 recognizes a person who can be determined to be similar to the specific person described above (hereinafter referred to as a similar person) based on the human similarity information held in the human similarity information holding unit 31. . Then, the recommended content extraction unit 32 extracts content having content information in which a similar person is included in the performer information Pm in FIG. 3 (the corresponding component of the vector is 1) as recommended content. That is, the content information itself that includes a similar person in the performer information Pm (the corresponding component of the vector is 1) or a part thereof is also extracted as the content information of the recommended content or a part thereof. The A detailed example of this series of processing will be described later with reference to FIGS. 8 to 11 as similar person extraction processing.

  Note that the process in which the recommended content extraction unit 32 extracts recommended content using a similar person is not limited to the above-described processing example, and any processing can be employed. Specifically, for example, a vector Uman related to a person who can be a character of content rather than a user, for example, a vector Uman composed of various information based on the number and frequency of content in which the character is a character. Created in advance for each person, executes a matching process between the vector Uman related to the similar person and the vector PP ′, and extracts the content having the vector PP ′ matched with the vector Uman as the recommended content. May be.

  The recommended content extraction unit 32 generates a list of recommended content extracted in this way as recommendation information, and outputs it to the recommendation information output unit 33.

  Note that the recommended content extraction unit 32 may not use the extracted content as a recommended content immediately, but instead as a recommended candidate, and then determine the recommended candidate selected by an appropriate method as the recommended content. In other words, if the process of whether or not to include the extracted content in the recommended content list is regarded as a selection process, and it is also permitted to include all of the extracted content in the recommended content list, All the extracted contents can be regarded as recommended candidates.

  The recommendation information output unit 33 appropriately outputs the recommendation information provided from the recommended content extraction unit 32 to the television display device 11 and the recording / playback device 12 (FIG. 1). Also, the recommendation information output unit 33 causes the recommendation information holding unit 34 to appropriately hold the recommendation information.

  Next, of the processes of the content recommendation processing apparatus 10 having the functional configuration of FIG. 2, a process of generating or updating person similarity information (hereinafter referred to as person similarity information generation / update process), A series of processing (hereinafter referred to as content recommendation processing) until recommendation information is generated and output using person similarity information or the like will be described individually in that order.

  The flowchart of FIG. 5 shows an example of person similarity information generation / update processing.

  In step S <b> 1, the person similarity information generation unit 30 in FIG. 2 extracts a plurality of performers based on the content information held in the content information holding unit 24.

  Specifically, for example, the person similarity information generation unit 30 extracts all the persons (persons) included in the performer information Pm (see FIG. 4) of each content information regarding each content.

  In step S2, the person similarity information generation unit 30 sets a processing target similarity criterion.

  Here, the processing target similarity criterion will be described.

  The degree of similarity between persons can be calculated by paying attention to a predetermined characteristic among various characteristics of persons and calculating how similar the predetermined characteristic is according to a predetermined method. The characteristics of a person that can be a target of such calculation are referred to as similarity criteria here.

  In other words, if the feature causes a difference between each person and can express the difference qualitatively, for example, if the feature amount can be extracted for each person and the difference can be expressed by the feature amount. Any feature can be adopted as a similarity criterion. Specifically, for example, in the present embodiment, a facial feature of a person (hereinafter simply referred to as “face”), a voice uttered by the person (hereinafter simply referred to as “voice”), and a person It is assumed that the characteristics of the language used when speaking and writing (hereinafter simply referred to as “language”) are adopted as similarity criteria.

  Therefore, in the present embodiment, a predetermined one similarity criterion among “face”, “voice”, and “language” is set as a processing target similarity criterion by the processing in step S2.

  In step S3, the person similarity information generating unit 30 registers a plurality of performers extracted in the process of step S1 in the similarity table based on the processing target similarity criterion.

  Here, the similarity table refers to, for example, a table as shown in FIG. 6, that is, a table describing the similarity between performers for a predetermined similarity criterion. That is, in the present embodiment, as shown in FIG. 6, a similarity table of “face”, “voice”, and “language” is held in the person similarity information holding unit 31 as person similarity information. It will be.

  Therefore, when the similarity table for the processing target similarity criterion is not yet held in the person similarity information holding unit 31, in the process of step S3, all the plurality of performers extracted in the process of step S1 are registered. A similarity table (however, a numerical value indicating similarity is an undescribed similarity table) is newly generated and held in the person similarity information holding unit 31.

  Specifically, for example, the processing target similarity criterion is “face”, and “actor A”, “actor B”, “singer C”, “singer D”, “moderator E”, and When “moderator F” is extracted, a new similarity table for “face” in FIG. 6 (however, each numerical value indicating similarity is not described) is newly generated by the process of step S3. Then, it is held in the person similarity information holding unit 31.

  On the other hand, when the similarity table for the processing target similarity criterion is already held in the person similarity information holding unit 31, in the process of step S3, among the plurality of performers extracted in the process of step S1 Only performers who are not registered in the similarity table are newly registered. That is, the contents of the similarity table are updated.

  Specifically, for example, when the processing target similarity criterion is “face” and the similarity table of “face” in the state of FIG. 6 is already held in the person similarity information holding unit 31, the person G is When extracted in the process of step S1, although not shown, a column of “person G” is newly created to the right of the column of “moderator F” by the process of step S3, and “moderator F” A row of “person G” is newly created below the row of.

  In step S <b> 4, the person similarity information generation unit 30 acquires performer information based on a similarity calculation method according to the processing target similarity criterion. And in step S5, the person similarity information generation part 30 calculates the similarity between each performer according to the similarity calculation method according to the process target similarity standard, respectively.

  Here, an outline of a specific example of the similarity calculation method according to the processing target similarity criterion will be described.

  For example, when “face” is set as the processing target similarity criterion, the following similarity calculation method can be employed.

  That is, the person similarity information generation unit 30 extracts the face data of each performer. This extraction method is not particularly limited. For example, if face data of a predetermined performer is held in advance in the outside or the content information holding unit 24, the face data can be extracted as it is. Further, for example, when such face data does not exist, still image data including the face of the performer to be extracted is extracted from the video data of the content data held in the content information holding unit 24. A method of generating and extracting face area data as face data from the still image data can be employed.

  Next, the person similarity information generation unit 30 extracts feature amounts corresponding to the “face” similarity calculation method from each face data of a plurality of performers.

  For example, when a method using the Euclid distance in the space where the principal component analysis is performed as the similarity calculation method of “face”, luminance information is extracted from the face data (image data), Data obtained as a result of dimensional compression of the luminance data by principal component analysis is extracted as a feature amount according to the similarity calculation method of “face”. The feature amount corresponding to the “face” similarity calculation method is an example of the information of the performer in the process of step S4.

  Therefore, in the process of step S5, the similarity between each person is calculated based on an appropriate distance measure in the feature quantity space between the feature quantities of each extracted person. That is, brightness information is extracted from each face data (image data) of a plurality of performers, and each brightness information is obtained as a result of dimension compression by principal component analysis. Become. Each Euclid distance in the space where the principal component analysis is performed between the feature amounts of each person is calculated. Each of the Euclid distances is normalized so that it can be compared with other similar criteria (“voice” or “language”) (for example, a value in the range of 0 to 1 here). It is calculated as the similarity between them.

  For example, when “voice” is set as the processing target similarity criterion, the following similarity calculation method can be employed.

  That is, the person similarity information generation unit 30 extracts the audio waveform data of each performer. This extraction method is not particularly limited. For example, if the voice waveform data of a predetermined performer is held in advance in the outside or the content information holding unit 24, the voice waveform data can be extracted as it is. . Further, for example, when such audio waveform data does not exist, the audio waveform data from the section of the content data held in the content information holding unit 24 from the section where the performer to be extracted utters Can be used.

  Next, the person similarity information generation unit 30 extracts feature amounts corresponding to the “voice” similarity calculation method from the respective audio waveform data of the plurality of performers.

  For example, if a method using an acoustic model such as HMM (Hidden Markov Model) or GMM (Gaussian Mixtures Model) is adopted as the similarity calculation method for “voice”, such as MFCC (Mel Frequency Cepstrum Coefficient) The feature amount is extracted as a feature amount according to the similarity calculation method of “voice”. The feature amount corresponding to the “voice” similarity calculation method is an example of performer information referred to in the process of step S4.

  Therefore, in the process of step S5, the similarity between each person is calculated based on an appropriate distance measure in the feature quantity space between the feature quantities of each extracted person. That is, for example, the feature quantity distribution extracted from the voice waveform data of each of a plurality of performers is modeled as a mixed normal distribution, and the voice of a certain speaker (performer) is modeled by another (other performer). The normalized score (when applied to) is used as a distance measure, and each distance measure is normalized so that it can be compared to other similar criteria ("face" or "language") ( For example, here, a value in the range of 0 to 1) is calculated as the similarity between the performers.

  For example, when “language” is set as the processing target similarity criterion, the following similarity calculation method can be employed.

  That is, the person similarity information generating unit 30 extracts the frequency of words appearing in the text information of each performer. As the text information of the performer, for example, the utterance in the content of the performer can be adopted by voice recognition. The speech recognition method in this case is not particularly limited, and for example, a method using an acoustic model such as an HMM can be employed. For example, text information such as a performer's work or an interview article can be adopted as the performer's text information. What is necessary is just to acquire text information, such as a performer's work and an interview article, from the outside or the content information holding | maintenance part 24 grade | etc.,.

  Next, the person similarity information generation unit 30 extracts feature amounts corresponding to the “language” similarity calculation method from each piece of text information of a plurality of performers.

  For example, when the so-called vector space method is adopted as the “language” similarity calculation method, each word that appears in each text information of a plurality of performers is converted into the “language” similarity calculation method. It is extracted as a feature amount according to the above. The feature amount corresponding to the “language” similarity calculation method is an example of the information of the performer in the process of step S4.

  Therefore, in the process of step S5, the similarity between each person is calculated based on an appropriate distance measure in the feature quantity space between the feature quantities of each extracted person. That is, each appearing word is considered as an independent axis, and each of a plurality of performers is expressed as a multidimensional vector (one point on the multidimensional space) of the multidimensional space with the appearance frequency as a value. The cosine distance between each multidimensional vector for each of the performers was calculated and normalized so that each cosine distance could be compared to other similar criteria ("face" or "voice"). A value (for example, a value in the range of 0 to 1 here) is calculated as the similarity between the performers.

  Note that the similarity calculation method corresponding to each of “face”, “voice”, and “language” is not limited to the above-described example, and it is needless to say that any other method can be adopted.

  Thus, when the similarity between each performer is calculated by the process of step S5, the process proceeds to step S6.

  In step S6, the person similarity information generation unit 30 describes the similarity between the performers in the similarity table of the processing target similarity criterion.

  Specifically, for example, the processing target similarity criterion is “face”, and “actor A”, “actor B”, “singer C”, “singer D”, “moderator E”, and “Moderator F” is extracted, and “Actor A”, “Actor B”, “Singer C”, “Singer D”, “Moderator E”, and “Moderator F” are registered in step S3. It is assumed that the similar similarity table is newly generated. In such a case, as a result of the processing in step S6, the similarity table of “face” in which each numerical value indicating the similarity as shown in FIG. 6 is described in the corresponding item is completed. It is held in the similarity information holding unit 31.

  Thereby, the similarity between each person can be easily understood. For example, paying attention to the similarity table of “face” in FIG. 6, the similarity between “Actor A” and “Actor B” is “0.6”, and “Actor A” and “Singer C” are similar. It can be easily seen that the similarity is “0.3”. Thus, it can be easily understood that the face of “Actor A” is more similar to the face of “Actor B” than the face of “Singer C”.

  Returning to FIG. 5, in step S <b> 7, the person similarity information generation unit 30 determines whether or not to create / update another similarity criterion similarity table.

  If it is determined in step S7 that another similarity criterion similarity table is to be created / updated, the process returns to step S2, and the subsequent processes are repeated. In other words, in the present embodiment, the loop processing of steps S2 to S7 is repeatedly executed for each of “face”, “voice”, and “language”, and as a result, as shown in FIG. Each similarity table of “face”, “voice”, and “language” is completed and held in the person similarity information holding unit 31 as person similarity information.

  Thereby, in step S7, it is determined that the similarity table of another similarity criterion is not created / updated, and the person similarity information generation / update process ends.

  Since the content held by the content information holding unit 24 is sequentially updated, the person similarity information generating unit 30 appropriately executes the person similarity information generating / updating process, for example, periodically, The content of the person similarity information held in the similarity information holding unit 31 may be updated as appropriate.

  When such person similarity information generation / update processing is executed one or more times and the person similarity information is held in the person similarity information holding unit 31, the content recommendation processing device 10 is shown in FIG. Content recommendation processing can be executed. That is, the flowchart of FIG. 7 shows an example of content recommendation processing.

  In step S <b> 21, the recommended content extraction unit 32 of FIG. 2 extracts a specific person that can be determined to be liked by the user from the user preference information held in the user preference information holding unit 29.

  Note that the extraction of the specific person means extraction of personal information of the specific person. Specifically, for example, in the present embodiment, as described above, information called Person (person) having a numerical value (preference degree) equal to or higher than a predetermined threshold is extracted as a specific person in the performer information Pup of FIG. Is done.

  In this case, it should be noted that when there are a plurality of persons (persons) having a numerical value (preference degree) equal to or greater than a predetermined threshold, the plurality of persons are respectively extracted as specific persons. That is, the number of specific persons extracted in the process of step S21 is not limited to one, and there may be a plurality of persons.

  In step S22, the recommended content extraction unit 32 determines the content in which the specific person appears as a recommended candidate. In other words, in the present embodiment, the recommended content extracting unit 32 recommends a content having content information in which a specific person is included in the performer information Pm in FIG. 3 (the corresponding component of the vector is 1) as a recommended candidate. Determine as.

  In step S23, the recommended content extraction unit 32 extracts a similar person of a specific person based on the human similarity information held in the human similarity information holding unit 31 (hereinafter referred to as a similar person extraction process). Execute. A detailed example of the similar person extraction process will be described later with reference to FIGS.

  In step S <b> 24, the recommended content extraction unit 32 determines content in which a similar person appears as a recommended candidate. In other words, in the present embodiment, the recommended content extracting unit 32 recommends a content having content information in which a similar person is included in the performer information Pm in FIG. 3 (the corresponding component of the vector is 1) as a recommended candidate. Determine as.

  In step S25, the recommended content extraction unit 32 selects recommended content from the recommended candidates.

  In step S <b> 26, the recommended content extraction unit 32 outputs a list of recommended content (recommendation information) via the recommendation information output unit 33.

  In the example of FIG. 7, both the content in which a specific person appears and the content in which a similar person appears are determined as recommended candidates, but either one may be determined as recommended content. That is, a list of contents in which a specific person appears can be output as recommendation information as in the conventional case, and a list of contents in which a similar person appears can also be output as recommendation information.

  Next, a detailed example of the process of step S23, that is, the similar person extraction process, will be described with reference to the flowchart of FIG.

  In step S41, the recommended content extraction unit 32 determines one or more types of use similarity criteria.

  The use similarity criterion is a specific person in order to extract a similar person among the above-described various similarity standards, that is, in the present embodiment, “face”, “voice”, and “language”. The similarity criterion used when calculating the similarity to.

  In step S42, the recommended content extraction unit 32 sets a processing target similarity criterion as a predetermined one of the usage similarity criteria.

  In step S43, a predetermined one of the specific persons is set as a process target specific person.

  That is, as described above, since there may be a plurality of specific persons, the similar person extraction process in the example of FIG. 8 is provided with the process of step S43. Further, the processes of steps S45 to S47 described later are provided. Processing is provided. In other words, if it is assumed that there is only one specific person, the processes in steps S43 and S45 to S47 can be omitted.

  In step S <b> 44, the recommended content extraction unit 32 acquires the similarity of each performer with respect to the processing target specific person from the person similarity information held in the person similarity information holding unit 31.

  Specifically, for example, the similarity table of “face” in the state of FIG. 6 is held in the person similarity information holding unit 31, the processing target similarity criterion is “face”, and the processing target specific person is “ In the case of “Actor A”, as the similarity of each performer to the processing target specific person, the similarity “0.6” of “Actor B” to “Actor A” and “Singer C” of “Actor A” Similarity “0.3”, “Singer D” similarity “0.8” to “Actor A”, “Moderator E” similarity “0.5” to “Actor A”, and “Actor A” The degree of similarity “0.4” of “Moderator F” with respect to “

  In step S45, the recommended content extraction unit 32 determines whether there is a specific person that has not yet been set as the process target specific person.

  If there are a plurality of specific persons, and there is a person who has not yet been set as a process target specific person among the plurality of specific persons, it is determined as YES in the process of step S45. Then, the process returns to step S43, and the subsequent processes are repeated.

  That is, the loop process of steps S43 to S45 is repeatedly executed for each of one or more specific persons, and the similarity of each performer for each of the one or more specific persons is acquired. Thereafter, it is determined as NO in the process of step S45, and the process proceeds to step S46.

  In step S46, the recommended content extraction unit 32 determines whether or not there are a plurality of specific persons.

  When there are a plurality of specific persons, the recommended content extraction unit 46 determines YES in step S46, and in step S47, based on the similarities of the specific persons with respect to the predetermined performer, a predetermined content is determined. Calculate the overall similarity to the performers. Such total similarity is calculated for each performer. Hereinafter, such processing in step S47 is referred to as total similarity calculation processing for a plurality of specific persons. Details of the total similarity calculation processing for a plurality of specific persons will be described later with reference to FIG. When the total similarity calculation process for a plurality of specific persons in step S47 ends, the process proceeds to step S48.

  On the other hand, if there is one specific person, it is determined as NO in step S46, and the overall similarity calculation process of the plurality of specific persons in step S47 is not executed, and the process directly proceeds to step S48. move on.

  In step S48, the recommended content extraction unit 32 determines whether there is a use similarity criterion that has not yet been set as the processing target similarity criterion.

  If it is determined YES in the process of step S48, that is, a plurality of types of use similarity criteria are determined in the process of step S41, and the processing target use similarity criteria are included in the plurality of types of use similarity criteria. If there is a type that has not yet been set, the process returns to step S42, and the subsequent processes are repeated.

  That is, the loop process of steps S42 to S48 is repeatedly executed for each of one or more types of usage similarity criteria, and the similarity of each performer for a specific person is acquired for each of one or more types of usage similarity criteria. The However, when there are a plurality of specific persons, the total similarity is calculated. Thereafter, it is determined NO in the process of step S48, and the process proceeds to step S49.

  In step S <b> 49, the recommended content extraction unit 32 determines whether there are multiple types of use similarity criteria.

  If it is determined as YES in step S49, that is, if a plurality of types of use similarity criteria have been determined in the process of step S41, the recommended content extraction unit 46 determines each of the plurality of types of use similarity criteria in step S50. The total similarity of each performer with respect to a specific person is calculated based on each similarity of each performer with respect to the specific person. Hereinafter, such processing in step S50 will be referred to as a plurality of use similarity criterion total similarity calculation processing. Details of the overall similarity calculation processing of the plurality of use similarity criteria will be described later with reference to FIG. When the overall similarity calculation process of the plurality of use similarity criteria in step S50 ends, the process proceeds to step S51.

  On the other hand, when it is determined as NO in step S49, that is, when one type of use similarity criterion is determined in the process of step S41, the total similarity calculation process of the plurality of use similarity criteria in step S50 is performed. Without being executed, the process proceeds to step S51 as it is.

  In step S51, the recommended content extraction unit 32 determines performers whose similarity is equal to or greater than a threshold as similar persons.

  Specifically, for example, the similarity table of “face” in the state of FIG. 6 is held in the person similarity information holding unit 31, the use similarity criterion is only “face”, and the specific person is “actor A”. It is assumed that the processing of steps S41 to S50 described above is executed when there is only one person. In such a case, the similarity compared with the threshold value in the process of step S51 is the similarity “0.6” of “Actor B” to “Actor A” and the similarity of “Singer C” to “Actor A”. Degree “0.3”, “singer D” similarity “0.8” to “actor A”, “moderator E” similarity “0.5” to “actor A”, and “actor A” The degree of similarity “0.4” of “Moderator F” with respect to Therefore, for example, when “0.7” is set as the threshold, “singer D” having a similarity “0.8” is determined as a similar person.

  Further, for example, when the total similarity calculation processing of a plurality of specific persons in step S47 or the total similarity calculation processing of a plurality of use similarity criteria in step S50 has been executed, the total obtained as a result of those calculation processing The similar similarity is compared with a threshold value, and a performer whose overall similarity is equal to or greater than the threshold value is determined as a similar person.

  Thus, when a similar person is determined in the process of step S51, the similar person extraction process ends, that is, the process of step S23 of FIG. 7 described above ends, and the process of next step S24 performs this process. The content in which the similar person determined in the process of step S51 appears is determined as a recommendation candidate.

  Next, the details of each process of steps S47 and S50 in the similar person extraction process, that is, the details of the total similarity calculation process of a plurality of specific persons and the total similarity calculation process of a plurality of used similarity criteria Will be explained individually in that order.

  The flowchart of FIG. 9 shows a detailed example of the total similarity calculation process for a plurality of specific persons.

  In step S <b> 71, the recommended content extraction unit 32 extracts each degree of preference of a plurality of specific persons from the user preference information held in the user preference information holding unit 29.

  Specifically, for example, in the present embodiment, each numerical value associated with each of a plurality of specific persons in the performer information Pup of FIG. 4 is extracted as each degree of preference of the plurality of specific persons. That is, for example, when “personA” is included in the specific person, “10” is extracted as the degree of preference. For example, when “personB” is included in the specific person, “43” is extracted as the degree of preference.

  In step S72, the recommended content extraction unit 32 sets a predetermined one of the plurality of performers as a process target performer.

  In step S <b> 73, the recommended content extraction unit 32 weights each similarity of the processing target performer for each of the plurality of specific persons using the corresponding specific person's preference degree. In step S74, the recommended content extraction unit 32 determines the total addition of the weighted similarities as the similarity of the process target performer.

  In step S75, the recommended content extraction unit 32 determines whether there is a performer that has not yet been set as a process target performer.

  If it is determined as YES in the process of step S75, that is, if there is a performer that has not yet been set as a process target performer, the process returns to step S72, and the subsequent processes are repeated.

  That is, the loop processing of steps S72 to S75 is repeatedly executed for each of the plurality of performers, and the overall similarity of each of the plurality of performers is calculated.

  Specifically, for example, “actor A” and “singer D” are specific persons, and the respective values obtained by normalizing the respective preference degrees extracted in the process of step S71 are “0.7” and “0”. .3 ". Then, “0.7” is adopted as the weight value of “Actor A”, and “0.3” is adopted as the weight value of “Singer D”.

  Further, it is assumed that the overall similarity for “face” is calculated, and the target person of the similarity calculation, that is, a plurality of performers are “actor B”, “singer C”, “moderator E”, and It is assumed that it is “Moderator F”. In addition, it is assumed that the respective degrees of similarity between the persons are the values shown in the “face” similarity table in the state of FIG.

  That is, the similarity to “Actor A” is “0.6” for “Actor B”, “0.3” for “Singer C”, and “0.5” for “Moderator E”. It is assumed that “Moderator F” is “0.4”.

  The similarity to “Singer D” is “0.4” for “Actor B”, “0.5” for “Singer C”, and “0.5” for “Moderator E”. It is assumed that “Moderator F” is “0.5”.

  In this case, when “actor B” is set as the process target performer in the process of step S72, the similarity “0.6” of “actor B” with respect to “actor A” is the weight value “ It is weighted with “0.7” to become “0.6 * 0.7”, and the similarity “0.4” of “Actor B” with respect to “Singer D” is weighted with a weight value “0.3”. “0.4 * 0.3”.

  Therefore, in the process of step S74, the calculation of the following equation (1) is executed, and the calculation result “0.54” is determined as the overall similarity of “actor B”.

  0.6 * 0.7 + 0.4 * 0.3 = 0.54 (1)

  Similarly, when “singer C” is set as a performer to be processed, the calculation of the following expression (2) is executed in the processes of steps S73 and S74, and the calculation result “0.36” is “ It is determined as the overall similarity of “Singer C”.

  0.3 * 0.7 + 0.5 * 0.3 = 0.36 (2)

  Further, when “moderator E” is set as a process target performer, the calculation of the following expression (3) is executed in steps S73 and S74, and the calculation result “0.5” is “ It is determined as the overall similarity of the moderator E ”.

  0.5 * 0.7 + 0.5 * 0.3 = 0.5 (3)

  Further, when “moderator F” is set as the performer to be processed, the calculation of the following equation (4) is executed in the processes of steps S73 and S74, and the calculation result “0.43” is “ It is determined as the overall similarity of the moderator F ”.

  0.4 * 0.7 + 0.5 * 0.3 = 0.43 (4)

  Therefore, in this numerical example, the overall similarity is “Actor B”> “Moderator E”> “Moderator F”> “Singer C”.

  In this way, when the overall similarity of each of the plurality of performers is calculated, it is determined as NO in the process of step S75, and the overall similarity calculation process of the plurality of specific persons is completed. That is, the process of step S47 of FIG. 8 described above ends, and the process proceeds to step S48.

  In this case, if there is only one type of use similarity criterion determined in the process of step S41, the total similarity calculation processing of the plurality of use similarity criteria in step S50 is not executed, so that the plurality of specific persons in this step S47 The overall similarity calculation processing result, that is, the overall similarity of each performer is compared with the threshold value in the process of step S51.

  Specifically, for example, when the overall similarity of each performer is the numerical example described above, and the threshold is “0.4”, “Actor B” having the overall similarity “0.54” And “Moderator E” having the overall similarity “0.5” are determined as similar persons.

  Next, with reference to the flowchart of FIG. 10, an example of the overall similarity calculation process of a plurality of use similarity criteria, that is, a detailed example of the process of step S50 of FIG.

  In step S <b> 91, the recommended content extraction unit 32 acquires the genre of content that can be a recommendation candidate from the content information held in the content information holding unit 24.

  In step S92, the recommended content extraction unit 32 sets a predetermined one of the acquired genres as a processing target genre.

  In step S <b> 93, the recommended content extraction unit 32 determines each weight coefficient of a plurality of similarity criteria for the processing target genre.

  In step S94, the recommended content extraction unit 32 sets a predetermined one of a plurality of performers as a process target performer.

  In step S <b> 95, the recommended content extraction unit 32 weights each degree of similarity of the processing target performer for each of the plurality of use similarity criteria using the corresponding use similarity criteria weighting factor. In step S <b> 96, the recommended content extraction unit 32 determines the total addition of the weighted similarities as the similarity of the process target performer.

  In step S97, the recommended content extraction unit 32 determines whether there is a performer that has not yet been set as a process target performer.

  If it is determined as YES in the process of step S97, that is, if there is a performer that has not yet been set as a process target performer, the process returns to step S94, and the subsequent processes are repeated.

  That is, the loop processing of steps S94 to S98 is repeatedly executed for each of the plurality of performers for the processing target genre, and the overall similarity of each of the plurality of performers is calculated. Thereafter, it is determined as NO in the process of step S97, and the process proceeds to step S98.

  In step S98, the recommended content extraction unit 32 determines whether there is a genre that has not yet been set as the processing target genre.

  If it is determined as YES in the process of step S98, that is, if there is a genre that has not yet been set as the process target genre, the process returns to step S92, and the subsequent processes are repeated.

  That is, the loop process of steps S92 to S98 is repeatedly executed for each genre acquired in the process of step S91, and the overall similarity of each of the plurality of performers is calculated. Thereafter, it is determined as NO in the process of step S98, and the process proceeds to step S99.

  In step S99, the recommended content extraction unit 32 determines whether there are a plurality of genres acquired in the process of step S91.

  If there are a plurality of genres acquired in the process of step S91, it is determined YES in the process of step S99, and the process proceeds to step S100. In step S100, the recommended content extraction unit 32 calculates a genre integrated similarity for each of a plurality of performers.

  Here, the genre integrated similarity for a predetermined performer uses each similarity determined in the process of step S96 for each of the plurality of genres (hereinafter referred to as genre similarity) for the predetermined performer. Thus, it means one similarity calculated according to a predetermined calculation rule. The predetermined calculation rule is not particularly limited, and for example, a calculation rule that weights each genre similarity and sets the average value of each genre similarity after the weight as the genre integrated similarity can be adopted. Further, for example, a calculation rule that sets the total addition value of each genre similarity after weighting as the genre integrated similarity can be adopted.

  When the process of step S100 is completed, the overall similarity calculation process for a plurality of use similarity criteria is ended. That is, the process of step S50 of FIG. 8 described above ends, and the process proceeds to step S51.

  On the other hand, if there is only one genre acquired in the process of step S91, it is determined NO in the process of step S99, and the process of step S100 is not executed, and a plurality of usage similarities are obtained. The reference total similarity calculation process ends. That is, the process of step S50 of FIG. 8 described above ends, and the process proceeds to step S51.

  As described above, the total similarity calculation processing of a plurality of usage similarity criteria is characterized in that the usage similarity criteria to be emphasized are changed according to the contents (genre in this case). That is, in order to change the use similarity criterion to be emphasized depending on a predetermined genre, in the example of FIG. 10, the weight coefficient determined in the process of step S93 is used as a value indicating the importance level.

  In other words, the weighting factor can be arbitrarily determined according to how the weighting factor is emphasized. That is, the process of step S93 only needs to finally determine the weighting coefficient, and the processing content is not particularly limited.

  For example, a process of dynamically updating a user's response to past recommended content, that is, a process of calculating each weighting coefficient each time according to a predetermined calculation method can be employed as the process of step S93.

  Further, for example, a process of presetting and holding each weighting coefficient of each similarity standard for each genre and extracting each held weighting coefficient may be adopted as the process of step S93. Specifically, for example, weight coefficient information as shown in FIG. 11 is held in the person similarity information holding unit 31 and the like in FIG. 2 in advance, and the recommended content extraction unit 32 performs the processing target genre as processing in step S93. By acquiring the “face” coefficient α, the “voice” coefficient β, and the “language” coefficient γ, the values can be determined as the weighting coefficients.

  In this case, as the processing of steps S95 and S96, the following equation (5) is calculated, and the overall similarity of the processing target performer for the processing target genre is determined.

  Total similarity = α × “face” similarity + β × “voice” similarity + γ × “language” similarity (5)

  In Expression (5), the “face” similarity refers to the similarity of the process target performer when the use similarity criterion is “face”. That is, α × “face” similarity is the result of weighting the “face” similarity by the process of step S95. Similarly, “voice” similarity refers to the similarity of performers to be processed when the use similarity criterion is “voice”. That is, β × “voice” similarity is the result of weighting the “voice” similarity by the process of step S95. Further, the “language” similarity refers to the similarity of performers to be processed when the use similarity criterion is “language”. In other words, γ × “language” similarity is the result of weighting the “language” similarity by the process of step S95.

  Further, the “total similarity” calculated by Expression (5) is the similarity for one genre that is the processing target genre, that is, the “genre similarity” described above in the description of the processing in step S100. ". Therefore, when there are a plurality of genres acquired in the process of step S91, each genre similarity is calculated for each of the plurality of genres using Equation (5). In the process of step S100, one similarity obtained by integrating the genre similarities for each of a plurality of genres, that is, a genre integrated similarity is calculated.

  If “face” is not determined as the use similarity criterion in the process of step S41 of FIG. 8 described above, the “face” coefficient α = 0 is handled, and the “voice” coefficient β and the “language” coefficient γ are as follows. Then, normalization may be performed again so that the added value becomes 1. That is, this process may be executed as the process of step S93.

  Similarly, when “voice” is not determined as the use similarity criterion in the process of step S41 of FIG. 8 described above, the “voice” coefficient β = 0 is handled, and the “face” coefficient α and the “language” coefficient γ are processed. May be normalized again so that the added value becomes 1. That is, this process may be executed as the process of step S93. Further, when “language” is not determined as the use similarity criterion in the process of step S41 of FIG. 8 described above, the “language” coefficient γ = 0 is handled, and the “face” coefficient α and the “voice” coefficient β are as follows. Then, normalization may be performed again so that the added value becomes 1. That is, this process may be executed as the process of step S93.

  Further, instead of the processing in steps S95 and S96, the processing target genre is obtained by integrating three similarity measures, that is, “face”, “voice”, and “language” similarities using a Bayesian network. You may employ | adopt the process which determines the comprehensive similarity of a process target performer.

  Further, as the contents, the genre is adopted in the example of FIG. 10, but is not particularly limited to the example of FIG. That is, by simply replacing “genre” in FIG. 10 with arbitrary “contents content”, it is possible to easily realize the total similarity calculation processing of a plurality of use similarity criteria for arbitrary “contents content”.

  Furthermore, the feature of the total similarity calculation processing of a plurality of use similarity criteria, that is, the feature of changing the use similarity criteria to be emphasized according to the content (genre in this case) is another process, for example, the step of FIG. It can also be applied to the process of S41. That is, a method of determining according to the content genre can be adopted as the method of determining the use similarity criterion for the processing in step S41. Specifically, for example, if the genre of the content is information or liberal arts, “Language” to emphasize the utterance content of the performer, and “Voice” to emphasize the voice quality of the voice actor in the case of animation, music In step S41, a method of determining “voice” for emphasizing the voice quality of the singer in the case of “sound” and “face” for emphasizing the face information of the performer in the case of a drama or movie is used as the use similarity criterion. Can be adopted.

  The person similarity information generation / update process in FIG. 5 and the content recommendation process in FIG. 7 and the like have been described as being executed by the content recommendation processing apparatus 10 having the functional configuration in FIG. That is, the description has been made on the assumption that the person similarity information generation / update process of FIG. 5 and the content recommendation process of FIG. 7 and the like are executed by the same device. However, since the person similarity information generation / update process in FIG. 5 and the content recommendation process in FIG. 7 and the like are independent processes, it is not always necessary to execute both processes in the same apparatus. Absent.

  For example, the content recommendation processing device 10 includes a content information holding unit 24, a user preference information holding unit 29, a person similarity information holding unit 31, a recommended content extraction unit 32, a recommended information output unit 33, and a recommended information holding unit in FIG. It is possible to delegate only the functions of the other functional blocks in FIG. 2 to another apparatus.

  Specifically, for example, the distribution server 5 of FIG. 1 may have the function of the person similarity information generation unit 30. In this case, the distribution server 5 periodically executes the person similarity information generation / update process of FIG. 5, and the process result, that is, the generated or updated person similarity information, is recommended via the network 8 as a content recommendation. It can be distributed to the processing device 10. In practice, a very large number of users individually own the respective content recommendation processing devices 10. Therefore, in consideration of this, the person similarity information generation / update processing of FIG. It is preferable to execute on the distribution server 5 side rather than on the processing device 10 side.

  By the way, the series of processes described above can be executed by hardware, but can also be executed by software.

  In this case, for example, a personal computer shown in FIG. 12 may be employed as at least a part of the content recommendation processing device 10 and the like of FIG.

  In FIG. 12, a CPU (Central Processing Unit) 101 executes various processes according to a program recorded in a ROM (Read Only Memory) 102 or a program loaded from a storage unit 108 to a RAM (Random Access Memory) 103. To do. The RAM 103 also appropriately stores data necessary for the CPU 101 to execute various processes.

  The CPU 101, ROM 102, and RAM 103 are connected to each other via a bus 104. An input / output interface 105 is also connected to the bus 104.

  The input / output interface 105 includes an input unit 106 including a keyboard and a mouse, an output unit 107 including a display, a storage unit 108 including a hard disk, and a communication unit 109 including a modem and a terminal adapter. It is connected. The communication unit 109 controls communication performed with other devices (not shown) via a network including the Internet.

  A drive 110 is connected to the input / output interface 105 as necessary, and a removable medium 111 made up of a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is appropriately attached, and a computer program read from them is loaded. These are installed in the storage unit 108 as necessary.

  When a series of processing is executed by software, a program constituting the software executes various functions by installing a computer incorporated in dedicated hardware or various programs. For example, a general-purpose personal computer is installed from a network or a recording medium.

  As shown in FIG. 12, the recording medium containing such a program is distributed to provide a program to the user separately from the apparatus main body, and a magnetic disk (including a floppy disk) on which the program is recorded. , Removable recording media (packages) consisting of optical disks (including compact disk-read only memory (CD-ROM), DVD (digital versatile disk)), magneto-optical disks (including MD (mini-disk)), or semiconductor memory (Media) 111, but also a ROM 102 on which a program is recorded and a hard disk included in the storage unit 108 provided to the user in a state of being incorporated in the apparatus main body in advance.

  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 order, but is not necessarily performed in chronological order, either in parallel or individually. The process to be executed is also included.

  Further, in the present specification, the system represents the entire apparatus including a plurality of apparatuses and processing units.

  Further, for example, in the above-described embodiment, only the television broadcast program and the distribution streaming data are employed as the content, but it is naturally not limited to the above-described embodiment. That is, the content is widely generated by human creative activities, and the present invention can be applied to such content as a whole.

  For example, movies, music, theatre, literary arts, photographs, comics, animations, computer games and other characters, figures, colors, sounds, actions or videos, or combinations thereof, or information related thereto are provided via an electronic computer. The program for this is an example of content, and any of them can be applied to the present invention. That is, these various contents can be targeted for recommendation. Furthermore, in this case, there is no particular need to limit the content recommendation based on a specific person that the user likes, and content recommendation based on a specific object can be performed widely. In this case, by performing the series of processes described above, it is possible to recognize a similar object similar to the specific object and perform content recommendation based on the similar object.

It is a figure explaining a television program broadcast and delivery of stream data. It is a functional block diagram which shows the functional structural example of the content recommendation processing apparatus of FIG. 1 as an information processing apparatus with which this invention is applied. It is a figure which shows an example of the content information hold | maintained at the content information holding | maintenance part of FIG. It is a figure which shows an example of the user preference information hold | maintained at the user preference information holding part of FIG. 3 is a flowchart illustrating an example of person similarity information generation / update processing executed by the content recommendation processing device of FIG. 2. 6 is a diagram illustrating an example of person similarity information generated by the person similarity information generation / update process of FIG. 5 and held in the person similarity information holding unit 31 of FIG. It is a flowchart explaining the example of a content recommendation process which the content recommendation processing apparatus of FIG. 2 performs. It is a flowchart explaining the detailed example of the similar person extraction process of step S23 of FIG. It is a flowchart explaining the detailed example of the total similarity calculation process of several specific persons of step S47 of FIG. It is a flowchart explaining the detailed example of the total similarity calculation process of the some use similarity reference | standard of step S50 of FIG. It is a figure which shows an example of the weight coefficient information utilized by the total similarity calculation process of the some use similarity reference | standard of FIG. It is a block diagram which shows the structural example of the personal computer as an image processing apparatus with which this invention is applied.

Explanation of symbols

  5 distribution server, 10 content recommendation processing device, 24 content information holding unit, 29 user preference information holding unit, 30 person similarity information generating unit, 31 person similarity information holding unit, 32 recommended content extracting unit, 33 recommended information output unit , 101 CPU, 102 ROM, 103 RAM, 108 storage unit, 111 removable recording medium

Claims (9)

In an information processing device that recommends content to a user,
Preference information holding means for holding preference information related to a specific person preferred by the user;
Person similarity information holding means for holding person similarity information indicating each degree of similarity based on a predetermined similarity criterion between a plurality of persons including the specific person;
A similar person that recognizes the specific person based on the preference information held in the preference information holding means and can determine that the specific person is similar to the specific person based on the person similarity information held in the person similarity information holding means An information processing apparatus comprising: a recommendation unit that recognizes at least one of the content in which the specific person appears and the content in which the similar person appears as a candidate of recommended content to be recommended to the user. The information processing apparatus according to claim 1, wherein the similarity based on the predetermined similarity criterion is a similarity based on facial features of a person. The information processing apparatus according to claim 1, wherein the similarity based on the predetermined similarity criterion is a similarity based on a feature of a voice uttered by a person. The information processing apparatus according to claim 1, wherein the similarity based on the predetermined similarity criterion is a similarity based on characteristics of a language used by a person. The person similarity information includes information indicating each similarity based on each of a plurality of similarity criteria,
The recommendation means is:
For each of the plurality of similarity criteria, obtaining each similarity between the similar person candidate and the specific person from the person similarity information,
Based on the respective similarities for each of the acquired plurality of similar criteria and the importance of the plurality of similar criteria changing according to the content, the overall similarity of the similar person candidates is calculated. Operate,
The information processing apparatus according to claim 1, wherein it is determined whether to recognize the similar person candidate as the similar person based on the total similarity. The recommendation means is:
Recognizing a plurality of persons as the specific person,
Obtaining each similarity between the candidate for similar person and each of the plurality of specific persons from the person similarity information;
Based on the obtained similarities, the overall similarity of the similar person candidates is calculated,
The information processing apparatus according to claim 1, wherein it is determined whether to recognize the similar person candidate as the similar person based on the total similarity. Similarity calculation means for calculating each similarity based on the predetermined similarity criterion between the plurality of persons including the specific person, and causing the person similarity information holding means to store the calculation results as the person similarity information; The information processing apparatus according to claim 1, further comprising: Preference information holding means for holding preference information related to a specific person preferred by the user;
An information processing method of an information processing apparatus for recommending content to the user, comprising: person similarity information holding means for holding person similarity information indicating each degree of similarity based on a predetermined similarity criterion between a plurality of persons including the specific person There,
A similar person that recognizes the specific person based on the preference information held in the preference information holding means and can determine that the specific person is similar to the specific person based on the person similarity information held in the person similarity information holding means And determining at least one of the content in which the specific person appears and the content in which the similar person appears as a recommended content candidate to be recommended to the user. Preference information holding means for holding preference information related to a specific person preferred by the user;
Person similarity information holding means for holding person similarity information indicating each degree of similarity based on a predetermined similarity criterion between a plurality of persons including the specific person,
A program executed by a computer that controls a recommendation device that recommends content to the user,
A similar person that recognizes the specific person based on the preference information held in the preference information holding means and can determine that the specific person is similar to the specific person based on the person similarity information held in the person similarity information holding means And determining at least one of the content in which the specific person appears and the content in which the similar person appears as a recommended content candidate to be recommended to the user.

Images