JP2011244340A - Program recommendation apparatus, method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To recommend programs more appropriate to a user in a video distribution service using the Internet line.SOLUTION: The frequency of each word included in viewed programs is obtained from a user's viewing history, and the frequency of each word is also obtained from a video recording history. The frequency of each word included in programs scheduled to be broadcasted is also obtained. By totalling the frequencies, weights are obtained. According to the frequencies and weights, the degree of recommendation of each program to the user is obtained. On the basis of the degree of the recommendation, the programs more appropriate to the user are recommended.

Description

  The present invention relates to a program recommendation apparatus, method, and program, and more particularly, to a program recommendation apparatus, method, and program for viewing a television program of a user provided by a video distribution service using an Internet line.

  A number of information recommendation methods have been proposed. For example, in order to recommend a news article, the user is given a five-step evaluation for each news article, the past evaluation history is viewed, the degree of preference similarity between users is measured, and the news article preferred by a user with a similar preference Has been proposed (see, for example, Non-Patent Document 1).

  Similarly to the above-described method, the user is evaluated for each piece of information, and attention is paid to text included in the information evaluated by the user in the past evaluation history. There has been proposed a method for recommending information in which words similar to the text appear, considering that the text reflects the user's interest (for example, see Non-Patent Document 2).

  Also, let the user specify the text of the part of interest, estimate the user's interest based on the appearance frequency of the words contained therein, search the information in the community based on the word representing the user's interest, There is a method of presenting only information of a portion that is of interest to the user (for example, see Patent Document 1).

JP 11-338869 A

Resnick, P. et al., "GroupLens: An open architecture for collaborative filtering of netnews". CSCW '94: Proceedings of the 1994 ACM conference on Computer supported cooperative work, ACM Press New York, 1994. Mooney R.J. et al, `` Content-Based Book Recommending Using Learning for Text Categorization '', Proceedings of the 5th ACM Conference on Digital Libraries, ACM Press New York, 2000

  Multi-channel video distribution services using internet lines are progressing. Under these circumstances, it is difficult for the user to find a program that is really interesting. It is also difficult to grasp and record all the programs of interest.

  When a program is recommended using the technique disclosed in Patent Literature 1, a program related to an interesting word designated by the user is recommended. In this case, there is a problem that a word must be specified. If the specified word is not enough, the program is not recommended to the user.

  When a program is recommended by the methods described in Non-Patent Documents 1 and 2, a program is recommended to the user using information on the program viewed by the user, but information recorded by the user in the program information is not considered. Or if the information of the recorded program is used, the information that the user watched is not considered. Without using both viewing and recording information, it is not possible to recommend a more appropriate program to the user.

  The present invention has been made in view of the above points, and in a video distribution service using an Internet line, a program recommendation device capable of recommending a program more appropriately by using a log relating to user viewing and recording, and An object is to provide a method and a program.

In order to solve the above-mentioned problems, the present invention (Claim 1) is directed to a program recommendation that uses a user's viewing history, particularly information about whether or not a program has been recorded by a user, in a video distribution service using an Internet line. A device,
Program information acquisition means for acquiring program information in a video distribution service using an Internet line and storing the information in a program information storage means;
Word extraction means for extracting words from summary text of contents in program information;
A statistic calculating means for calculating a statistic of each word in each program information and storing it in a statistic storage means;
Viewing information acquisition means for acquiring viewing information of viewing or recording of each user in a video distribution service using the Internet line, and storing the viewing information in the viewing information storage means;
From the acquired viewing information, the user's interest amount for the word is calculated from the statistics of each word calculated from the summary text of the content stored in the program information storage means of the program that is the target of the viewing information. , An interest amount calculating means for storing the identifier of the user and the interest amount storage means together with the word;
From the statistics for each word of each user stored in the interest amount storage means and the statistics for each word of only the pre-broadcast program among the programs stored in the program information storage means, the user's question Interest matching degree calculating means for calculating the interest matching degree for the user and storing the user identifier in the interest matching degree storage means together with the program identifier;
Of the programs in the program information storage means, refer to the viewing information storage means to select only the recorded programs, and from the statistics of each word of the program to be recorded stored in the statistics storage means Calculating a good interest amount for the word of the user, and storing a good interest amount calculating means for storing the user identifier and the good interest amount storage means together with the word;
From the statistics for each word of each user stored in the good interest amount storage means and the statistics for each word of only the pre-broadcast program among the programs stored in the program information storage means, the user's A good interest match degree calculating means for calculating a good interest match degree for a program, and storing the user identifier in the good interest match degree storage means together with the identifier of the program;
Of the programs stored in the program information storage means, the program information storage for all combinations of the broadcasted programs and the programs that have been viewed among the users stored in the program information storage means. Whether or not the user has recorded the program stored in the means, the interest match degree stored in the interest match degree storage means, and the good interest match degree stored in the good interest match degree storage means. A weight parameter calculating means for calculating a weight parameter and storing it in a weight parameter storage means;
Given a user identifier, for each pre-broadcast program stored in the user's program information storage means, the interest match degree stored in the interest match degree storage means and the good interest match degree storage means The recommended score for each program of the user is calculated using the degree of good interest matching stored in the weight parameter and the weight parameter stored in the weight parameter storage means, and the identifiers of the top N questions of the recommended score are recommended. Program recommendation means for selecting as a program.

Further, the present invention (Claim 2), in the statistic calculation means,
When calculating the statistic of each word, a TF value (word appearance frequency) and a TFIDF value (word appearance frequency and reverse appearance frequency) for extracting words regarded as important are used.

Further, the present invention (Claim 3) is a weight parameter calculating means,
Whether or not the user has recorded the program stored in the program information storage unit, the interest match level stored in the interest match level storage unit, and the good match level stored in the good interest match level storage unit Means for calculating a weight parameter by logistic regression analysis based on the degree of interest match.

Further, the present invention (Claim 4) is a program recommendation method that uses a user's viewing history, particularly information on whether or not a program has been recorded by a user, in a video distribution service using an Internet line,
Program information storage means, statistics amount storage means, viewing information storage means, interest amount storage means, interest match degree storage means, good interest amount storage means, good interest match degree storage means, weight parameter storage means are stored in the storage medium. Program information acquisition means, word extraction means, viewing information acquisition means, interest amount calculation means, interest match degree calculation means, good interest amount calculation means, good interest match degree calculation means, weight parameter calculation means, program recommendation means, In a device having
A program information acquisition unit acquires program information in a video distribution service using an Internet line, and stores the program information in a program information storage unit;
A word extracting means for extracting a word from the summary text of the content in the program information;
A statistic calculating unit calculates a statistic of each word in each program information and stores it in the statistic storage unit;
Viewing information acquisition means acquires viewing information of each user's viewing or recording in the video distribution service using the Internet line, and stores the viewing information in the viewing information storage means; and
From the statistics of each word calculated by the interest amount calculating means from the acquired viewing information and the summary text of the contents stored in the program information storage means of the program that is the target of the viewing information, the word of the user An interest amount calculation step of calculating an interest amount with respect to the user and storing it in the interest amount storage means together with the identifier of the user and the word;
The interest match degree calculation means is a statistic for each word of each user stored in the interest amount storage means, and a statistic for each word of only the pre-broadcast program among the programs stored in the program information storage means An interest matching degree calculation step of calculating an interest matching degree with respect to the question of the user and storing it in the interest matching degree storage means together with the identifier of the user and the identifier of the program;
The amount of interest calculation means refers to the viewing information storage means among the programs in the program information storage means, selects only the recorded programs, and is the recording target program stored in the statistics storage means A good interest amount calculation step of calculating a good interest amount for the user from the statistic of each word, and storing the user's identifier and the good interest amount storage unit together with the word;
The good interest coincidence calculating means calculates the statistics for each word of each user stored in the good interest amount storage means and the words of only the pre-broadcast programs among the programs stored in the program information storage means. From the statistics, the good interest match degree for the program of the user is calculated, and the good interest match degree calculation step of storing the identifier of the user and the good interest match degree storage means together with the identifier of the program;
The weight parameter calculation means includes all combinations of broadcasted programs among programs stored in the program information storage means and programs that have been viewed among users stored in the program information storage means. About whether the user has recorded the program stored in the program information storage means, the interest match degree stored in the interest match degree storage means, and the good interest match degree storage means A weight parameter calculating step for calculating a weight parameter using the good interest match degree and storing the weight parameter in a weight parameter storage means;
When the program recommendation means is given a user identifier, the interest match degree stored in the interest match degree storage means for each program before the broadcast stored in the program information storage means of the user, and good Using the good interest match degree stored in the interest match degree storage means and the weight parameter stored in the weight parameter storage means, a recommendation score for each program of the user is calculated, and the top N of the recommended scores are calculated. And a program recommendation step of selecting an identifier of the question as a recommended program.

Further, the present invention (Claim 5) is, in the statistic calculation step,
When calculating the statistic of each word, a TF value (word appearance frequency) and a TFIDF value (word appearance frequency and reverse appearance frequency) for extracting words regarded as important are used.

Further, the present invention (claim 6), in the weight parameter calculation step,
Whether or not the user has recorded the program stored in the program information storage unit, the interest match level stored in the interest match level storage unit, and the good match level stored in the good interest match level storage unit Based on the degree of interest match, a weight parameter is calculated by logistic regression analysis.

  The present invention (Claim 7) is a program recommendation program for causing a system to function as each means constituting the program recommendation apparatus according to any one of Claims 1 to 3.

  As described above, according to the present invention, an Internet line, which was impossible with the conventional technology, was realized by realizing a method of recommending a program more suited to the user's interest using not only viewing information but also recorded information. It is possible to recommend an appropriate program by the user in the video distribution service using the. This leads to an increase in viewing / recording in video distribution services using the Internet line.

It is a schematic diagram of the program recommendation apparatus in one embodiment of the present invention. It is an example of the program information table in one embodiment of this invention. It is an example of the viewing-and-listening information table in one embodiment of this invention. It is an example of the interest information table in one embodiment of this invention. It is an example of the good interest information table in one embodiment of this invention. It is an example of the interest matching degree table in one embodiment of this invention. It is an example of the good interest agreement degree table in one embodiment of this invention. It is a flowchart of the process of the program dictation in one embodiment of this invention. It is a flowchart of the detailed process of step 120 of FIG. 8 in one embodiment of this invention. It is a flowchart of the detailed process of step 130 of FIG. 8 in one embodiment of this invention. It is a flowchart of the process at the time of the program recording in one embodiment of this invention. It is a flowchart of the detailed process of step 220 of FIG. 11 in one embodiment of this invention. It is a flowchart of the detailed process of step 230 of FIG. 11 in one embodiment of this invention. It is a flowchart of the process at the time of user logon in one embodiment of this invention. It is a flowchart of the process at the time of the weight parameter update in one embodiment of this invention.

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

  FIG. 1 shows the configuration of a program recommendation device according to an embodiment of the present invention.

  The program recommendation device 1 shown in FIG. 1 includes a program information acquisition unit 10, a viewing information table update unit 20, an interest information table update unit 30, a good interest information table update unit 40, an interest match degree table update unit 50, and a good interest match degree. The table updating unit 60, the program recommendation unit 70, the weight parameter updating unit 80, the storage unit 90, the communication unit A0, and the input / output unit B0 are connected to the network 3, and the input / output unit B0 is It is connected to an external device 4 such as an input device or a display device.

  The storage unit 90 includes a program information table 91, a viewing information table 92, an interest information table 93, a good interest information table 94, an interest match level table 95, a good interest match level table 96, a disk storing a weight parameter recording unit 97, and the like. It is a storage medium. Below, each table in the memory | storage part 90 is demonstrated.

<Program information table>
As shown in FIG. 2, the program information table 91 includes a program ID field, a content field, and a broadcast time field.

  The program ID field is an identifier that identifies the program, and is set by the program information acquisition unit 10.

  In the content field, a summary about the content of the program is set as text data by the program information acquisition unit 10.

  In the broadcast time field, the broadcast time of the program is set by the program information acquisition unit 10.

<Viewing information table>
As shown in FIG. 3, the viewing information table 92 includes a program ID field, a user ID field, and a viewing / recording type field.

  The program ID field corresponds to the identifier of the program that appears in the program information table, and is set by the viewing information table update unit 20.

  In the user ID field, an identifier for identifying a user who has watched / recorded a program is set by the viewing information table update unit 20.

  In the viewing / recording type field, the viewing information table update unit 20 sets a value for determining whether the viewing / recording is viewing or recording.

<Interest information table>
As shown in FIG. 4, the interest information table 93 includes a user ID field, a word field, and a feature amount field.

  In the user ID field, the user ID is set by the interest information table update unit 30.

  In the word field, the word w is set by the interest information table update unit 30.

The feature amount field is a statistic interest ₀ (u, w) of the word w that appears in the content of the program viewed by the user u, and is set by the interest information table update unit 30.

<Good interest information table>
As shown in FIG. 5, the good / interest information table 94 includes a user ID field, a word field, and a feature field.

  In the user ID field, the user ID is set by the good / interest information table update unit 40.

  In the word field, the word w is set by the good interest information table update unit 40.

The feature amount field is a statistic interest ₁ (u, w) of the word w appearing in the content of the program recorded by the user u, and is set by the good interest information table update unit 50.

<Interest match table>
The interest match degree table 95 includes a user ID field, a program ID field, and an interest match degree field, as shown in FIG.

In the interest match degree field, the similarity score score ₀ (p, u) between the contents of the program u viewed in the past by the user u of the row and the program p of the row is set by the interest match degree table update unit 40.

<Good interest match degree table>
As shown in FIG. 7, the good interest match degree table 96 includes a user ID field, a program ID field, and a good interest match degree field.

In the good interest match degree field, the similarity score ₁ (p, u) between the contents of the program u recorded in the past of the user u in the row and the program p in the row is set by the good interest match table update unit 60. The

<Weight parameter recording section>
The weight parameter recording unit 97 includes weights λ ₀ and λ for calculating a recommendation score score (p, u) for the program recommendation unit 70 to recommend the program p to the user u as shown in the following equation (1). ₁ is stored.

重み λ₀、λ₁ は、重みパラメータ更新部８０によって設定される。

The weights λ ₀ and λ ₁ are set by the weight parameter update unit 80.

  Below, the process at the time of program viewing in the above configuration will be described.

  FIG. 8 is a flowchart of processing when viewing a program according to an embodiment of the present invention.

  Step 110) The viewing information table updating unit 10 sets a program ID field and a user ID field in the viewing information table 92 according to the viewed program and the viewed user. In addition, the viewing / recording type field is set to “viewing”. Such a row v is inserted into the viewing information table 92.

  Step 120) The interest information table update unit 30 updates the interest information table 82 based on the information in the row v.

  Step 130) The interest coincidence degree table updating unit 50 updates the interest coincidence degree table 84 based on the information in the row v.

  Next, the processing of step 120 in FIG. 8 will be described in detail.

  FIG. 9 is a flowchart of detailed processing of step 120 in FIG. 8 according to the embodiment of the present invention. Now, assume that the user in row v and the target program are u and p, respectively.

  Step 121) The program information acquisition unit 10 refers to the program information table 91 and acquires the value of the content field of the row where the program ID is p.

  Step 122) The interest information table update unit 30 acquires the word set W included in the contents of the program p using the segmentation function of the morphological analyzer.

  Step 123) The interest information table updating unit 30 performs, for each word w included in the word set W, a statistic in the content of the program p, for example, an appearance frequency (Document 3 (Manning, CD et al., “ “TF value”) described in “Introduction to information retrieval”, 2008)) tf (p, w) is calculated. In calculating the weight in the content of the program p of the word w, not only the appearance frequency but also another scale (such as “TFIDF value” described in Document 3) may be used.

Step 124) The interest information table update unit 30 refers to the interest information table 93 and obtains the feature value field interest ₀ (u, w) in the row where the user ID field value is u and the word field value is w. To do. However, if there is no row with the user ID field value u and the word field value w, the value of interest ₀ (u, w) is 0.

Step 125) The interest information table update unit 30 sets the value of the feature amount field of the row in which the value of the user ID field is u and the value of the word field is w in the interest information table 93 as interest ₀ (u, w) + tf ( Update or insert as p, w).

  Next, the processing of step 130 in FIG. 8 will be described in detail.

  FIG. 10 is a flowchart of detailed processing of step 130 in FIG. 8 according to the embodiment of the present invention. Suppose now that each user in row v is u.

  Step 131) The interest match degree table update unit 50 refers to the interest information table 93, and sets W as the set of word field values that appear in the row with the user ID u.

Step 132) The interest coincidence degree table updating unit 50 refers to the interest information table 93 for each word w included in the word set W, and stores the feature amount field in the row where the user ID is u and the word is w. Get the value interest ₀ (u, w).

  Step 133) The interest match degree table updating unit 50 refers to the program information table 91, and acquires a program set P in which the broadcast time field value is later than the current time.

  Step 134) The interest coincidence degree table updating unit 50 calculates the statistic of each word w in the word set W in the content field of each program p in the program set P (for example, “TF value” described in Document 3) tf Calculate (p, w). If the word w is not included in the body field of program p, tf (p, w) = 0.

Step 135) The interest coincidence degree table updating unit 50 calculates the interest coincidence degree score ₀ (p, u) for the user u of the program p using the following equation (2).

式（２）における tf(p,w) は、番組 p における単語 w の重みを意味し、興味情報テーブル９３の特徴量 interest₀(u,w)はユーザ u の単語 w に対する興味量を表す。分子は、両者の内積であるが、内積は番組 p に含まれる語数が多ければ多いほど大きくなる傾向を考慮し、分母により正規化する。これは、文献３（Manning, C.D. et al., "Introduction to information retrieval", 2008）記載のコサイン類似度と呼ばれる尺度であり、この尺度によりユーザ u の興味と番組 p の関連の強さが求まる。

In equation (2), tf (p, w) means the weight of the word w in the program p, and the feature amount interest ₀ (u, w) of the interest information table 93 represents the interest amount of the user u with respect to the word w. The numerator is the inner product of the two, and the inner product is normalized by the denominator in consideration of the tendency that the larger the number of words included in the program p, the larger. This is a measure called cosine similarity described in Reference 3 (Manning, CD et al., “Introduction to information retrieval”, 2008). This measure determines the strength of the relationship between user u's interest and program p. .

Step 136) The interest coincidence degree table update unit 50 stores the interest coincidence degree table 95 in the interest coincidence degree field of the row where the user ID is u and the program ID is p. The interest coincidence score ₀ (p, u) Update to be

    Next, processing at the time of program recording will be described.

  FIG. 11 is a flowchart of processing at the time of program recording in one embodiment of the present invention.

  Step 210) The viewing information table updating unit 10 sets a program ID field and a user ID field in the viewing information table 92 according to the viewed program and the viewed user. The viewing / recording type field is set to “recording”. Such a row r is inserted into the viewing information table 92.

  Step 220) The good / interest information table update unit 50 updates the good / interest information table 94 based on the information in the row r.

  Step 230) The good interest coincidence degree table update unit 50 updates the good interest coincidence degree table 96 based on the information of the row r.

  Next, the processing of step 220 in FIG. 11 will be described in detail.

  FIG. 12 is a flowchart of detailed processing of step 220 in FIG. 11 according to the embodiment of this invention. Assume that the user in row r and the target programs are u and p, respectively.

  Step 221) The program information acquisition unit 10 refers to the program information table 91 and acquires the value of the content field of the row where the program ID is p.

  Step 222) The good interest information table update unit 40 acquires the word set W included in the contents of the program p by using the division writing function of the morphological analyzer.

  Step 223) The good interest information table updating unit 40 performs, for each word w included in the word set W, a statistic in the content of the program p, for example, an appearance frequency (Document 3 (Manning, CD et al., “ “TF value”) described in “Introduction to information retrieval”, 2008)) tf (p, w) is calculated. In calculating the weight in the content of the program p of the word w, not only the appearance frequency but also another scale (such as “TFIDF value” described in Document 3) may be used.

Step 224) The good interest information table update unit 40 refers to the good interest information table 94, and the value of the feature field in the row where the value of the user ID field is u and the value of the word field is w interest ₁ (u, w) To get. However, if there is no row with the user ID field value u and the word field value w, the interest ₁ (u, w) value is 0.

Step 225) The good interest information table update unit 40 sets the value of the feature amount field in the good interest information table 94 in the row where the value of the user ID field is u and the value of the word field is w interest ₁ (u, w) + Update or insert to be tf (p, w).

  Next, the processing of step 230 in FIG. 11 will be described in detail.

  FIG. 13 is a flowchart of detailed processing of step 230 in FIG. 11 according to the embodiment of this invention. Suppose now that each user on line r is u.

  Step 231) The good interest coincidence degree table updating unit 60 refers to the good interest information table 94, and designates a set of word field values appearing in the row with the user ID u as W.

Step 232) The good interest coincidence degree table updating unit 60 refers to the good interest information table 94 for each word w included in the word set W, and the feature amount of the row where the user ID is u and the word is w Get the field value interest ₁ (u, w).

  Step 233) The good interest coincidence degree table update unit 60 refers to the program information table 91, and acquires a program set P in which the value of the broadcast time field is later than the current time.

  Step 234) The good interest coincidence degree table updating unit 60 calculates the statistics of each word w in the word set W in the content field of each program p in the program set P (for example, “TF value” described in Document 3). tf (p, w) is calculated. If the word w is not included in the body field of program p, tf (p, w) = 0.

Step 235) The good interest match degree table update unit 60 calculates the interest match score ₁ (p, u) for the user u of the program p using the following equation (3).

上記の式（３）における tf(p,w) は、番組 p における単語 w の重みを意味し、良興味情報テーブル９４の特徴量 interest₁(u,w)はユーザ u の単語 w に対する良興味量を表す。分子は、両者の内積であるが、内積は番組 p に含まれる語数が多ければ多いほど大きくなる傾向を考慮し、分母により正規化する。これは、文献３（Manning, C.D. et al., "Introduction to information retrieval", 2008）記載のコサイン類似度と呼ばれる尺度であり、この尺度によりユーザ u の良興味と番組 p の関連の強さが求まる。

In the above equation (3), tf (p, w) means the weight of the word w in the program p, and the feature amount interest ₁ (u, w) in the good interest information table 94 is the good interest of the user u for the word w. Represents an amount. The numerator is the inner product of the two, and the inner product is normalized by the denominator in consideration of the tendency that the larger the number of words included in the program p, the larger. This is a measure called cosine similarity described in Reference 3 (Manning, CD et al., “Introduction to information retrieval”, 2008). This measure shows the strength of the relationship between the good interest of the user u and the program p. I want.

Step 236) The good interest match degree table update unit 60 stores the good interest match degree table 96, and the value of the good interest match degree field in the row where the user ID is u and the program ID is p is the good interest match degree score ₁ ( Update to be p, u).

  FIG. 14 is a flowchart of processing at the time of user logon according to an embodiment of the present invention.

When user u logs on,
Step 310) The program recommendation unit 70 refers to the interest match degree table 95 and obtains the row I ⁰ _u whose user ID field is u.

Step 320) The program recommendation unit 70 refers to the good interest match degree table 96, and obtains the row I ¹ _u whose user ID field is u.

Step 330) The program recommendation unit 70 refers to the weight parameter recording unit 97 and acquires the weight parameters λ ₀ and λ ₁ .

  Step 340) The program recommendation unit 70 refers to the program information table 91, and acquires a program set P in which the broadcast time field value is later than the current time.

Step 350) The program recommendation unit 70, for each program p in the program set P, the interest match score of the row with the program ID p in the user ID I ⁰ _{u in} the interest match table 95. ₀ (p, u) and the good interest match score score ₁ (p, u) of the line with the program ID p of the user ID I ¹ _{u in} the good interest match table 96 is acquired. Then, the recommendation score score (p, u) of the program p to the user u is calculated using the equation (1).

  Step 360) The program recommendation unit 70 determines a program to be recommended to the user u using the value of the recommendation score score (p, u) for the user u of each program p in the previous program set P. At that time, an arbitrary number N (N = 1, 2, 3,...) That is set in advance in descending order of the value of the recommended score field may be set as a program to be recommended, and a certain threshold is set for the recommended score. A program that has a recommendation score exceeding that value may be recommended.

  Next, weight parameter update processing by the weight parameter update unit 80 will be described.

  FIG. 16 is a flowchart of processing when updating the weight parameter according to the embodiment of the present invention. The weight parameter may be automatically calculated using a logistic regression model described in Reference 4 (Hosmer, DW et al., “Applied logistic regression”, 2000) as in the present embodiment, or Reference 5 ( Cristianini, N. et al., "An introduction to support Vector Machines: and other kernel-based learning methods", 2000) A weight parameter may be set. The update of the weight parameter may be started by an explicit input from the system administrator, or may be periodically batch processed.

  Step 410) The weight parameter updating unit 80 acquires the user set U appearing in the user ID field of the viewing information table 92.

  Step 420) The weight parameter updating unit 80 refers to the program information table 91 and acquires a program set P in which the broadcast time field value is earlier than the current time.

Step 430) The weight parameter updating unit 80 refers to the interest match degree table 95 and the good interest match degree table 96, and combines each question of the program set P with each question of each user of the user set U ( For p, u), the interest match score ₀ (p, u) and the good interest match score ₁ (p, u) are obtained, and l (p, u) is calculated using the following equation.

上記の式（４）は、Σ_iλ_iscore_i(p,u) の値を確率化したもので、Σ_iλ_iscore_i(p,u) の値が高ければ高いほど値が１に近づき、Σ_iλ_iscore_i(p,u) の値が低ければ低いほど０に近づく。

Equation (4) _{is, Σ i λ i score i (} p, u) obtained by the probability of the value _{of, Σ i λ i score i (} p, u) the greater the value is 1 higher the value of The closer Σ _i λ _i score _i (p, u) is, the closer it is to 0.

Step 440) The weight parameter update unit 80 has a user ID u and a program ID p in the viewing information table 92 for each program p in the program set P and each user u in the user set U. When there is a row where the value of the viewing / recording field is “recording” b (p, u) = α, the user ID is u, the program ID is p, and the value of the viewing / recording field is “recording” Does not exist, but b (p, u) = β if there is a “view” line, and b (p, u) = 0 if there is no line with user ID u and program ID p And At this time, L (λ ₀ , λ ₁ ) is calculated using the following equation.

α、βは人手によって、例えば 0.8、0.5 に設定する。

α and β are manually set to 0.8 and 0.5, for example.

The above equation (5) is a measure called the likelihood of λ ₀ and λ ₁ , and whether u has recorded or watched p in the past b (p, u) and the equation for each program user ( This is a scale that represents how much the set λ ₀ and λ ₁ match the past viewing information from l (p, u) represented by 4). Therefore, λ ₀ and λ ₁ that maximize Equation (5) are considered to be the most appropriate weight parameters.

Step 450) The weight parameter updating unit 80 uses the quasi-Newton method or the like described in Reference 6 (Liu, DC et al., “On the limited memory BFGS method for large scale optimization”, 1989). L (lambda ₀ represented by), lambda ₁₎ to maximize lambda _0, calculates the lambda _1.

Step 460) The weight parameter updating unit 80 records the weight parameters λ ₀ and λ ₁ calculated earlier in the weight parameter recording unit 97.

  It should be noted that the processing of each component of the program recommendation device can be constructed as a program and installed in a computer used as the program recommendation device for execution, or distributed via a network.

  In addition, the constructed program can be stored in a portable storage medium such as a hard disk, a flexible disk, or a CD-ROM, and can be installed or distributed in a computer.

  The present invention is not limited to the above-described embodiment, and various modifications and applications can be made within the scope of the claims.

DESCRIPTION OF SYMBOLS 1 Program recommendation apparatus 2 External terminal 3 Network 4 External apparatus A0 Communication part B0 Input / output part 10 Program information acquisition part 20 Viewing information table update part 30 Interest information table update part 40 Good interest information table update part 50 Interest coincidence degree table update part 60 good interest match level table update unit 70 program recommendation unit 80 weight parameter update unit 90 storage unit 91 program information table 92 viewing information table 93 interest information table 94 good interest information table 95 interest match level table 96 good interest match level table 97 weight Parameter storage

Claims (7)

In a video distribution service using an Internet line, a program recommendation device that uses information about whether a user has recorded a viewing history, particularly whether a program has been recorded by a user,
Program information acquisition means for acquiring program information in the video distribution service using the Internet line and storing the program information in the program information storage means;
Word extraction means for extracting words from summary text of contents in the program information;
A statistic calculating means for calculating a statistic of each word in each program information and storing it in a statistic storage means;
Viewing information acquisition means for acquiring viewing information of viewing or recording of each user in the video distribution service using the Internet line, and storing the viewing information in the viewing information storage means;
From the acquired viewing information, the user's interest in the word from the statistics of each word calculated from the summary text of the content stored in the program information storage means of the program that is the target of the viewing information An interest amount calculating means for calculating an amount, and storing the user identifier in the interest amount storage means together with the word;
From the statistics for each word of each user stored in the amount-of-interest storage means and the statistics for each word of only the pre-broadcast program among the programs stored in the program information storage means, the user Interest matching degree calculation means for calculating the interest matching degree for the question and storing the user identifier and the program identifier together with the interest matching degree storage means;
Of the programs in the program information storage means, referring to the viewing information storage means, only recorded programs are selected, and each of the programs to be recorded stored in the statistics storage means A good interest amount calculating means for calculating a good interest amount for the user from the statistic of the word, and storing the user identifier in the good interest amount storage means together with the word;
From the statistics for each word of each user stored in the good interest amount storage means and the statistics for each word of only the pre-broadcast program among the programs stored in the program information storage means, A good interest match degree calculating means for calculating a good interest match degree for the program of the user, and storing the user identifier in the good interest match degree storage means together with the identifier of the program;
Of all the programs stored in the program information storage means, the program information for all combinations of broadcasted programs and programs that have been viewed among the users stored in the program information storage means. Whether or not the user has recorded the program stored in the storage means, the interest match degree stored in the interest match degree storage means, and the good interest match stored in the good interest match degree storage means A weight parameter calculating means for calculating a weight parameter using the degree and storing the weight parameter in the weight parameter storage means;
Given a user identifier, for each pre-broadcast program stored in the program information storage means of the user, the interest match degree stored in the interest match degree storage means and the good interest match The recommendation score for each program of the user is calculated using the good interest match degree stored in the degree storage means and the weight parameter stored in the weight parameter storage means, and the top N of the recommended scores are calculated. Program recommendation means for selecting an identifier of a question as a recommended program;
A program recommendation device characterized by comprising: The statistic calculation means includes:
2. A TF value (word appearance frequency) and a TFIDF value (word appearance frequency and reverse appearance frequency) for extracting a word regarded as important are used when calculating the statistic of each word. The program recommendation device described. The weight parameter calculating means includes
Whether the user has recorded the program stored in the program information storage means, the interest match degree stored in the interest match degree storage means, and the good interest match degree storage means. The program recommendation device according to claim 1, further comprising means for calculating a weight parameter by logistic regression analysis based on the degree of good interest match. In a video distribution service using an Internet line, a program recommendation method that uses information about whether a user has recorded a viewing history, particularly whether a program has been recorded by a user,
Program information storage means, statistics amount storage means, viewing information storage means, interest amount storage means, interest match degree storage means, good interest amount storage means, good interest match degree storage means, weight parameter storage means are stored in the storage medium. Program information acquisition means, word extraction means, viewing information acquisition means, interest amount calculation means, interest match degree calculation means, good interest amount calculation means, good interest match degree calculation means, weight parameter calculation means, program recommendation means, In a device having
The program information acquisition unit acquires program information in a video distribution service using the Internet line, and stores the program information in the program information storage unit;
A word extracting step in which the word extracting means extracts a word from the summary text of the content in the program information;
A statistic calculating unit calculates a statistic of each word in each program information and stores it in the statistic storage unit;
The viewing information acquisition unit acquires viewing information of viewing or recording of each user in the video distribution service using the Internet line, and stores the viewing information in the viewing information storage unit;
From the statistics of each word calculated from the summary text of the content stored in the program information storage means of the user and the program that is the target of the viewing information, the interest amount calculating means from the acquired viewing information, An interest amount calculation step of calculating an interest amount of the user for the word, and storing the user identifier in the interest amount storage unit together with the word;
The interest match degree calculating means includes a statistical quantity for each word of each user stored in the interest amount storage means, and each of the programs stored in the program information storage means for only the program before broadcasting. An interest matching degree calculating step of calculating an interest matching degree with respect to the question of the user from a statistic of a word, and storing the user's identifier and the interest matching degree storage unit together with the identifier of the program;
The amount-of-interest calculation means selects only the recorded programs from among the programs stored in the program information storage means with reference to the viewing information storage means, and stores the recorded programs stored in the statistics storage means A good interest amount calculation step of calculating a good interest amount for the word of the user from the statistics of the word as a target of the program and storing the user's identifier and the word together with the good interest amount storage unit When,
The good interest coincidence calculation means calculates only the pre-broadcast programs among the statistics for each word of each user stored in the good interest quantity storage means and the programs stored in the program information storage means. A good interest match degree calculating step of calculating a good interest match degree for the program from the statistic of each word, and storing the user's identifier in the previous good interest match degree storage means together with the identifier of the program; ,
Of the programs stored in the program information storage unit, the weight parameter calculation unit includes all broadcasted programs and programs that have been viewed among the users stored in the program information storage unit. Whether the user has recorded the program stored in the program information storage means, the interest match degree stored in the interest match degree storage means, and the good interest match degree storage means A weight parameter calculating step of calculating a weight parameter using the stored good interest match degree and storing it in the weight parameter storage means;
When the program recommendation means is given a user identifier, the interest match degree stored in the interest match degree storage means for each pre-broadcast program stored in the program information storage means of the user Using the good interest match degree stored in the good interest match degree storage means and the weight parameter stored in the weight parameter storage means, the recommendation score for each program of the user is calculated, and the recommendation A program recommendation step of selecting identifiers of the top N questions in the score as recommended programs;
A program recommendation method characterized by: In the statistic calculation step,
5. When calculating the statistic of each word, a TF value (word appearance frequency) and a TFIDF value (word appearance frequency and reverse appearance frequency) for extracting a word regarded as important are used. The program recommendation method described. In the weight parameter calculating step,
Whether the user has recorded the program stored in the program information storage means, the interest match degree stored in the interest match degree storage means, and the good interest match degree storage means. The program recommendation method according to claim 4, wherein the weight parameter is calculated by logistic regression analysis based on the degree of good interest match.   The program recommendation program for functioning a system as each means which comprises the program recommendation apparatus of any one of Claim 1 thru | or 3.

Images