WO2004114155A1

WO2004114155A1 - Content recommending device, method, and program

Info

Publication number: WO2004114155A1
Application number: PCT/JP2003/007860
Authority: WO
Inventors: Seishi Okamoto; Hiroya Inakoshi; Akira Sato; Takahisa Ando
Original assignee: Fujitsu Limited
Priority date: 2003-06-20
Filing date: 2003-06-20
Publication date: 2004-12-29

Abstract

An input section receives information on an object user to whom a content is recommended. A content recommendation candidate determining section searches for contents selected in the past by similar users similar to the object user entered, and determines content recommendation candidates recommended to the object user. A search section searches a case database for the contents selected in the past by the object user. A score calculating section calculates the score representing the similarity of each content of the content recommendation candidates with each content selected by the object user. A recommendation content determining section determines the content to be recommended to the object user according to the scores of the content recommendation candidates.

Description

Content recommendation device, method and program

The present invention relates to a content recommendation device, method, and program for recommending useful content to a user, and more particularly to a content recommendation device, method, and program using profile data and user case data, which characterize the content.

Content recommendation device, method, and program for recommending content. Rice field

Background art

A shopping site on the Internet provides a service that recommends useful content to customers who have a history of purchasing content based on their purchase history. This service saves customers from having to search for the content that suits them from a vast amount of content, and can lead to sales expansion for shopping site operators.

Conventionally, as a method of recommending useful content for each user, for example, a profile data is created by characterizing the user and the content, and a content recommendation is requested based on the created profile data. A method of searching for a user similar to the target user and recommending content that has been frequently browsed or purchased by the similar user and content highly evaluated by the similar user has been adopted (Patent Document 1).

According to this content recommendation method, users who browse or purchase the same content as the target user or users who perform similar evaluations are set as similar users, and similar users often browse. It is possible to recommend useful content for the target user by recommending content that has been highly evaluated by similar users who are often purchased.

(Patent Document 1)

Japanese Patent Application Laid-Open No. 2000- 1 4 8 8 6 4

However, according to such a conventional content recommendation method, in a group of users similar to the target user, the content having a high content evaluation with a high number of browsing and purchase times may be recommended to the target user, and the target user The problem may be that useful content is not always recommended to the public. This near point Tsu is to be recommended to the target user, which is determined by whether it is useful for similar users occurs because the usefulness to the target user is not reflected directly _(e.g., similar user to the target user If 10 people are selected and 9 of them have high reputation and one of them has high reputation, nine similar users give high reputation to the target user. However, for the target user, the latter content, in which one similar user gives a high rating, may be useful.

The present invention provides a content recommendation device, a method, and a method for recommending content that is truly useful to a target user based on the similarity with the target user's case, using content that has been selected by a similar user as a content recommendation candidate. The purpose is to provide the program. Disclosure of the invention

The present invention is based on a user database in which characteristics of a plurality of users are recorded and a profile in which characteristics of a target user are recorded from an example database in which examples of contents selected by the users are recorded. A computer that recommends the best content for the target user In the content recommendation device, a similar user search unit that searches for a similar user similar to the profile from the user database, a content recommendation candidate determination unit that sets content selected by the similar user in the past as a content recommendation candidate from the case database, A search unit for searching for a content previously selected by a specific user recorded in a case database; a score calculation unit for calculating each similarity between a content recommendation candidate and a previously selected content; and a target user based on each similarity. And a recommended content determining unit that determines content to be recommended.

Therefore, according to the present invention, a content recommendation candidate determined based on a selection history example such as past browsing or purchase of a similar user similar to the target user, and a history of past selections of the target user. Since the recommended content is determined by calculating the similarity from the case, if the content is highly similar to the selection history of the target user, even if there is only one similar user who selected the content, The content can be recommended.

Here, the content recommendation candidate determination unit is characterized in that the content selected by the target user in the past is not a recommendation candidate. Also, the content recommendation candidate determination unit deletes the content that the target user has already selected from the determined content recommendation candidates. This prevents the content already selected by the target user from being unnecessarily recommended.

The recommended content determination unit recommends a predetermined number of profile contents in the descending order of the score from the content recommendation candidates. In this case, the content that the target user has frequently viewed or purchased in the past will be recommended as useful content for the user. Recommend the content of a predetermined number of profiles. Score If you recommend high content, you may be more likely to recommend only the same content. Even if the content has a low score, it is considered to be useful content for the target user because it is selected by similar users.

The recommended content determination unit may recommend the content of a predetermined number of profiles in descending order of the score and the content of a constant number of profiles in descending order of the score from the content recommendation candidates. The case database stores a case in which a specific user previously added the content selected by the user to the evaluation value given by the target user, and the score calculation unit sets the content recommendation candidate similar to the content having a high evaluation value to: The score for each content recommendation candidate is calculated using the evaluation value such that the score of the content recommendation candidate similar to the case content having a high similarity and a low evaluation value is low.

The score calculation unit calculates, for each content recommendation candidate, the total sum of similarities with a plurality of case contents as a score. The score calculation unit calculates, for each content recommendation candidate, an average of the similarities with a plurality of case contents as a score. The score calculation unit calculates the similarity to the case content for each content recommendation candidate, and obtains the maximum value or the minimum value among the calculated similarities as the score.

The present invention provides a content recommendation method. That is, the present invention provides a user database that records the characteristics of each of a plurality of users, and a case database that records cases of contents selected by the users, based on a profile that records the characteristics of the target user. A content recommendation method for recommending an optimal content for the target user,

A similar user search step for searching for a similar user similar to the profile from the user database;

From the case database, the content selected by a similar user in the past A content recommendation candidate determining step in which the content recommendation candidate is used as a content recommendation candidate; a search step for searching for the content selected by the specific user in the past recorded in the case database;

A score calculation step for calculating each similarity between the content recommendation candidate and the content selected in the past;

A recommended content determination step for determining content to be recommended to the target user from each similarity;

It is characterized by having.

The present invention provides a program for content recommendation. This program is based on a user data base that records the characteristics of each of a plurality of users, and a profile that records the characteristics of the target user from a case database that records cases of the content selected by the user. In the content recommendation device computer that recommends the optimal content for the target user,

A similar user search step for searching similar users similar to the profile from the user database;

A content recommendation candidate determining step of selecting a content selected by a similar user in the past from the case database as a content recommendation candidate; a search step of searching for a content previously selected by a specific user recorded in the case database;

A score calculation step for finding each similarity between the content recommendation candidate and the content selected in the past;

A recommended content determination step of determining content recommended to the target user from each similarity;

Is executed.

The details of the content recommendation method and the program in the present invention are basically the same as those of the content recommendation device. BRIEF DESCRIPTION OF THE FIGURES

Figure 1 is an explanatory diagram of the network environment to which the present invention is applied;

FIG. 2 is a block diagram of a functional configuration of the content recommendation device according to the present invention; FIG. 3 is an explanatory diagram of user profile data;

Figure 4 is an illustration of profile data characterized by content; Figure 5 is an illustration of purchase and listing example data;

Fig. 6 is an explanatory diagram of the processing content by the content candidate determining unit of Fig. 2; Fig. 7 is an explanatory diagram of the processing content of the search unit, score calculation unit, and recommended content determining unit of Fig. 2;

FIG. 8 is a flowchart of the content recommendation processing of the present invention using the functional configuration of FIG. 2;

Figure 9 is an explanatory diagram of the evaluation matrix given by the user to the content;

Figure 10 is an explanatory diagram of the evaluation matrix given by the similar user to the target user;

Figure 11 is an explanatory diagram of the evaluation matrix given by the target user;

Figure 12 is an explanatory diagram of the profile list corresponding to the content to which the target user has given an evaluation;

Figure 13 is an illustration of the profile list corresponding to the top two content recommendation candidates for similar users of Figure 10;

Fig. 14 is an explanatory diagram of importance, which is a profile attribute value calculated based on the content to which the target user has given an evaluation in the past; BEST MODE FOR CARRYING OUT THE INVENTION

FIG. 1 is an explanatory diagram of a network environment to which content recommendation processing according to the present invention is applied. Here, an example of content recommendation at a book sales site will be described. In FIG. 1, the content recommendation processing of the present invention is executed by the server 10. For server 10 The user devices 14-1 and 14-2 are connected via the evening network 12. The user devices 14-1 and 14-2 are equipped with WWW browsers, and perform content recommendation requests by specifying user information by accessing the server 10 via the Internet 12. The user database 16, the case database 18, and the profile database 20 are connected to the server 10. These databases may be data files such as XML files. The server 10 includes a CPU 22, a memory 24 and a cache 26 as a hardware configuration. The memory 24 is loaded with a program for executing the content recommendation process of the present invention, and the CPU 22 executes the program to execute the content recommendation process.

FIG. 2 is a block diagram of a functional configuration of a content recommendation process according to the present invention executed by the server 10 of FIG. 2 and FIG. 1. In FIG. 2, a server 10 functioning as a content recommendation processing device includes a user information input unit 28, a content recommendation candidate determination unit 30, a search unit 32, a score calculation unit 34, a recommended content determination unit 36, An output section 38 is provided. For the server 10, a user database 16, a case database 18, and a profile database 20 are provided. The user database 16 stores, for example, user data that characterizes users who use the content recommendation processing of the present invention as registered users. The case database 18 stores the history of the user browsing or purchasing the content as case data. The profile database 20 stores a profile database that characterizes the content of a web page, a product, and the like as a set of pairs of attributes and attribute values or a set of keywords. The user information input unit 28 inputs information of a target user who is a content recommendation target in response to a user recommendation request input regarding content transmitted from the user devices 141-1 and 14-12 in FIG. . Recommended content The supplementary decision unit 30 searches the case database 18 for contents previously browsed or purchased by a similar user similar to the target user input in the user information input unit 28, and based on the search result, the target user is searched. Determine content recommendation candidates. In this case, a similar user similar to the target user is searched using the user database 16. The search unit 32 searches the case database 18 for the case contents browsed or purchased by the target user in the past, and searches the profile corresponding to the searched case contents from the profile file base 20. The score calculation unit 34 calculates the similarity of the target content searched by the search unit 32 to the profile of the example content for each content recommendation candidate profile determined by the content recommendation candidate determination unit 30. Is calculated. The recommended content determination unit 36 determines the content to be recommended to the target user based on the score for each content recommendation candidate calculated by the score calculation unit 34. In determining the recommended content in this case, basically, the specified number of contents are determined and recommended in descending order of the score, but in addition to this, the specified number of contents are determined in descending order of the score and recommended to the target user. For example, if the product is a book, if the target user has purchased only the book of the author A, if the recommended content is selected in descending order of the score, only the book of the author A may be recommended. By daringly recommending content with a low score, it is possible to recommend books other than the author A selected by similar users. The output unit 38 has input the recommendation request for the content determined by the recommended content determination unit 36. ■ The output unit 38 transfers the content to the user device.

FIG. 3 is an explanatory diagram of the user profile data 40 stored in the user database 16 of FIG. The user profile data 40 is composed of a set of attribute-attribute-value pairs for the user following the user ID. In other words, the attributes are age, gender, There are occupations, marriages, number of children, and addresses, and user-specific attribute values are stored for each.

FIG. 4 is an explanatory diagram of profile data stored in the profile database 20 of FIG. In the evening, the profile data used in the present invention includes the profile data 42 shown in FIG. 4 (A) and the profile data 44 shown in FIG. 4 (B). The profile data 42 in Fig. 4 (A) characterizes the content of a web page or product as a set of attribute-attribute-value pairs. In this example, the profile data 40 has a book name, an author name, and a publisher as attributes following the content ID, and stores attribute values corresponding to the attributes. Therefore, this profile data 42 is profile data that characterizes a book as content. In contrast, the profile data 44 in Fig. 4 (B) characterizes contents such as web pages and products as a set of keywords. In the following description, the case where the profile data 42 that characterizes the content by the set of the attribute-attribute-value pairs in Fig. 4 (A) is stored in the profile database 20 is an example. To explain.

FIG. 5 is an explanatory diagram of case data stored in the case database 18 of FIG. FIG. 5 (A) shows basic case data 46, in which the relation type, date / time, user ID, and content ID are set as attributes, and the corresponding attribute values are stored. The relationship type stores the user's action on the content, for example, “buy” or “browse”. In this example, “buy” is registered. In addition, the date and time of the case data 46 can be used when performing a search by deciding the designated period in case search. In the present invention, the user ID can be used as a key for case search by similar users and case search by target users. Furthermore, the content ID can be used when searching for the corresponding profile data 42 shown in FIG. 4 (A). FIG. 5 (B) shows other case data 48 stored in the case database 18. This case data 48 stores information of the evaluation newly given to the content of the user. That is, in this example, for the content ID = 20000, the user with the user ID = 77777 gives the evaluation value “0.8” as the evaluation. In this way, by storing the evaluation value for the user's content in the case data 48, in the content recommendation process of the present invention, the case data of the target user is evaluated based on the evaluation value. Can be narrowed down to specific case data. Also, this evaluation value can be used for calculating the similarity in the score calculation.

Next, referring to FIGS. 6 and 7, the details of the content recommendation processing according to the present invention in the super-processor 10 of FIG. 2 will be described. FIG. 6 is an explanatory diagram of the processing performed by the content recommendation candidate determination unit 30 of FIG. The content recommendation candidate determination unit 30 searches the user database 16 by a similar user search 52 using the recommendation target user 50 obtained by the recommendation request input, Search for similar users 5 4— 1 to 5 4—n. Subsequently, the similar user case search 5 6 is used to search the case database 18 using the user ID of the similar user 5 4—1 to 5 4—n as a search key. Similar user case data 5 8—1 to 5 8 — Search for 11. Subsequently, in content candidate determination 60, content recommendation candidates 60-0 to 60-0-n are determined from similar user case data 58-1-1 to 58-n retrieved from the case database 18. I do. The following methods are used for the content candidate determination 60

(1) Contents that are frequently purchased and viewed

(2) Highly rated content

(3) Content recently purchased by similar users

Among the contents determined according to this determination method, for example, the higher specified number is designated as content recommendation candidates 60-1 to 60-n. In addition, When deciding a recommendation candidate, it is not necessary to recommend the case content of the target user who made the recommendation request because the target user has already purchased or browsed the content, so the case content of the target user is deleted. I do. Subsequently, for each of the content recommendation candidates 6 0— l to 6 0 — n, a profile search 62 2 of the content recommendation candidates is performed, and the profile data 20 4 — 1 to 6 4 — corresponding to the content recommendation candidates is obtained from the profile database 20. Search for n.

FIG. 7 is an explanatory diagram of the processing content of the search unit 32 in FIG. 2 and also briefly shows the processing of the score calculation unit 34 and the recommended content determination unit 36. In FIG. 7, the search unit 32 uses the user ID obtained from the recommendation target user database 50 who made the recommendation request input, and performs a case search 66 of the target user from the case database 18 through the case search 18 of the target user. Case data 6 8—:! Search for 6 8—n. In the case of this target user case search 6 6

(1) Search for the number of designated cases

(2) Search with specified period

(3) Search for cases with evaluation values equal to or greater than the specified threshold

And so on. Then, using the content ID stored in the target user case data 6 8 — 1 to 6 8 — n as a search key, a profile search for case contents 70 from the profile database 20 to the profile file corresponding to the case contents One evening 7 2— Search for 1 to 7 2 -n.

Next, the score calculation unit 34, for each of the profile data 6 4 — 1 to 6 4 — n corresponding to the content recommendation candidates obtained by the content recommendation candidate determination unit 30 of FIG. The score C for each content recommendation candidate 60-1 to 60-0-n is calculated using the profile data 72-1-1 to 72-n corresponding to the case content of the target user obtained in the above. For example, the similarity between profile X and profile y S im (x, y) = S um w (ai) d (xi, yi) (1) where w (ai) is the weight for the profile feature ai

d (x i, y i) is calculated using the distance of X and y to the profile feature a i. Based on the equation (1), the content candidate 60 0— :! determined by the content recommendation candidate determination unit 30 in FIG. All profile data 7 2 1 1 to 7 2 _ n of the example content obtained by the search unit 3 2 in Fig. 7 for each of the profile data 6 4 1 to 6 4 n Degrees are calculated, and the sum is used as the score of each content recommendation candidate. That is, the score of the content recommendation candidate is calculated by the following equation.

Score (C) = SumSim (P, Pj) (2) where P is the profile of content C

Here, the score of the content recommendation candidate is calculated not by the sum of the similarities according to the formula (2) but by the average of the similarities or the highest or lowest value among the similarities. It may be a score.

Further, the profile data 72-1 to 72-n of the case contents searched by the search unit 32 in Fig. 7 are given evaluation information from the target user as shown in the case data 48 in Fig. 5 (B). In this case, based on the evaluation value of this evaluation information, a score calculation is performed such that the similarity increases for content with high evaluation and the score decreases for content with low evaluation. For example, using the evaluation value, Calculate the score of the recommendation candidate.

Score (C) = SumEva1 (Pi) * Sim (P, Pj)

(3) where Eva1 (Pi) is the target user's profile Pi

An evaluation value given to the content of this example Further, as another example of the score calculation 74 in the score calculation section 34, a score calculation based on the weight of the profile feature value may be performed. In this score calculation, the weights W 1, W 2, to Wn of each profile feature value are obtained from the frequency of profile feature values and statistical information in the set of case contents searched by the search unit 32. The sum of the feature value W of the profile of the content recommendation candidate obtained by the content recommendation candidate determination unit 30 is calculated as a score.

Also, the score calculation 74 in the score calculation section 34 includes a case where the evaluation data of the target user is stored in the profile data 72-1-1 to 72-n of the case content searched by the search section 32. In the descending order of the evaluation value, the score for each content recommendation candidate is calculated using only the profile data of the specified number of case contents or the open file data of the case content whose evaluation value is equal to or higher than the specified threshold. You may make it calculate.

Subsequently, the recommended content determination unit 36 determines the content recommended to the target user based on the content recommendation candidate score calculated by the score calculation unit 34 by the content recommendation 76. The determination method in the content recommendation 76 can be, for example, any of the following.

(1) Recommend a specified number of contents in descending order of score.

(2) Recommend a specified number of contents in order from the lowest score. (3) When the specified number of KOs are recommended in descending order of the score, the specified number of KOs are simultaneously recommended in order of decreasing score.

(4) Recommend those who have a score above the threshold.

(5) Recommend content that has a score below the threshold.

(6) Set upper and lower thresholds, and recommend content with a score above the upper threshold and content with a score below the lower threshold.

Such a decision for content recommendation is selected from the content with the higher score or the content with the lower score. FIG. 8 is a flowchart of the content recommendation processing of the present invention having the functional configuration of FIG. 2, and this processing procedure simultaneously shows the contents of a program for the content recommendation processing executed by the server 10. The procedure of the content recommendation process of the present invention is as follows. Step S1: Input information of the recommendation target user who has made the content recommendation request.

Step S2: Search the user database 16 for a user similar to the target user.

Step S3: Search case data of similar users from the case database 18 and determine a content recommendation candidate based on this. At the same time, a profile database corresponding to the content recommendation candidate is searched from the profile database ₂₀ .

Step S4: Delete the content corresponding to the target user's case in the content recommendation candidates.

Step S5: Search case data of the target user from the case database 18 to determine case contents.

Step S6: Search the profile of the case content from the profile database 20. Step S7: Calculate the total sum of similarities between each example content profile and the content recommendation candidate profile as a score.

Step S8: Based on the score, determine the content to be recommended from the content recommendation candidates, and respond to the target user who made the recommendation request.

Next, a specific example of the content recommendation process according to the present invention will be described by taking as an example a case where the content recommendation process is performed using the evaluation value given by the user to the content in the profile data.

FIG. 9 is an explanatory diagram of the evaluation matrix 78 given by the user to the content. This evaluation matrix 78 is created from a case database in which three users A, B, and C store evaluations previously given to each of the contents CI, C2, C3, and C4. Here, the content evaluation in the evaluation matrix 78 is given on a five-point scale from 1 to 5, where 1 is the lowest rating and 5 is the highest rating. For example, user A gives content C 1 the lowest rating, content C 2 and 3 the highest rating, and content C 4 the bad rating. In such an evaluation matrix 78, for example, the similarity between user A and user B is determined by the Pearson correlation function (Communications of ACM, Vol. 40, No. 3, 197 7) The following equation is calculated using the correlation function of.

P C C (A, B) = C o V (A / B) / (V (A) * V (B))

= 0.9.2 3 0 9 3

Where V (A) and V (B) are the standard deviations of the evaluations of users A and B, respectively G ov (A, B) is the covariance of the evaluations of users A and B On the other hand, the similarity between user A and user C is calculated by the following equation.

P C C (A, C) = C o V (A / C) / (V (A) * V (Ο)

=-0.96 0 7 8 4 Here, Pearson's correlation function takes a value of 1 or more and 1 or less, and user A and user B close to 1 are very similar, and user A close to 1 1 And User C are not quite similar. Therefore, the content recommendation processing according to the present invention based on such a Pearson correlation function will be specifically described as follows.

First, in the present invention, as shown in the processing content of the content recommendation candidate determination unit 30 in FIG. 6, similar users are obtained for the target users who receive the content recommendation request, and the content having a high evaluation from the similar users is obtained. Is a content recommendation candidate. This method of selecting a content highly evaluated by similar users as a content recommendation candidate can be referred to as preprocessing of the content recommendation process of the present invention, which is the same content as the conventional content recommendation method.

FIG. 10 is an evaluation matrix 80 showing the evaluations given by the users to the contents of users D, E, and F selected as similar users to the target user Z who receives the content recommendation. In the evaluation matrix 80, the similar users D, E, F, and the contents C11, C12, C13, and C14 are given a five-level evaluation from 1 to 5. Here, if the number of contents to be recommended to the target user Z is specified as 2, the contents C 11 and the contents C 12 in the evaluation matrix 80 are highly evaluated by all similar users, so that the recommendation to the target user Z is made. Will be done. This means that, in the present invention, the content recommendation candidate determination unit 30 in FIG. 6 has determined that the content C 11 and the content C 12 have been determined as content recommendation candidates. I do.

Figure 11 shows the evaluation matrix 82 of the target user Z who receives the recommendation of the content, and shows the past evaluation of the content C 21, C 22 C 23, and C 24 of the target user Z. I have. Further, FIG. 12 shows a profile list 84 corresponding to the content for which the target user Z has been evaluated, as in the evaluation matrix 82 of FIG. The profile list 84 stores contents C 21, C 22, C 23, and C 24 as attributes, and collectively stores the attribute values. That is, in the profile list 84, each (content, genre) pair constitutes profile data for each content. From the evaluation matrix 82 of FIG. 11 and the profile list 84 of the content of FIG. 12, it can be seen that the target user Z has a high evaluation of the content of the genre “mystery”. Here, assuming that the profiles of the content recommendation candidates C 11 and C 12 determined by the evaluation matrix 80 of FIG. 10 are the profile list 86 of FIG. 13, the similar user of FIG. Assuming that two highly-reputed contents C ll and C 12 are recommended to the target user Z, in this content recommendation, the social science content C 1, which is a genre with a low reputation in the target user Z, is used. However, in the present invention, the past evaluation of the target user Z for the content recommendation candidates C 11 and C 12 evaluated by the similar users obtained from FIG. By determining the importance of the attribute values of the profile of the case content profile, and re-evaluating using this importance, it is possible to make the target user Z truly among the content recommendation candidates that have been highly evaluated by similar users. Tetsuyu Of recommended content is realized.

Figure 14 shows the importance list 88 that indicates the importance of each profile attribute value calculated based on the content that the target user Z has evaluated in the past. Here, the importance takes a value of 1 or less. In the present invention, Using the importance obtained in the importance list 88 in FIG. 14, the scores of the content recommendation candidates C11 and C12 based on the evaluation of the similar users in FIG. 10 are recalculated. In other words, the score of content C 11 is 0.1 because the genre is social science, and the score is 1.0 because content C 12 is a mystery. As a result, of the two content recommendation candidates C 11 and C 12, the content C 12 with a high score is determined as the content to be recommended to the target user Z by recalculating the score. For this reason, the target user Z can recommend the attribute value of interest, in this example, the content whose genre is mystery to the target user, and the similar user evaluates it but the target user Z is interested. No profile attribute values, in this example we can exclude recommendations for target users of content whose genre is social science.

In the above embodiment, a book is taken as an example of the content to be recommended.However, the present invention is not limited to this. I can do it.

Although the present invention has been described with reference to the corresponding drawings, the present invention is not limited to these embodiments, and various other modifications may be made without departing from the scope or spirit of the present invention. And changes are possible. In particular, in the present embodiment, the content recommendation at the product sales site has been described. However, the content recommendation for lending a product may be considered. Also, it may be used for content recommendation for distribution of moving images. Furthermore, it may be a content recommendation for product introduction. Further, the present invention is not limited by the numerical values shown in the above embodiments. Industrial applicability

As described above, according to the present invention, content recommendation is received. The content recommendation candidate recommended based on the past browsing and purchases of other users similar to the target user to be viewed or the evaluation given to the user's content, and the past browsing and purchase of the target user, and Uses a history example such as an evaluation given by the user to calculate the cost for each content recommendation candidate, and re-evaluates the content to be recommended, so that the content is more useful for the target user. Can be recommended.

Claims

The scope of the claims

1. From the user database that records the characteristics of each of a plurality of users and the case database that records the cases of the content selected by the user, the target user can be identified based on the profile that records the characteristics of the target user. In the content recommendation device that recommends the most suitable content,

A similar user search unit for searching for a similar user similar to the profile from the user database;

A content recommendation candidate determining unit that uses the content selected by the similar user in the past as a content recommendation candidate from the case database; and a search unit that searches the content selected by the specific user in the past recorded in the case database. When,

A score calculation unit for calculating each similarity between the content recommendation candidate and the previously selected content;

A content recommendation device, comprising: a recommended content determination unit that determines content recommended to the target user from each of the similarities.

2. In the content recommendation device of claim 1,

The candidate deciding unit is characterized in that the user selected in the past by the target user is not a recommended candidate.

3. The content recommendation apparatus according to claim 1, wherein the recommended content determination unit recommends a predetermined number of profiles in descending order of score from the content recommendation candidates.

4. The content recommendation device according to claim 1, wherein the recommendation content determination unit recommends a predetermined number of profile contents from the content recommendation candidates in ascending order of score.

5. In the content recommendation apparatus according to claim 1, the recommendation content determination unit includes a predetermined number of profile contents in a descending order of the score and a predetermined number of opening files in a descending order of the score from the content recommendation candidates. A content recommendation device characterized by recommending the content of the content.

6. In the content recommendation device of claim 1,

The case database stores a case in which the specific user has previously added the evaluation value given by the target user to the selected content, and the score calculation unit stores the content similar to the content having the high evaluation value. The recommendation candidate is characterized in that a score for each content recommendation candidate is calculated using the evaluation value so that the score of the content recommendation candidate similar to the case content having a high similarity and a low evaluation value is low. Content recommendation device.

7. The content recommendation device according to claim 1, wherein the score calculation unit calculates, for each of the content recommendation candidates, a total sum of similarities with a plurality of case contents as a score.

8. The content recommendation apparatus according to claim 1, wherein the score calculation unit calculates, for each of the content recommendation candidates, an average of similarity with a plurality of case contents as a score.

9. In the content recommendation method according to claim 1, the score calculation unit calculates a similarity with the case content for each of the content recommendation candidates, and a maximum value or a minimum value of the calculated similarities. A content recommendation device characterized in that values are obtained as scores.

10. Based on a user database that records the characteristics of each of a plurality of users and a case database that records cases of the content selected by the user, based on a profile that records the characteristics of the target user. In the content recommendation method that recommends the most suitable content for the target user,

A content recommendation candidate determining step in which the content selected by the similar user in the past from the case database is a content recommendation candidate;

A search step for searching for the content previously selected by the specific user recorded in the case database;

A score calculating step of calculating each similarity between the content recommendation candidate and the previously selected content;

A recommended content determining step of determining content recommended to the target user from each similarity;

A content recommendation method comprising:

11. Based on a user database that records the characteristics of each user and a case database that records examples of content selected by the user, based on a profile that records the characteristics of the target user. A computer of a content recommendation device that recommends optimal content for the target user,

A recommended content determining step of determining, from each of the similarities, content to be recommended to the target user;

A program characterized by executing