JP2010140433A - Content recommendation system, content recommendation method and content recommendation program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a content recommendation system recommending, even when the taste of a content user changes with time, a content according to the taste at that time. <P>SOLUTION: When profile information read by a reading means 86 is not stored in a profile storage means 81 as the profile information of the user, a profile registration means 88 registers the profile information and interest level of the user in the profile storage means 81 by use of a relevancy read by the reading means 86. Accordingly, the taste of the user in recommendation is estimated to recommend a content without explicit update of the profile information on a system. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a content recommendation system, a content recommendation method, and a content recommendation program for recommending content according to a user's preference.

  Various content recommendation systems and content recommendation methods for recommending content suitable for users have been proposed.

  Patent Document 1 describes an information presentation support system that enables information filtering that accurately reflects user preferences. As shown in FIG. 11, the information presentation support system described in Patent Document 1 includes a user registration unit 71, a history information management unit 72, a popular article extraction unit 73, a preference extraction unit 74, and a filtering unit 75. The article DB 76 and the article presentation unit 77 are provided. Among them, the preference extraction unit 74 uses the preference information set in the user registration unit 71 and the browsing history for each user generated by the history information management unit 72, for each user, for each user group, or for all users. Preference information is extracted. The filtering unit 75 filters the articles stored in the article DB 76 using the preference information extracted by the preference extraction unit 74. The popular article extraction unit 73 ranks the articles in the order of the number of accesses based on the history information for each user stored in the history information management unit 72. Further, the article presentation unit 77 presents the processing result of the filtering unit and the processing result of the popular article extraction unit 73 together to the user.

  The preference extraction unit 74 uses an article vector composed of a plurality of vector components in which the types of keywords (words) appearing in the article or article group corresponding to the user's log are the number of dimensions, and the frequency of appearance of each keyword is weighted. Is generated. The preference extraction unit 74 dimensionally compresses the generated article vector by principal component analysis. Then, the preference extraction unit 74 calculates the eigenvector of the correlation matrix of the dimension-compressed article vector, performs KL analysis by Karhunen-Loeve (hereinafter referred to as KL) conversion, and obtains the vector space obtained as a result. Extracted as personal preference information of the user.

  Patent Document 2 describes a document management apparatus that enables efficient browsing of a document that matches a changing user's preference. In the document management apparatus described in Patent Document 2, when a document is registered in the document management apparatus, the document analysis unit analyzes the contents of the document to create a document summary, and obtains document characteristics from the document summary. . The contents of the user information storage unit are updated with the document characteristics obtained by the document analysis unit as user preference characteristics.

  Patent Document 3 describes an information providing method that can provide optimal information following a user's interest / preference even if it changes with time. The information providing method described in Patent Document 3 calculates the degree of similarity between a requested document and each document in a document database storing a plurality of documents when a document presentation request is received from a viewer. To do. The next document to be presented to the viewer is determined based on the calculated similarity and the browsing history of each past document of the viewer.

Japanese Unexamined Patent Publication No. 11-15840 (paragraphs 0026 to 0033, FIG. 1) JP 2008-186058 A (paragraphs 0023 to 0026, FIGS. 1, 4, and 5) JP 2003-167907 A (paragraph 0006)

  The information presentation support system described in Patent Literature 1 filters information using initial profile setting data (initial preference setting information) set in the user registration unit 71. Therefore, when the user's preference changes with the passage of time, the initial information is used unless the user explicitly updates. Therefore, there is a problem that a recommendation result suitable for the user's preference at the time of recommendation cannot be obtained.

  For example, consider a case where the user's preference changes with time, and the user becomes interested in information related to a preference field that is not set as initial preference setting information. At this time, even if the information presentation support system reconstructs the association between the preference information and the keyword using the browsing history, the information presentation support system cannot generate the association between the preference information item and the keyword that are not set in the system.

  The document management method described in Patent Document 2 distributes a document by analogizing the user's preference characteristics only from the document registered by the user. Since the user cannot correct the taste characteristic estimated by analogy, the user cannot perform document distribution in which the user's preference is reflected in the taste characteristic. That is, there is a problem that the document is distributed based only on the preference that does not reflect the user's preference characteristic.

  Also, the information providing method described in Patent Document 3 calculates the similarity between a document browsed in the past and a presentation candidate document when the user browses the document. The amount of calculation of similarity is proportional to the product of the number of documents browsed in the past and the number of candidate documents. Therefore, if the number of target documents (that is, documents viewed in the past and candidate documents) increases, The calculation time of the degree may not be within a realistic range.

  In addition, since the information providing method described in Patent Document 3 is based on the premise that there is information on a document browsed by the user in the past, in the case of a new user, an appropriate document should be provided to the user. There is a problem that cannot be done.

  Accordingly, the present invention provides a content recommendation system, a content recommendation method, and a content recommendation program that can recommend content according to the preference at that time even when the content user's preference changes with time. With the goal.

  A content recommendation system according to the present invention includes a profile storage unit that stores, for each user, a profile item that indicates a user's preference field, and an interest level that indicates the degree of interest of the user with respect to the profile item, and user content browsing. Browsing history storage means for storing history, content feature word related storage means for storing feature word identification information that is identification information of feature words characterizing content for each content, and degree of association between profile items and feature words Is stored in the profile feature word relationship storage means for storing each of the profile items and feature word identification information, and in the user's content browsing history and content feature word related storage means stored in the browsing history storage means. Content characterized by feature words based on feature word identification information for each content User feature word generation means for generating user feature word related information including the number of browsing times of the viewed user for each user and feature word identification information, and user feature word related information generated by the user feature word generation means Reading means for reading the profile item and the degree of association corresponding to the feature word identification information included in the information from the profile feature word relation storage means, and the profile item read by the reading means is stored in the profile storage means as a user profile item If the relevance level read by the reading means is used, the interest level update means for updating the interest level of the user's profile item stored in the profile storage means, and the profile item read by the reading means is the user's profile item. If it is not stored as a profile item in the profile storage means, The profile registration means for registering the user's profile item and the interest level in the profile storage means, the interest level corresponding to the profile item stored in the profile storage means, and the profile feature word relationship storage means. A feature word interest level calculating means for calculating a feature word interest level that is an interest level for each feature word of the user based on the relevance level of the feature word identification information corresponding to the profile item, and a content feature word related storage Content interest level calculating means for calculating an interest level for each content of the user based on the feature word identification information corresponding to the content stored in the means and the feature word interest level is provided.

  The content recommendation method according to the present invention browses content characterized by a feature word based on a user's content browsing history and feature word identification information that is characteristic word identification information that characterizes the content defined for each content. A user feature word generation step for generating user feature word related information including the number of times the user browses, and a profile item indicating a degree of association between the profile item indicating the user's preference field and the feature word And reading the profile item and the degree of association corresponding to the feature word identification information included in the user feature word related information generated in the user feature word generation step from the profile feature word relation storage means storing for each feature word identification information Step and profile items read in the reading step are profiled as user profile items When stored in the memory means, the interest level update step for updating the interest level for the user's profile item stored in the profile storage means using the relevance level read in the reading step, and the profile read in the reading step If the item is not stored in the profile storage means as the user's profile item, the profile registration means for registering the user's profile item and the degree of interest in the profile storage means using the relevance read in the reading step; Based on the degree of interest corresponding to the profile item stored in the profile storage means and the degree of association between the feature word identification information corresponding to the profile item stored in the profile feature word relationship storage means, Feature word interest level calculating step for calculating the feature word interest level that is an interest level The interest level of each content of a user, characterized in that it comprises a content of interest calculation step of calculating, based on the feature word identification information and characteristic word interest corresponding to the content.

  The content recommendation program according to the present invention is a content characterized by a feature word based on feature word identification information that is identification information of a feature word that characterizes a user's content browsing history and content defined for each content. User feature word generation processing for generating user feature word related information including the number of times the user has browsed, and the degree of association indicating the degree of association between the profile item indicating the user's preference field and the feature word Profile item corresponding to the feature word identification information included in the user feature word related information generated in the user feature word generation step from the profile feature word relation storage means for storing for each profile item and feature word identification information. The reading process to be read and the profile items read by the reading process are profiled as user profile items. Profile stored in the file storage means, using the degree of relevance read in the reading step, the interest level update process for updating the interest level for the user's profile item stored in the profile storage means, and the profile read in the read process Profile registration processing and profile for registering a user's profile item and degree of interest in the profile storage means using the relevance read in the reading step when the item is not stored in the profile storage means as the user's profile item The interest for each feature word of the user based on the degree of interest corresponding to the profile item stored in the storage means and the degree of association between the feature word identification information corresponding to the profile item stored in the profile feature word relation storage means A feature word interest level calculation process for calculating a feature word interest level that is a degree, and The interest for each content use's, characterized in that to execute the content of interest calculation processing for calculating, based on the feature word identification information and characteristic word interest corresponding to the content.

  According to the present invention, even if the content user's preference changes with time, the content can be recommended according to the preference at that time.

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

  FIG. 1 is a block diagram showing an embodiment of a content recommendation system according to the present invention. The content recommendation system in the present invention includes a content recommendation server 1, and the content recommendation server 1 is connected to a content reference unit 3 via a communication network 2. The communication network 2 is, for example, a public network or a closed network such as a LAN (Local Area Network), but the form of the communication network 2 is not particularly limited. The content reference unit 3 requests the content recommendation server 1 to recommend content. The content recommendation server 1 receives identification information (hereinafter referred to as a user identifier) for identifying a user transmitted from the content reference unit 3 and a user's recommendation request, and recommends content that meets the recommendation request. As a result, the content is transmitted to the content reference unit 3. The content reference unit 3 requests the content received as the recommendation result from, for example, a content providing server (not shown) that manages the content, and acquires the content. Note that there may be one content providing server or a plurality of content providing servers. Since the communication network 2 and the content providing server are well known to those skilled in the art and are not directly related to the present invention, detailed description thereof is omitted. Moreover, although the content reference part 3 is implement | achieved by PC (Personal Computer), a portable terminal, etc., the content reference part 3 is not limited to PC or a portable terminal.

  The content recommendation server 1 includes a profile storage unit 11, a browsing history storage unit 12, a content feature word related storage unit 13, a user feature word matching unit 14, a user feature word related storage unit 15, and a profile update unit. 16, a profile feature word relationship storage unit 17, and a content query unit 18.

  The profile storage unit 11 includes identification information (hereinafter referred to as a profile identifier) that identifies an item (hereinafter referred to as a profile item) that represents a user identifier and a field of preference, and a user's interest in the profile item. It is a storage device that stores profile information including a value to be expressed (hereinafter referred to as an interest level). Profile information can be referred to as user preference information. An example of profile information stored in the profile storage unit 11 is shown in FIG. In the profile information shown in FIG. 2, for example, the user identified by the user identifier X is interested in the profile item identified by the profile identifiers p001 and p002, and the degree of interest is 4 and 2, respectively. It shows that.

  For example, the profile information is input from an input unit (not shown) of the content recommendation server 1. For example, the control unit (not shown) of the content recommendation server 1 stores the input profile information in the profile storage unit 11. Remember me.

  The browsing history storage unit 12 can specify a user identifier of a user who has browsed the content, information for identifying the content browsed by the user (hereinafter referred to as a content identifier), and the number of times the content has been browsed. It is a storage device that stores browsing history information including information (hereinafter referred to as browsing number specifying information). The browsing number specifying information may be the number of times the content is browsed, or may be a history of browsing date and time when the content was browsed. When the browsing number specifying information is a history of browsing date and time when browsing the content, the number of browsing times of the content may be calculated by counting the history of browsing date and time. In the following description, the browsing history storage unit 12 will be described by taking as an example a case where the number of times of browsing the content is stored as the browsing count specifying information. The number of times the content has been browsed is referred to as the number of browsing times. The browsing history storage unit 12 may store, as browsing history information, for example, a browsing log of content stored by the content providing server and an access log of a proxy server (not shown) connected to the communication network 2. As the content identifier, the browsing history storage unit 12 may use, for example, a content URI (Uniform Resource Identifier) or a document ID of a document management system or the like. An example of browsing history information stored in the browsing history storage unit 12 is shown in FIG. In the browsing history information shown in FIG. 3, for example, the user identified by the user identifier X is browsing the content identified by the content identifiers c001 and c003, and the browsing count is 4 times and 1 time respectively. It shows that there is.

  The content feature word related storage unit 13 stores, for example, content feature information including a content identifier of content published on the content providing server and a word that characterizes the content (hereinafter referred to as a feature word). It is. A feature word is one or more words or phrases that characterize content. The content feature word-related storage unit 13 stores, as a feature word, a word or phrase extracted from the content using a technique such as morphological analysis or TF-IDF (Term Frequency-Inverse Document Frequency) in association with the content. May be. Or the content feature word related storage part 13 may memorize | store the word and phrase provided by the rule defined based on the content etc. as a feature word. An example of the content feature information stored in the content feature word related storage unit 13 is shown in FIG. The content feature information shown in FIG. 4 indicates that, for example, a word characterizing the content identified by the content identifier c001 is a feature word identified by the feature word identifiers w001 and w002. The content feature information is input from, for example, an input unit (not shown) of the content recommendation server 1. For example, the control unit (not shown) of the content recommendation server 1 stores the input content feature information as a profile storage unit. 11 may be stored. Alternatively, the control unit may periodically generate content feature information based on the rule, and store the generated content feature information in the content feature word storage unit 13.

  Based on the browsing history information stored in the browsing history storage unit 12 and the content feature information stored in the content feature word related storage unit 13, the user feature word matching unit 14 Information indicating how long the content containing the content has been browsed by now (hereinafter referred to as user characteristic word related information) is generated.

  The user feature word related information includes a user identifier of a user who has browsed the content, information for identifying a feature word of the content browsed by the user (hereinafter referred to as a feature word identifier), a user identifier, A numerical value (hereinafter also referred to as frequency) representing the number of appearances of the pair with the feature word identifier (that is, the number of times the user of the content has browsed the content characterized by a certain feature word) and these pieces of information. Including the date and time of generation.

  The user feature word related storage unit 15 is a storage device that stores the user feature word related information generated by the user feature word matching unit 14. The user feature word matching unit 14 registers the generated user feature word related information in the user feature word related storage unit 15. At this time, the user feature word matching unit 14 causes the user feature word storage unit 15 to store the date and time when the combination of the feature word identifier, the feature word identifier, and the number of appearances is generated as the last update date and time. An example of the user feature word related information stored in the user feature word related storage unit 15 is shown in FIG. In the user feature word related information shown in FIG. 5, for example, the frequency that the user identified by the user identifier X browses the content characterized by the feature word identified by the feature word identifier w001 is four times. It shows that. In addition, the date and time when the frequency is calculated and registered in the user feature word related storage unit 15 is “July 25, 2008 00:00:00”.

  The profile feature word related storage unit 17 is a storage device that stores profile feature word related information representing the correlation between profile items and feature words. The profile feature word related information includes a profile identifier that identifies a profile item set by the user, a feature word identifier of the feature word related to the profile item, and a numerical value that represents the strength of the relationship between the profile item and the feature word. (Hereinafter referred to as “relevance”). An example of profile feature word related information stored in the profile feature word related storage unit 17 is shown in FIG. The profile feature word related information shown in FIG. 6 indicates, for example, that the degree of association between the profile item identified by the profile identifier p001 and the feature word identified by the feature word identifiers w001 and w002 is 4 and 2, respectively. .

  The profile update unit 16 updates the profile information stored in the profile storage unit 11 based on the user feature word related information and the profile feature word related information. For example, the profile update unit 16 performs the above-described processing when the user feature word related information stored in the user feature word related storage unit 15 is updated.

  The content query unit 18 receives the user identifier and the user recommendation request transmitted from the content reference unit 3, and transmits the content that matches the recommendation request to the content reference unit 3 as a recommendation result.

  The content query unit 18 acquires the profile information of the user specified by the user identifier from the profile storage unit 11 and selects a feature word associated with the profile item of the profile information from the profile feature word related storage unit 17. Further, the content query unit 18 acquires the content identifier of the content corresponding to the feature word from the content feature word related storage unit 13 and transmits it to the content reference unit 3. At this time, the content query unit 18 interests the user for each content calculated based on the degree of interest of the user stored in the profile storage unit 11 and the degree of association stored in the profile feature word related storage unit 17. The degree is also transmitted to the content reference unit 3. As described above, the content query unit 18 transmits the degree of interest for each content together so that the content recommendation server 1 can recommend the content to the user.

  The user feature word matching unit 14, the profile update unit 16, and the content query unit 18 are realized by the CPU of the content recommendation server 1 that operates according to a program (content recommendation program), for example. For example, the program is stored in storage means (not shown) of the content recommendation server 1, and the CPU reads the program, and the user feature word matching unit 14, the profile update unit 16, and the content query unit 18 according to the program. May operate as

  Next, the operation will be described. FIG. 7 is a flowchart illustrating an example of a process for updating the user feature word related information. In the following description, the profile information shown in FIG. 2 is stored in the profile storage unit 11, the browsing history information shown in FIG. 3 is shown in the browsing history storage unit 12, and the content feature word related storage unit 13 is shown in FIG. Assume that content feature information is stored in advance.

  The user feature word matching unit 14 reads the browsing history information from the browsing history storage unit 12 (step S21), and acquires the content identifier of the content browsed for each user from the browsing history information (step S22). Next, the user feature word matching unit 14 reads the content feature information having the content identifier acquired in step S22 from the content feature word related storage unit 13 (step S23). Then, the user feature word matching unit 14 combines the browsing history information read in step S21 and the content feature information read in step S23 with a content identifier to generate a pair of a user identifier and a feature word identifier. The number of appearances is counted (step S24). That is, the user feature word matching unit 14 counts the number of times the user has browsed the content characterized by a certain feature word for each user and feature word.

  The user feature word matching unit 14 updates the user feature word related information stored in the user feature word related storage unit 15 using the set of the user identifier and the feature word identifier generated in step S24. To do. Specifically, in the user feature word matching unit 14, the user feature word related information including the set of the user identifier and the feature word identifier generated in step S 24 is already stored in the user feature word related storage unit 15. It is determined whether or not (step S25). When the user feature word related information is already stored in the user feature word related storage unit 15 (YES in step S25), the user feature word matching unit 14 is stored in the user feature word related storage unit 15. The number of appearances counted in step S24 is added to the frequency of the user characteristic word related information being updated. Then, the user feature word matching unit 14 updates the last update date and time of the user feature word related storage unit 15 at the time when the update process is performed (hereinafter referred to as update operation execution time) (step S26Y).

  On the other hand, if it is determined in step S25 that the user feature word related information is not stored in the user feature word related storage unit 15 (NO in step S25), the user feature word matching unit 14 generates in step S24. The set including the user identifier, the feature word identifier, the number of appearances, and the update operation execution time is stored as new user feature word related information in the user feature word related storage unit 15 (step S26N).

  Further, the user feature word matching unit 14 performs the profile update process to be described later, so that the user feature word related information newly generated in step S24 (hereinafter referred to as the difference of the user feature word related information). You may notify to the profile update part 16 (step S27). In this case, the profile update unit 16 can perform a profile update process, which will be described later, every time user feature word related information is generated. The difference in the user feature word related information is the user feature word related information generated based on the browsing history information targeted in step S21. That is, the difference between the user feature word related information is the difference between the user feature word related information whose frequency is updated by the user feature word matching unit 14 in step S26Y and the user feature word related information newly generated in step S26N. Both are included.

  For example, the user feature word matching unit 14 may perform the update process when one new browsing history is added to the browsing history storage unit 12. Or the user feature word matching part 14 may implement the said update process regularly like a daily batch. In addition, when the update process of a user characteristic word relevant information is implemented regularly and it does not carry out every time one new browsing history is added to the browsing history memory | storage part 12, the user characteristic word matching part 14 is a user characteristic. The time when the word related information update process is completed may be stored in the user characteristic word related storage unit 15 as the last update date and time.

  FIG. 8 is a flowchart illustrating an example of a process (profile update process) for updating the profile information stored in the profile storage unit 11. The profile update unit 16 may perform the profile update process in response to the update process of the user feature word related information in step S26Y or step S26N. Or the profile update part 16 may implement a profile update process, when the notification of the difference of the user feature word related information is received from the user feature word matching part 14 at step S27. In the following description, a case will be described in which the profile update unit 16 performs a profile update process when receiving a difference in user feature word related information from the user feature word matching unit 14.

  In the following description, it is assumed that the initial value of the profile feature word related information shown in FIG. 6 is stored in the profile feature word related storage unit 17 in advance. Note that. For example, the control unit (not shown) of the content recommendation server 1 uses a known technique such as factor analysis from the profile information, the browsing history information, and the content feature information. The word related information may be derived, and the derivation result may be stored in the profile feature word related storage unit 17. Further, for example, the control unit may appropriately recalculate profile feature word related information from the latest profile information and browsing history information to improve accuracy.

  When the profile update unit 16 receives the difference (hereinafter referred to as difference information) of the user feature word related information from the user feature word matching unit 14, the profile associated with the user identifier included in the difference information. All the information is extracted from the profile storage unit 11 (step S31). Next, the profile update unit 16 uses the profile identifier included in the profile information extracted in step S31 to extract the profile feature word information associated with the profile identifier from the profile feature word related storage unit 17. In this way, the profile update unit 16 acquires the degree of association between each profile identifier and the feature word identifier included in the difference information (step S32). Then, the profile update unit 16 sets a correction value for each of the profile word extracted in step S31, (steps S33S to S33E) and the feature word identifier corresponding to the profile identifier, and for each feature word identifier included in the difference information. The degree of interest in the profile information is corrected by calculating (step S34). The profile update unit 16 updates the interest level stored in the profile storage unit 11 with the corrected interest level.

  For example, the profile update unit 16 distributes the number of appearances included in the difference information received from the user feature word matching unit 14 for each profile identifier according to the degree of association of the feature word identifier acquired in step S <b> 32, thereby increasing the interest level. It may be corrected.

  For example, it is assumed that the difference information (user identifier, feature word identifier, frequency, last update date / time) is (X, w001, 1, 2008/7/25 00:00:00). Further, it is assumed that (p001, w001, 4) and (p002, w001, 1) are stored as file feature word related information (profile identifier, feature word identifier, relevance) in the profile feature word related storage unit 17. . In this case, the profile update unit 16 adds the value distributed to these data including w001 stored in the profile feature word related storage unit 17 to the degree of interest. In the case of the above example, 1 × (4 ÷ 5) = 0.8 is distributed to p001, and 1 × (1 ÷ 5) = 0.2 is distributed to p002, and the degree of interest of the profile storage unit 11 is set as a correction value. to correct.

  Furthermore, the profile update part 16 implements the following processes as step S35. First, the profile update unit 16 sets a profile identifier corresponding to the user identifier included in the difference information received from the user feature word matching unit 14 (hereinafter referred to as a profile identifier in the profile storage unit 11). Extract from Next, the profile update unit 16 sets a profile identifier corresponding to the feature word identifier included in the difference information received from the user feature word matching unit 14 (hereinafter referred to as a profile identifier in the profile feature word related storage unit 17). Extracted from the profile feature word related storage unit 17. Next, the profile update unit 16 includes a profile identifier that is included in the profile identifier of the profile feature word related storage unit 17 but is not included in the profile identifier of the profile storage unit 11 (hereinafter referred to as an insufficient profile identifier). Is acquired (step S35). The profile update unit 16 extracts the degree of association corresponding to the feature word identifier and the insufficient profile identifier included in the difference information received from the user feature word matching unit 14 from the profile feature word association storage unit 17 as in step S32 ( Step S36).

  And the profile update part 16 correct | amends the interest degree of profile information for every profile identifier and characteristic word identifier similarly to step S33S-S33E (step S37S-S37E, step S38).

  Finally, the profile update unit 16 uses a profile item that does not exist in the profile information (that is, the profile identifier) for the user identified by the user identifier of the received difference information based on the correction values calculated in steps S34 and S38. ) Is newly added (step S39). When it determines with adding at step S39 (YES in step S39), the profile update part 16 adds to the profile memory | storage part 11 with an appropriate degree of interest as a user's profile item identified by the user identifier of the received difference information. (Step S3A).

  As a specific method of determination, for example, the profile update unit 16 compares the correction value calculated in step S34 with the correction value calculated in step S38. At this time, when the correction value calculated in step S38 is larger than the correction value calculated in step S34, a new addition may be made. In addition, when newly adding, the profile update unit 16 may correct the correction value calculated in step S38 to an appropriate value and add it. As described above, the correction value calculated in step S38 is a reference value for determining whether or not to newly register a profile item in the profile storage unit 11, and thus can be referred to as a registration reference value.

  Processing in the case of adding in steps S39 to S3A will be described using a specific example. For example, the profile update unit 16 calculates the average value V1 of the correction values calculated in step S34. If the correction value V2 for each profile identifier calculated in step S38 exceeds V1, the profile update unit 16 adds the profile identifier to the profile storage unit 11. At this time, for example, the profile update unit 16 may uniformly set “1” as the degree of interest. Alternatively, the profile update unit 16 may calculate the ratio V2 / V1 between V1 and the correction value V2 and round the calculation result to an integer as the degree of interest.

  As described above, the profile update unit 16 uses the profile identifier stored in the profile feature word related storage unit 17 even for the profile identifier not stored in the profile storage unit 11 in advance by the process of step S38. Update profile information. Therefore, the interest level of the profile item identified by the profile identifier can be reflected in the user's profile information without explicitly updating the profile information on the system.

  In the above description, the case where the profile update unit 16 performs the profile update process when receiving the difference of the user feature word related information from the user feature word matching unit 14 has been described. In addition, the profile update unit 16 may periodically perform profile update processing such as daily batch processing.

  For example, the profile update unit 16 corrects user feature word related information registered in the user feature word related storage unit 15 after the date and time of the previous batch processing (hereinafter referred to as the last batch processing date and time). A value may be calculated to correct the interest level of the profile information. In this case, the profile update unit 16 may use user feature word related information whose last update date and time in the user feature word related storage unit 15 is later than the last batch processing date and time as difference information.

  Further, the profile update unit 16 may clear the frequency of the user feature word related information that is the target of the batch process to 0 after the batch process ends. By doing so, the profile update unit 16 can calculate the correction value as in the above embodiment.

  Further, the profile update unit 16 may calculate the correction value using the accumulated value (that is, without clearing the frequency to 0 after the batch process is completed). By doing so, the profile update part 16 can correct | amend an interest degree according to the browsing condition which the user browsed so far. Note that profile update processing by batch processing is not limited to these methods.

  FIG. 9 is a flowchart illustrating an example of content recommendation processing. Upon receiving the user identifier and the user recommendation request from the content reference unit 3 (step S41), the content query unit 18 reads the profile information corresponding to the received user identifier from the profile storage unit 11 (step S42). For each of the profile identifiers included in the read profile information (steps S43S to S43E), the content query unit 18 reads the corresponding feature word identifier and the degree of association from the profile feature word related storage unit 17, for example, in step S42. The degree of interest of the profile identifier and its degree of association are multiplied for each feature word identifier and accumulated (step S44). This cumulative value is the degree of interest for each feature word of the user, and can be called the feature word interest level.

  Further, the content query unit 18 obtains a content identifier associated with the feature word identifier from the content feature word related storage unit 13 for each feature word identifier read from the profile feature word related storage unit 17 in step S44 (S45S to S45E). Search is performed (step S46). The content query unit 18 calculates the score by accumulating the feature word interest level accumulated in S44 for each content identifier searched in step S46 (step S47).

  The content query unit 18 arranges the list of content identifiers searched in step S46 in the descending order of the score calculated in step S47, and transmits the list to the content reference unit 3.

  According to the present invention, the user feature word matching unit 14 generates user feature word related information based on the browsing history information of the user and the feature words of the content browsed by the user. Then, the profile update unit 16 reads the profile identifier and the degree of association corresponding to the feature word included in the user feature word related information received from the user feature word matching unit 14 from the profile feature word relationship storage unit 17. When the read profile information is stored in the profile storage unit 11 as user profile information, the profile update unit 16 updates the interest level using the read association level. On the other hand, when the read profile information is not stored in the profile storage unit 11, the profile information is newly registered in the profile information storage unit 11. Thereafter, the content query unit 18 calculates the feature word interest level based on the interest level with respect to the profile information stored in the profile information storage unit 11 and the relevance level of the feature word corresponding to the profile information. Then, the content query unit 18 calculates a score based on the feature word and the feature word interest level. Therefore, even if the content user's preference changes with time, the content can be recommended according to the preference at that time.

  In other words, items that are not set as profile information by the user can also be newly added as profile items by determining the presence or absence of interest using the browsing history. Therefore, even if the profile information on the system is not explicitly updated, content can be recommended by guessing the user's preference at the time of recommendation.

  Hereinafter, the present invention will be described with reference to specific examples, but the scope of the present invention is not limited to the contents described below.

  In the following description, the user specified by the user identifier X (hereinafter referred to as user X) includes the feature word specified by the feature word identifier w004, and the content specified by the content identifier c004 is 2008. The user characteristic word related information update and profile update processing when viewed once on August 7, 2012 will be described. The profile storage unit 11, the browsing history storage unit 12, the content feature word related storage unit 13, the profile feature word related storage unit 17, and the user feature word related information 15 have values shown in the examples of FIGS. Is set.

  The user feature word matching unit 14 reads the browsing history information (X, c004, 1) from the browsing history storage unit 12 (step S21), and acquires the content identifier c004 of the content browsed by the user X from the browsing history information. (Step S22). Next, the user feature word matching unit 14 reads the content feature information (c004, w004) having the content identifier c004 from the content feature word related storage unit 13 (step S23). The user feature word matching unit 14 combines the browsing history information (X, c004, 1) and the content feature information (c004, w004) with the content identifier c004, and a set of the user identifier and the feature word identifier (X, w004). The number of appearances is counted as 1 from the browsing history information (step S24).

  In this embodiment, since the combination of the user identifier and the feature word identifier does not exist in the user feature word related information 15 shown in FIG. 5 (NO in step S25), the user feature word matching unit 14 uses this set. It is memorize | stored in the user characteristic word related storage part 15 as user characteristic word related information (S26N). The user feature word matching unit 14 notifies the profile update unit 16 of the difference (X, w004, 1) of the user feature word related information in order to cause the profile update unit 16 to perform the profile update process (step S27).

  When the profile update unit 16 receives the difference (X, w004, 1) of the user feature word related information from the user feature word matching unit 14, the profile update unit 16 uses the user identifier X included in the difference information to Profile information (X, p001, 4) and (X, p002, 2) associated with X are extracted from the profile storage unit 11 (step S31). Next, the profile update unit 16 performs profile feature word information (p001, w001, 4), (p001, w002, 2), (p002, w003) including the profile identifiers p001 and p002 included in the profile information extracted in step S31. 3) is acquired from the profile feature word related storage unit 17 (step S32). In the present embodiment, the feature word identifier w004 included in the difference of the received user feature word related information does not include the feature word identifier acquired in step S32. For this reason, the profile update unit 16 does not correct the interest level of the profile information acquired in step S32 (steps S33S to S33E, step S34).

  Next, the profile update unit 16 extracts a profile identifier corresponding to the feature word identifier w004 included in the difference information (X, w004, 1) received from the user feature word matching unit 14 from the profile storage unit 11. In the present embodiment, the profile identifier corresponding to w004 does not exist in the profile storage unit 11. Next, the profile update unit 16 obtains the profile identifier p004 corresponding to the feature word identifier w004 included in the difference information (X, w004, 1) received from the user feature word matching unit 14 from the profile feature word related storage unit 17. Extract. This p004 is not included in the user profile information. Therefore, the profile update unit 16 extracts the relevance “1” corresponding to the feature word identifier w004 and the shortage profile identifier p004 included in the difference information received from the user feature word matching unit 14 from the profile feature word related storage unit 17. (Step S36).

  In this embodiment, the relevance “1” of the feature word identifier for the profile identifier p004 is used as the interest level as it is (S37S, S38, S37E). Finally, based on the degree of association “1” of the profile identifier p004 with the feature word identifier w004 in step S38, a profile item p004 not included in the profile information is newly added for the user X with the degree of interest “1” (step S39, Step S3A).

  FIG. 10 is a block diagram showing an outline of the content recommendation system in the present invention. The content recommendation system according to the present invention is a profile that stores, for each user, profile items (for example, profile items, profile identifiers) indicating the user's preference field, and interest levels indicating the user's interest in the profile items. Storage unit 81 (for example, profile storage unit 11), browsing history storage unit 82 (for example, browsing history storage unit 12) for storing a user's content browsing history, and feature word identification information that characterizes content Content feature word related storage means 83 (for example, content feature word related storage unit 13) that stores word identification information (for example, feature word identifier) for each content, and an association indicating the degree of association between the profile item and the feature word A profile feature that stores the degree for each profile item and feature word identification information Word relationship storage means 84 (for example, profile feature word related storage unit 17), user's content browsing history stored in browsing history storage means 82, and feature words for each content stored in content feature word related storage means 83 Based on the identification information, user feature word generation means 85 for generating user feature word related information including the number of browsing times of the user who has browsed the content characterized by the feature word for each user and feature word identification information. (For example, the user feature word matching unit 14) and profile feature word relation storage that stores profile items and association levels corresponding to feature word identification information included in the user feature word related information generated by the user feature word generation means 85. The reading means 86 (for example, the profile update unit 16) that reads from the means 83 and the profile item read by the reading means 86 are useful. When the profile storage unit 81 stores the profile item as a user's profile item, the degree of interest for updating the interest level for the user's profile item stored in the profile storage unit 81 using the relevance level read by the reading unit 86 The profile item read by the update unit 87 (for example, the profile update unit 16) and the reading unit 86 is not stored as the user's profile item in the profile storage unit 81 (for example, the profile information read by the reading unit 86 is insufficient). Profile registration unit 88 (for example, profile update unit 16) for registering the user's profile item and interest level in the profile storage unit 81 using the relevance read by the reading unit 86. Stored in profile storage means 81 Based on the degree of interest corresponding to the profile item and the degree of association between the feature word identification information corresponding to the profile item stored in the profile feature word relationship storage unit 84, the feature word that is the interest level for each feature word of the user Based on the feature word interest level calculating means 89 (for example, the content query unit 18) for calculating the degree of interest, the feature word identification information corresponding to the content stored in the content feature word related storage means 83, and the feature word interest level. The content interest level calculating means 90 (for example, the content query unit 18) for calculating the interest level (for example, score) of each content of the user.

  With such a configuration, even if the content user's preference changes with time, the content can be recommended according to the preference at that time.

  That is, even when the profile information read by the reading unit 86 is not stored in the profile storage unit 81 as the user's profile information, the profile registration unit 88 uses the relevance read by the reading unit 86 to store the profile information. The profile information and interest level of the user are registered in the means 81. Therefore, even if the profile information on the system is not explicitly updated, content can be recommended by guessing the user's preference at the time of recommendation.

  In the above embodiment, the user feature word related storage unit (for example, the user feature word related storage unit 15) that stores the user feature word related information and the user generated by the user feature word generation unit 85 are used. User feature word registration means (for example, user feature word matching unit 14) for registering feature word related information in the user feature word related storage means, and reading means 86 is stored in the user feature word storage means. A configuration is disclosed in which a profile item and a degree of association of a user corresponding to feature word identification information included in the user feature word related information are read from the profile feature word relationship storage means.

  Further, the user feature word registration means registers the user feature word related information and the update date and time (for example, the last update date and time) when the user feature word related information is updated, and the reading means 86 uses the information. User feature word related information stored in the user feature word storage means and registered in the user feature word related storage means after the date and time when the interest level update means 87 updated the previous interest level (for example, the last update date and time). User feature word related information or usage registered in the user feature word related storage means in the user feature word related storage means after the date (for example, the last update date and time) when the profile registration means 88 registered the degree of interest last time. The profile item and the degree of association of the user corresponding to the feature word identification information included in the person feature word related information may be read from the profile feature word relationship storage means 84 with reference to the update date and time. (E.g., reading method of the difference information in a batch process).

  In the above-described embodiment, the profile item and the degree of relevance that the reading unit 86 reads the registration reference value (for example, the correction value V2) indicating the criterion for determining whether or not to register the user's profile item and the degree of interest are included. A registration reference value calculation means (for example, profile update unit 16) to be calculated based on the registration reference value calculated by the registration reference value calculation means by the profile registration means 88 is a predetermined value (for example, average value V1). The user's profile item and the degree of interest may be registered in the profile storage unit 81 when the value is greater than the value.

  Further, the browsing history storage unit 82 stores a browsing history including browsing number specifying information that can specify the number of times the user browses the content, and the interest level updating unit 87 identifies feature words included in the user characteristic word related information. The user's browsing frequency specifying information corresponding to the information is distributed according to the ratio of the relevance corresponding to the feature word identification information stored in the profile characteristic word relationship storage means 84, and according to the distributed browsing frequency specifying information. The degree of interest may be updated.

  The browsing history storage unit 82 stores a browsing history including browsing number specifying information that can specify the number of times the user browses the content, and the profile registration unit 88 includes feature word identification information included in the user characteristic word related information. Is distributed according to the ratio of relevance corresponding to the feature word identification information stored in the profile feature word relationship storage means 84, and is used according to the distributed browse number specifying information. A person's profile item and interest level may be registered in the profile storage means 81.

  The present invention is preferably applied to a content recommendation system that recommends content according to the user's preference.

It is a block diagram which shows one Embodiment of the content recommendation system by this invention. It is explanatory drawing which shows the example of profile information. It is explanatory drawing which shows the example of browsing history information. It is explanatory drawing which shows the example of content characteristic information. It is explanatory drawing which shows the example of user characteristic word related information. It is explanatory drawing which shows the example of profile characteristic word related information. It is a flowchart which shows the example of the update process of a user characteristic word related information. It is a flowchart which shows the example of a profile update process. It is a flowchart which shows the example of a content recommendation process. It is a block diagram which shows the outline | summary of the content recommendation system in this invention. It is explanatory drawing of the information presentation assistance system relevant to this invention.

Explanation of symbols

1 content recommendation system 11 profile storage unit 12 browsing history storage unit 13 content feature word related storage unit 14 user feature word matching unit 15 user feature word related storage unit 16 profile update unit 17 profile feature word related storage unit 18 content query unit 2 Communication network 3 Content reference section

Claims (10)

Profile storage means for storing, for each user, a profile item indicating a user's preference field, and an interest level indicating a degree of interest of the user with respect to the profile item,
Browsing history storage means for storing a user's content browsing history;
Content feature word related storage means for storing, for each content, feature word identification information that is identification information of a feature word that characterizes the content;
Profile feature word relationship storage means for storing a degree of association indicating a degree of association between the profile item and the feature word for each profile item and feature word identification information;
Based on the user's content browsing history stored in the browsing history storage means and the feature word identification information for each content stored in the content feature word related storage means, the content characterized by the feature word was browsed User feature word generation means for generating user feature word related information including the number of times the user browses for each user and feature word identification information;
A reading unit that reads a profile item and a degree of association corresponding to the feature word identification information included in the user feature word related information generated by the user feature word generation unit from the profile feature word relation storage unit;
When the profile item read by the reading unit is stored in the profile storage unit as the user's profile item, the use stored in the profile storage unit using the degree of association read by the reading unit Interest level update means for updating the interest level of the profile item of the person,
When the profile item read by the reading unit is not stored as the user's profile item in the profile storage unit, the profile stored in the user is stored in the profile storage unit using the degree of association read by the reading unit. Profile registration means for registering items and degrees of interest;
Based on the degree of interest corresponding to the profile item stored in the profile storage unit and the degree of association between the feature word identification information corresponding to the profile item stored in the profile feature word relationship storage unit, A feature word interest degree calculating means for calculating a feature word interest degree that is an interest degree for each word;
Content interest level calculation means for calculating the interest level for each content of the user based on the feature word identification information corresponding to the content stored in the content feature word related storage means and the feature word interest level. A content recommendation system characterized by this. User feature word related storage means for storing user feature word related information;
User feature word registration means for registering the user feature word related information generated by the user feature word generation means in the user feature word related storage means,
The reading unit reads from the profile feature word relationship storage unit the profile item and the degree of association of the user corresponding to the feature word identification information included in the user feature word related information stored in the user feature word storage unit. The content recommendation system according to 1. The user feature word registration means registers the user feature word related information in association with the update date and time when the user feature word related information is updated,
The reading means is user feature word related information stored in the user feature word storage means, and the usage registered in the user feature word related storage means after the date and time when the interest level update means updated the previous interest level. User feature word related information or feature word identification included in the user feature word related information registered in the user feature word related storage means in the user feature word related storage means after the date and time when the profile registration means registered the degree of interest last time The content recommendation system according to claim 2, wherein the profile item and the degree of association of the user corresponding to the information are read from the profile feature word relationship storage means with reference to the update date and time. A registration reference value calculating means for calculating a registration reference value indicating a criterion for determining whether or not to register the user's profile item and the degree of interest based on the profile item and the degree of association read by the reading means;
The profile registration means registers the user's profile item and the degree of interest in the profile storage means when the registration reference value calculated by the registration reference value calculation means is larger than a predetermined value. The content recommendation system according to any one of the above. The browsing history storage means stores a browsing history including browsing count specifying information capable of specifying the user's content browsing count,
The degree-of-interest update means is a degree of relevance corresponding to the feature word identification information stored in the profile feature word relation storage means, with respect to the user's browsing count specifying information corresponding to the feature word identification information included in the user feature word related information. The content recommendation system according to any one of claims 1 to 4, wherein the content recommendation system distributes the content according to the ratio and updates the degree of interest according to the distributed browsing frequency specifying information. The browsing history storage means stores a browsing history including browsing count specifying information capable of specifying the user's content browsing count,
The profile registration unit is configured to display the number-of-views specifying information of the user corresponding to the feature word identification information included in the user feature word related information with the degree of association corresponding to the feature word identification information stored in the profile feature word relation storage unit. The content according to any one of claims 1 to 5, wherein the content is distributed according to the ratio, and the profile item and the interest level of the user are registered in the profile storage unit according to the distributed browsing frequency specifying information. Recommendation system. Based on the user's content browsing history and feature word identification information that is characteristic word identification information that characterizes the content defined for each content, includes the number of times the user has viewed the content characterized by the feature word A user feature word generation step for generating user feature word related information;
From the profile feature word relation storage means for storing the degree of association between the profile item indicating the field of user preference and the feature word for each profile item and feature word identification information, the user feature word A reading step of reading the profile item and the degree of association corresponding to the feature word identification information included in the user feature word related information generated in the generation step;
When the profile item read in the reading step is stored in the profile storage means as the user's profile item, the user stored in the profile storage means using the relevance degree read in the reading step An interest level update step to update the level of interest in the profile item,
If the profile item read in the reading step is not stored as the user's profile item in the profile storage means, the relevance read in the reading step is used to store the profile of the user in the profile storage means. Profile registration means for registering items and degrees of interest;
Based on the degree of interest corresponding to the profile item stored in the profile storage unit and the degree of association between the feature word identification information corresponding to the profile item stored in the profile feature word relationship storage unit, A feature word interest level calculating step for calculating a feature word interest level that is an interest level for each word;
A content recommendation method comprising: a content interest level calculation step of calculating an interest level for each content of a user based on feature word identification information corresponding to the content and the feature word interest level. A user feature word registration step of registering the user feature word related information generated in the user feature word generation step in the user feature word related storage means for storing the user feature word related information;
The reading step reads the profile item and the degree of association of the user corresponding to the feature word identification information included in the user feature word related information stored in the user feature word storage means from the profile feature word relation storage means. 7. The content recommendation method according to 7. On the computer,
Based on the user's content browsing history and feature word identification information that is characteristic word identification information that characterizes the content defined for each content, includes the number of times the user has viewed the content characterized by the feature word User feature word generation processing for generating user feature word related information,
From the profile feature word relation storage means for storing the degree of association between the profile item indicating the field of user preference and the feature word for each profile item and feature word identification information, the user feature word A reading process for reading the profile item and the degree of association corresponding to the feature word identification information included in the user feature word related information generated in the generation step;
When the profile item read in the reading process is stored in the profile storage unit as the user's profile item, the user stored in the profile storage unit using the degree of association read in the reading step Interest level update process to update the level of interest in the profile item,
When the profile item read in the reading process is not stored in the profile storage unit as the user's profile item, the profile of the user is stored in the profile storage unit using the relevance read in the reading step. Profile registration process to register items and interests,
Based on the degree of interest corresponding to the profile item stored in the profile storage unit and the degree of association between the feature word identification information corresponding to the profile item stored in the profile feature word relationship storage unit, A feature word interest degree calculation process for calculating a feature word interest degree that is an interest degree for each word, and
A content recommendation program for executing a content interest level calculation process for calculating an interest level for each content of a user based on feature word identification information and feature word interest level corresponding to the content. On the computer,
Executing a user feature word registration process for registering the user feature word related information generated in the user feature word generation step in the user feature word related storage means for storing the user feature word related information;
In the reading process, the profile item and the degree of association of the user corresponding to the feature word identification information included in the user feature word related information stored in the user feature word storage unit are read from the profile feature word relationship storage unit. Item 9. The content recommendation program according to Item 9.

Images