JP5668010B2 - Information recommendation method, apparatus and program - Google Patents

Description

  The present invention sequentially learns the user's interest and the degree of influence from the information sender using the user's selection history for information such as content, and selects information to be recommended to the user based on the learning result. The present invention relates to an information recommendation method, apparatus, and program to be presented.

  There is a demand for a technique for recommending appropriate services and contents in accordance with user behavior and situations. In view of this, an information recommendation service has been proposed in which information suitable for each information user is selected or information is rearranged in an order suitable for each information user and displayed. For example, when the information is an advertisement, it is an advertisement recommendation service that displays an advertisement by selecting an advertisement suitable for each information user or rearranging in an order suitable for each information user, and when the information is news It is a news recommendation service that selects news that is appropriate for each information user, or displays the news in an order that is appropriate for each information user. If the information is restaurant information, it is appropriate for each information user. It is a restaurant information recommendation service that displays restaurant information by selecting restaurant information or rearranging in order suitable for each information user.

  As a method for realizing the above information recommendation service, the user's interest is estimated from the frequency of the concept or the like appearing in the user history on the premise that the concept tag (metadata) for summarizing the contents is assigned to each content. There is a method, and this method is a content-based filtering method (Content Based Filtering: CBF).

  Specifically, the content-based filtering technique presents products of the same brand (holding the same concept tag) when browsing products of a specific brand (holding information indicating the brand as a concept tag). In this case, the memory-based method presents a product of a specific brand if the specific brand is frequently browsed from the history of browsing in the past. In a simple method, products related to concept tags that appear more frequently in the browsing history are presented. As a method using a taxonomy (ontology) in such a method, for example, “Proposal of a multifaceted preference grasping method using an operation history” described in Non-Patent Document 1 is available.

In addition, the following methods are considered as methods for learning the user's interest and the influence from the information sender using the selection history of the user's content and recommending the content based on the learning result. . FIG. 22 is a diagram showing an example thereof.
That is, when the user starts to use the service, for example, a list of randomly selected contents is displayed regardless of the user's interest. Then, when the user selects the content that he / she is interested in from the displayed first list, for example, when browsing the detailed web page or registering the bookmark, the history of this selection action is used. The user's interest is learned, the content is selected again based on the learning result, and the list is displayed.

  In the example shown in FIG. 22, if the user selects “eel”, “gyudon”, and “ramen” in which the user is interested in the first display list and browses the details, the user is based on the selection history. Gives the estimated result that he likes “eel”, “gyudon” and “ramen”. Then, when the user requests display of the list in this state, a second list is displayed that has been re-sorted so that “eel”, “gyudon”, and “ramen” are ranked higher based on the above estimation result. Is done. In addition, when the user selects content again and browses the details of the second display list, the estimation result is updated based on the selection history reflecting the selection operation. A sorted third list is displayed. Similarly, learning proceeds according to the user's selection within the range of the displayed list, and a list re-sorted based on the result is displayed.

“Proposal of multifaceted preference grasp method using operation history”, IEICE Technical Report, vol. 110, no. 450, LOIS2010-76, pp. 67-72, March 2011

However, the conventional recommendation method described above has the following problems to be solved.
(1) Learning proceeds based on the list displayed for the first time. For this reason, gourmet genres that the user has learned to like are displayed at the top of the list in the subsequent list display, but those that are not initially displayed in the list (for example, “sushi” or “Okinawa cuisine” ")" Is not displayed at the top even if the user "likes" them. For this reason, as the number of times the user uses is increased, the gourmet genre whose interest is “unknown” due to the fact that the user's interest is not learned is less likely to be displayed at the top. Therefore, there is a problem that learning of only the existing concept proceeds and information is narrowed down before the comprehensiveness of various interest evaluations is sufficiently ensured.

  (2) In the example shown in FIG. 22, until the fourth display list is displayed, the number of times the list including “gyudon” is displayed is 3, and the number of times “gyudon” is selected, that is, the user However, the number of selection histories that can be considered “gyudon” is 5 times. On the other hand, the number of times the list including “eel” is displayed once and the number of times “eel” is selected is also one. However, the results of the interest estimation performed when the display list is displayed for the fourth time are “gyudon” “like”, “eel” “like”, “ramen” “like”. Not considered. For this reason, there is no distinction between “eel” that the user may have just seen in detail and “gyudon” estimated to be “like” from a plurality of selection histories. That is, there is a possibility that the interest estimation result at a stage that has not been sufficiently learned is used for sorting for displaying the list, that is, there is a possibility of erroneous evaluation.

  The present invention has been made paying attention to the above circumstances, and its main purpose is to provide an information recommendation method, apparatus, and program capable of ensuring sufficient and sufficient learning opportunities to be estimated and reducing erroneous evaluation. is there.

One aspect of the present invention in order to achieve the above object, in the information recommendation method or apparatus for presenting selected information to be recommended to the User chromatography The First List View content list and the detailed view content list of the user And calculating information indicating the degree of interest of the user with respect to the information to be presented, and from the list browsing content list of the user, the detailed browsing content list, and recommended information from other users, Information representing the degree of influence from other users when taking action based on the information is calculated. Then, for each piece of information to be presented or a set of the information, a feature history of the user based on a presentation history of the information to be presented to the user or a user behavior history based on the information to be presented. Information representing the reliability of the estimation result is calculated. Then, based on both the information indicating the degree of interest of the user with respect to the presentation target information and the information indicating the degree of influence, and the information indicating the calculated reliability, the information on the presentation target or the An evaluation score for the set is calculated, and a list of information to be recommended to the user is generated based on the calculated evaluation score. The information to be presented includes, for example, a concept tag and a recommendation tag.

One aspect of the present invention is also characterized by comprising the following aspects.
In the first aspect, when the information representing the reliability is calculated, the total number N of information corresponding to the reliability calculation target and the information presented to the user among the information corresponding to the reliability calculation target And the total value m of the user's action for the information corresponding to the reliability calculation target, the total number N of the acquired information, the total value n of the presentation information, and the total value m of the action The reliability coefficient is calculated based on at least one of the above.

  In the second aspect, when the list is generated, a first region presenting recommended information selected based on the calculated evaluation score, and a second region presenting arbitrarily selected recommendation information A list is generated so as to have a region.

  Therefore, according to one aspect of the present invention, when calculating the evaluation point for the information to be presented or the set thereof, the calculated information is the information indicating the degree of interest of the user or the information indicating the degree of influence. Since information indicating the reliability is taken into account, it is possible to secure a learning opportunity for another estimation target until the reliability of the estimation result is ensured, and to reduce erroneous learning. In addition, it is possible to reduce errors in list display due to erroneous evaluation of contents.

  In addition to the first area for presenting recommended information selected based on the evaluation score, the list is added with a second area for presenting arbitrarily selected recommendation information, so that learning progresses to some extent. Even at this stage, unlearned evaluation item targets can be continuously presented to the user, and a wide range of interests can be learned.

  That is, according to one aspect of the present invention, it is possible to provide an information recommendation method, apparatus, and program capable of ensuring a sufficient learning opportunity for an estimation target and reducing erroneous evaluation.

The figure which shows the whole structure of the information recommendation system which concerns on one Embodiment of this invention. The block diagram which shows the function structure of each apparatus of the information recommendation system shown in FIG. The figure which shows the outline | summary of the interest analysis process by the interest analysis apparatus shown in FIG. The schematic diagram for demonstrating operation | movement of the feature score calculation part of interest. The figure which shows the detail of the feature score calculation process of an interest. The figure which shows the detail of a concept system update process. The figure which shows an example of the recommendation tag provided to a content. The figure which shows an example of a list browsing content list and a detailed browsing content list. The figure which shows the 1st Example of the influence analysis process by the influence analysis apparatus shown in FIG. The figure which shows the 2nd Example of the influence analysis process by the influence analysis apparatus shown in FIG. The figure which shows the 3rd Example of the influence analysis process by the influence analysis apparatus shown in FIG. The figure which shows the 4th Example of the influence analysis process by the influence analysis apparatus shown in FIG. The flowchart which shows the outline | summary of the procedure of the reliability coefficient calculation process by the reliability coefficient calculation apparatus shown in FIG. 2, and the content scoring process using the calculation result. The figure which shows the relationship between the total value of the content display number for describing the 1st Example of a reliability coefficient calculation process, and a reliability coefficient. The conceptual diagram for demonstrating the 2nd Example of a reliability coefficient calculation process. The figure which shows the relationship between the total value of the content display number for demonstrating the 2nd Example of a reliability coefficient calculation process, and a reliability coefficient. The figure which shows the relationship between the total value of the content display number for demonstrating the 3rd Example of a reliability coefficient calculation process, and a reliability coefficient. The figure which shows the outline | summary of the process in the case of performing content evaluation based on an interest score and an influence score. The figure which shows an example of the memory | storage information of the reliability calculation origin database and reliability coefficient database which were provided in the reliability analyzer shown in FIG. The figure which shows the example of a display of the list information of the sorted gourmet genre. The figure which matches and shows the list information of the sorted shop, and the gourmet genre of the said shop. The figure for demonstrating an example of the information recommendation method considered conventionally.

Embodiments according to the present invention will be described below with reference to the drawings.
FIG. 1 is an overall configuration diagram of an information recommendation system according to the present invention. This system includes a client terminal 400, a content server 500, an interest analysis device 100, an influence analysis device 200, a content scoring device 300, and a reliability analysis device 600.

The client terminal 400 and the content server 500 are connected via a communication network. Further, the content server 500, the interest analysis device 100, the influence analysis device 200, the content scoring device 300, and the reliability analysis device 600 are also connected by a communication network. Furthermore, the interest analysis device 100, the influence analysis device 200, the content scoring device 300, and the reliability analysis device 600 are also connected to each other via a communication network.
By performing a browsing operation on the client terminal 400, the user obtains desired content from the content server 500 and presents the acquired content on the screen of the client terminal 400 for browsing.

  The client terminal 400 collects a content browsing history by a user operation, and browses a list browsing content list (first content list) in which a plurality of contents are browsed as a list, and a detailed browsing content list (in which a content body is browsed from the content list ( A second content list) to the content server 500.

  The content server 500 has a function of transferring the list browsing content list and the detailed browsing content list to the interest analysis device 100, the influence analysis device 200, and the reliability analysis device 600 via the communication network.

The interest analysis device 100 has a function of calculating a feature score of interest and a user interest score for each concept appearing in the content based on the list browsing content list and the detailed browsing content list, and estimating the user's interest.
However, the interest analysis apparatus 100 may update the conceptual system / user interest score database by using, for example, other content-based filtering (CBF) other than the user interest score calculation method described below. That is, it may be other than the method of estimating interest based on the list browsing content list and the detailed browsing content list.

  The influence analysis device 200 calculates the characteristic score and the influence score of the influence on other users (individuals or groups) who give “recommended tags” to the content based on the list browsing content list and the detailed browsing content list. And has a function of estimating the degree of influence from other users (individuals or groups) on the user.

  The reliability analysis apparatus 600 collects the list browsing content list and the detailed browsing content list, and the total number of information corresponding to the reliability coefficient calculation target (for example, the total number of stores, the total number of news, the total number of recommended tags, etc.) N , The cumulative value n of the number of display items of information corresponding to the reliability coefficient calculation target of the target user, the behavior history (for example, detailed page browsing history, The accumulated value m of the bookmark registration history etc. is obtained, and this information is stored as reliability calculation source data. Then, based on the stored reliability calculation source data N, n, m, a reliability coefficient T of information corresponding to the reliability coefficient calculation target is calculated, and the calculated reliability coefficient T Has the function of saving.

  The content scoring device 300 inquires of the interest analysis device 100 or the influence analysis device 200 about the user interest score or the influence score, respectively, and also sends the information corresponding to the reliability coefficient calculation target to the reliability analysis device 600. Queries the reliability coefficient T. Then, content evaluation is performed based on the user interest score or the influence score acquired from the interest analysis apparatus 100 or the influence analysis apparatus 200 in response to the inquiry and the reliability coefficient T acquired from the reliability coefficient analysis apparatus 600, respectively. It has a function of calculating a score, generating a list of content sorted based on the calculated evaluation score, that is, a sorted content list, and transmitting it to the content server 500.

The configurations of the client terminal 400, the content server 500, the interest analysis device 100, the influence analysis device 200, the content scoring device 300, and the reliability analysis device 600 will be described in detail below. FIG. 2 is a block diagram showing the functional configuration.
Each device in FIG. 2 includes a CPU (Central Processing Unit), and realizes each function described above by causing the CPU to execute a program stored in advance in a program memory.

(1) Interest Analysis Device The interest analysis device 100 includes a history information receiving unit 110, an interest feature score calculation unit 120, an interest concept system update processing unit 130, and a concept system / user interest score database 140. The concept system / user interest score database 140 stores a root concept node ID, a concept system table, and a user interest score table. The root concept node ID is the highest concept node ID in the concept system structure, and there is only one root node in the system.

  The concept system table stores a self-concept ID, a parent concept ID list, and a child concept ID list. All self-concept IDs in the concept system are stored in association with a parent concept ID and a child concept ID (however, if the self-concept is the lowest, there is no child concept ID), thereby defining the concept structure Is done.

  The user interest score table stores the value of the user interest score (TotalZ) in association with the concept ID and the user ID (client terminal ID). The definition and calculation method of TotalZ will be described later.

FIG. 3 shows an outline of processing of the interest analysis apparatus 100.
The history information receiving unit 110 receives the list browsing content list and the detailed browsing content list generated in the client terminal 400 via the content server 500. The list browsing content list is, for example, a list of content in which the user browses only the content titles in a list. The detailed browsing content list is a list of content that the user has viewed the content (details) of the content body. FIG. 3 illustrates a case where the list browsing content list includes contents 1 to 8 and the detailed browsing content list includes contents 1, 3, and 4. In FIG. 3, the content indicated by the hatched pattern indicates that the concept B appears in the contents 1, 6, 7, and 8.

The interest feature score calculation unit 120 calculates an interest feature score of each concept using a statistical model of concept selection using the browsed content list and the detailed browsing content list.
FIG. 4A shows a schematic diagram of the analysis parameter extraction process. For example, the case where the user browses the details of 10 (= a) contents from a list of 50 (= S) contents is shown. Here, there are 15 articles (= U) including the concept of “baseball” among the 50 contents displayed in a list, and the concept of “baseball” is included in 10 contents viewed by the user. This indicates that there are five (= b) contents including “”.

  The interest feature score calculation unit 120 calculates the feature score Z for each concept ID using the analysis parameters S, a, U, and b. FIG. 5 shows details of the feature score calculation process of interest. In FIG. 5, i represents a concept identifier, and j represents a cluster ID.

  Here, the cluster is an identifier uniquely assigned to the list browsing content list and the detailed browsing content list. When the user browses the list browsing content displayed at another time (time zone), another cluster ID is given. The conditions for newly assigning the cluster ID under conditions other than the time include when there is no operation for a certain period of time while the list browsing content list is displayed, or when the browsing user (user ID) is switched, the list browsing content list On the other hand, when a narrow search is performed from the viewpoint of the content genre or the like, the browsing mode may be switched in other browsing applications.

  When H1 (first probability) is the total number S of the list browsing contents included in the list browsing content list and the number of contents U in which the concept i appears among the list browsing contents, a detailed browsing content is randomly selected and viewed. In this case, the cumulative probability that the number of contents in which concept i appears is b or more. H2 (second probability) is the number of contents U where the concept i appears in the list browsing contents, and the number of contents where the concept i appears is b when the number of detailed browsing contents is randomly selected and viewed. The cumulative probability is In the present embodiment, the cumulative probabilities H1 and H2 are obtained by the hypergeometric distribution, but are not limited to this method. Examples of other distributions include a binomial distribution and a normal distribution.

  As shown in FIG. 4B, for example, among the 10 contents browsed by the user using the analysis parameters S, a, U, and b, there are 5 or more contents including the concept of “baseball”. A certain probability is “0.12”. Here, “0.12” corresponds to the value of the cumulative probability H1.

  As an example of using the value of H2, consider the case where b is 0 in the above analysis parameters. In this case, the probability when the number of appearances is 0 or less is calculated. Specifically, in FIG. 4B, the horizontal axis is the value of the item whose value is 0, so “0.02”.

  Then, as shown in FIG. 5, the interest feature score calculation unit 120 calculates the interest feature score Z by the inverse function of the cumulative distribution function of the standard normal distribution using the calculated cumulative probabilities H1 and H2. As shown in FIG. 4C, the feature score Z is obtained by the inverse function of the cumulative distribution function of the standard normal distribution with H1 as the cumulative probability. When H2 is used as the cumulative probability, the feature score is obtained with the sign of the inverse value of the inverse function of the cumulative distribution function of the standard normal distribution being negative. Using this feature score Z, the interest concept system update processing unit 130 described later obtains a user interest score (TotalZ) for the concept of “baseball”.

  The interest feature score calculation unit 120 generates an update target concept list and outputs it to the interest concept system update processing unit 130. The “update target concept list” is a set of a concept ID, a feature score Z calculated by the feature score calculation unit 120, and a weight w. The concept ID appearing in the update target concept list becomes a node (concept) to be updated in the next concept system update process. The feature score Z and the weight w are calculated for all the concept IDs in the appearance concept list to which the superordinate concept is added. The weight w is a value set for each concept in each cluster ID.

  Note that the weight w can be changed as follows when the initial value w = 1 and a user's characteristic operation is performed. For example, the user can register the content presented on the client terminal 400 as a favorite content, input an evaluation for the content, and the like. When the client terminal 400 can transmit such an operation history to the interest analysis apparatus 100, the following processing is performed.

  For example, when the content is registered as a favorite, the interest feature score calculation unit 120 increases the weight w = 1.5 for all concept IDs including the content. In addition, the value of weight w according to event information collected from content browsing time, browsing weather, temperature, humidity, season, day of the week, holiday, leisure time, user location information at browsing, scheduler, diary, etc. Can also be changed.

  The interest concept system update processing unit 130 updates the user interest score for each concept based on the relationship information (superordinate concept and subordinate concept) in the concept system using the feature score. Nodes included in the graph of the concept system represent concepts, and links represent relationships between concepts. The user interest score is held as a value of a node corresponding to each concept in the concept system. In the concept system, the nodes located at the higher level represent the abstract concept, and the nodes located at the lower level represent the specific concept. The concept system and concept ID (identifier assigned to each node) are designed and defined in advance by a service operator or the like.

FIG. 6 shows details of processing of the interest concept system update processing unit 130. The interest concept system update processing unit 130 uses the user interest score update formula for each concept i shown in FIG. 6 for the concept that has appeared in the content (appearance concept) and the concept ID of the superordinate concept of this appearance concept. The values of the scores TotalZ _in , X _in , and Y _in are obtained and stored in the user interest score table corresponding to the column of the user ID (client terminal ID) corresponding to the cluster ID in the concept system / user interest score database 140. Update each value that is present.

Here, Xi _(n-1) is the sum of the squares of the weights w of the update target concept list in the past (up to the previous time) for each concept ID (represented by identifier i here). Similarly, Yi _(n-1) is the sum of multiplication of the weight w of the past update target concept list and the feature score Z for each concept ID (represented by identifier i in this case).

In FIG. 6, n is an identifier indicating how many times the concept system update process is performed. A series of processes for obtaining the user interest score TotalZ _in is performed in cluster ID units. When the unit in which the series of processes is performed is counted once, n is the number of times this series of processes is performed. It is an identifier indicating i is an identifier of a concept ID. Z _in is a Z value for each updated content list of concept i. The Z _ij is a Z value for each of the list browsing content list and the detailed browsing content list, and has a relationship of Z _ij εZ _in . Weight w _in is the weight to be used in each process of updating the concept i. The weight w is the same as that set by the interesting feature score calculation unit 120.

For example, w _in the favorite registration, if the user a special operation other than the inspection of the public, such as with the other user has performed, and (interval to the end from the viewing start) content viewing time, related the degree of content and concepts, Contents are changed according to the viewing time of the content, whether it is the weather / temperature / humidity / season / day of the week / holiday / leisure at the time of browsing, user location information at the time of browsing, event information collected from the scheduler / diary, etc. It is also changed when specified by the user or service operator.

  Furthermore, the interest concept system update processing unit 130 extracts the subordinate concept of each concept ID (appearance concept and superordinate concept) in the “update target concept list”, and updates the subordinate concept node. In the extraction of subordinate concepts, for each concept ID of the “update target concept list”, the concept system table of the concept system / user interest score database 140 is referred to, a list of concept IDs is extracted from the child concept ID list, and each child ID is further extracted. The process of extracting the child concept list with reference to the concept system table for the concept ID of the concept ID list is repeated.

  The feature score Z used to update the interest level of the lower concept is, for example, the value having the largest absolute value of the feature score among the adjacent parent nodes, the value of the closest higher node, the average of the values of the parent nodes Or a probability-coupled value. In the “update target concept list”, the user interest score of the updated concept (the concept that appears in the content and the superordinate concept) is not updated.

(2) Client Terminal In FIG. 2, the client terminals 400-1 and 400-2 include a history collection unit 410, a history information transmission unit 420, a content presentation unit 430, a content request transmission unit 440, and a recommended tag addition unit. 450.

  The content request transmission unit 440 makes a content presentation request to the content server 500 in response to a user instruction (input). Specifically, for example, content request data including a client terminal ID (or user ID) and a request time is transmitted to the content server 500. The request time may be added in the content server 500. The client terminal ID (or user ID) is a number uniquely assigned to each terminal (or user), and is an ID that matches the user ID of the user interest score table of the conceptual system / user interest score database 140 described later. is there.

  The history collection unit 410 collects user operation histories and creates a list browsing content list and a detailed browsing content list. The history information transmission unit 420 transmits the list browsing content list and the detailed browsing content list created by the history collection unit 410 to the content server 500.

  When the influence analysis device 200 is used instead of the interest analysis device 100, a “recommended tag” is used. The recommended tag giving unit 450 is used when the user wants to recommend content to other people about the list display content list and the detailed browsing content displayed on the content presentation unit 430 of the client terminals 400-1 and 400-2. A recommended tag is given to the content by the operation input. Information indicating that a recommended tag has been added to the content (recommended information) is transmitted to the recommended tag input unit 570.

  As a preparation for estimating the degree of influence from other users, that is, information senders, the user can see recommended tags from other users (individuals or groups). That is, another user can add a “recommended tag” when he / she wants to recommend the browsed information to someone, and whether or not the “recommended tag” has been added can be viewed by another user.

  FIG. 7 shows an example of “recommended tags” given to the content. For example, when Mr. A gives a recommendation tag to “information 2” and “information 5” as a recommendation target for other users, “information 2” and “information 5” in the list browsing content list and the detailed browsing content list of AA "Recommended tag" (A recommended) is displayed.

(3) Content Server In FIG. 2, the content server 500 includes a content transmission processing unit 510, a sorted content list reception unit 520, a content list transmission unit 530, a content list input unit 540, and a history information transfer unit 550. The content request transfer unit 560, the recommended tag input unit 570, and the recommended tag transmission unit 580 are provided.

  The history information transfer unit 550 receives the list browsing content list and the detailed browsing content list received from the client terminals 400-1 and 400-2 via the communication network, the interest analysis device 100, the influence analysis device 200, and the reliability analysis device. Forward to 600.

  The content list input unit 540 receives a content list that displays a list of contents to be presented to the client terminals 400-1 and 400-2 used by the user by the service operator. The content list transmission unit 530 transmits the input content list to the content scoring device 300 via the communication network.

The content list includes a content ID, a concept ID / relevance list, a content body, a content registration time, and recommended tag information. The content ID is a unique ID assigned by the content server 500 to each content. The concept ID / relationship degree list stores a concept ID of a concept that appears in content and a set of values indicating the degree of relevance between the concept and the content. The concept ID / relevance degree list is set in advance for each content, and as a specific example, content 1 (sports article) has {baseball concept ID = 1, relevance = 0.5}, { A set of concept ID and degree of association is stored, such as “Soccer” concept ID = 2, degree of association = 0.8}, {“Golf” concept ID = 3, degree of association = 0.6}. The recommended tag information is stored in association with the content ID.
In the present embodiment, an example is described in which the user is stored for the first time by giving a recommendation tag by another user, but it may be stored in advance. For example, when the information sender is determined at the time of information presentation, such as a blog, it is assumed that the recommendation tag is given by the information being presented.

  The concept ID matches the value stored in the concept system / user interest score database 140. For example, the relevance is a value from 0 to 1, and the larger the relevance, the stronger the relevance. As the relevance, a value set by the service operator at the time of content registration or a value calculated by another system is used.

  The sorted content list receiving unit 520 receives a sorted content list and a client terminal ID (or user ID) obtained by sorting a part or all of the content list from the sorted content list transmission unit 310. The content transmission processing unit 510 transmits the sorted content list to the client terminals 400-1 and 400-2 corresponding to the client terminal ID (or user ID).

  The content request transfer unit 560 transfers content request data, which is a content presentation request transmitted from the content request transmission unit 440 of the client terminals 400-1 and 400-2, to the content evaluation processing unit 320.

(4) Impact Analysis Device The impact analysis device 200 includes a history information receiving unit 210, a human system database 220, an influence feature score calculation unit 230, and an influence level conceptual system update processing unit for each person / group. 240 and an influence concept system database 250 having an influence score for each person / group.

  The history information receiving unit 210 acquires a list browsing content list, a detailed browsing content list, and recommended tag information assigned to each list from the history information transfer unit 550. An example of a list browsing content list and a detailed browsing content list to which recommended tags are attached is shown in FIG.

  In FIG. 8, for example, Mr. AA considers a case where the client terminal 400-1 browses the details of 10 (= a) contents from the list of 50 (= S) contents. Here, of the 50 contents displayed in a list, 15 contents (= U) containing the concept of “baseball” and 10 contents with “Recommended by Mr. A” (= M), of which six (= d) contents include “Recommended by Mr. A” and “Baseball”. And among the 10 contents browsed by the user, 4 contents (= m) with “Mr. A's recommendation”, 5 contents (= b) containing the concept of “baseball”, of which It is assumed that there are four (= d) contents with “Recommended by Mr. A” and “Baseball”.

  When the user browses the information, if the browsing rate increases for the content with the recommended tag, the degree of influence from other users to whom the user's recommended tag is added is considered to be positive (+). In addition, when the browsing rate for information to which a recommendation tag is assigned decreases, the degree of influence from other users to whom the user's recommendation tag is assigned is regarded as negative (−). Moreover, when the browsing rate about the information with which the recommendation tag was provided did not change, it can be considered that there is no influence from the other user who provided the recommendation tag to a user.

Hereinafter, several embodiments of the influence analysis process in the example of FIG. 8 will be described.
(4-1) First embodiment
FIG. 9 is a diagram illustrating an influence analysis process according to the first embodiment. The influence degree characteristic score calculation unit 230 calculates the influence degree characteristic score Z ′ from Mr. A to Mr. AA using the above S, a, M, and m as analysis parameters. The characteristic score Z ′ of the influence degree can be calculated by replacing U and b with M and m in the calculation formula shown in FIG.

  In order to obtain the influence characteristic score Z ′, first, cumulative probabilities H1 and H2 are obtained. In H1, when the number of occurrences of “Recommended by Mr. A” among the S list contents is M, the detailed browsing in which “Recommended by Mr. A” appears when a detailed browsing content is randomly selected and viewed. This is the cumulative probability that the content is m or more. In addition, when the number of occurrences of “Recommended by Mr. A” out of the S list browsing contents is M, “Recommended by Mr. A” appears when a detailed browsing content is randomly selected and viewed. This is the cumulative probability that the number of detailed browsing contents to be m or less. Then, one of H1 and H2 that is less than 0.5 is adopted, and the feature score Z ′ is obtained by the inverse function of the cumulative distribution function of the standard normal distribution. When H2 is used as the cumulative probability, the sign of the return value of the inverse function of the standard normal distribution cumulative distribution function is set to be a negative characteristic score. If both H1 and H2 are 0.5 or more, the feature score is 0.

In the above case (S = 50, a = 10, M = 10, m = 4), H1 = 0.0966, H2 = 0.981, and the feature score is Z ′ = 1.3.
Thus, the characteristic score of the degree of influence on “Recommended by Mr. A” is calculated. The calculated feature score is passed to an influence concept system database 250 having an influence score for each person / group. In the influence degree concept system database 250 having an influence degree score for each person / group, an influence degree score is given to each concept constituting the same concept system as the interest estimation.

In the case of the first embodiment, since the degree of influence is regarded as equivalent for each concept, the characteristic score of the degree of influence is stored in the root concept of the concept system indicating the degree of influence of Mr. A. When the list content list is updated and the user's browsing history is also updated, the influence feature score is calculated again, and the influence score is updated. For the update, the influence score update formula of the following formula (1) is used. In the case of the first embodiment, the concept i is a root concept.

  In Equation (1), the weight w is an initial value w = 1, the user's operation history (registration as favorite content, evaluation input to the content, etc.), the browsing time and browsing status of the content, and the content And the recommended tag can be changed according to at least one of the relevance levels. In calculating the user influence score, the user influence score is more accurately based on the user's behavior and situation by reflecting the user's situation at the time of browsing and the characteristics of browsing operation (favorite registration, long-time browsing, etc.). It becomes possible to ask for.

  Also, the root concept score is given to the subordinate concepts that are descendants with the same value. After the update of the impact score for “Recommended by Mr. A” is completed, the characteristics related to the impact level are also applied in the same way as above for those who have been given a “recommended tag” other than Mr. A included in the same list browsing content. The score is calculated, and the influence score is updated using the above formula (1).

  That is, according to the first embodiment, the number of contents (M) in which each “recommended tag” appears in the list browsing content list which is a user selection candidate, and the detailed browsing content list selected by the user from the list browsing content list By calculating the number of contents (m) in which each “recommended tag” appears, and comparing and analyzing the M and m, the rareness of each “recommended tag” appears and the content of the selection candidate Since the “recommended tag” attached to can use the characteristics of the number of appearances of “not selected” as well as the event “relatively selected”, It is possible to estimate the degree of influence from a person with high accuracy.

(4-2) Second embodiment
In the second embodiment, a method for estimating the degree of influence for each specific concept (or specific genre) from Mr. A to Mr. AA will be described. FIG. 10 is a diagram showing an influence analysis process according to the second embodiment. The influence degree feature score calculation unit 230 calculates the influence degree feature score Z ′ for Mr. A using the analysis parameters S, a, D, and d. The characteristic score Z ′ of the influence degree in this case can be calculated by replacing U and b with D and d in the calculation formula shown in FIG.

In the case of the second embodiment, H1 selects a detailed browsing content at random when the number of the content having “A recommended” and “baseball” appearing at the same time among the S browsing content S is D. The cumulative probability that the number of detailed browsing contents in which “Mr. A's recommendation” and “baseball” appear at the same time will be d or more. In addition, when the number of contents where “A-san recommended” and “baseball” are simultaneously added among the list browsing contents is D, H2 is selected when a detailed browsing content is randomly selected and browsed. This is a cumulative probability that the number of detailed browsing contents in which “Mr. A's recommendation” and “baseball” appear simultaneously will be d or less. Then, one of H1 and H2 that is less than 0.5 is adopted, and the feature score Z ′ is obtained by the inverse function of the cumulative distribution function of the standard normal distribution. When H2 is used as the cumulative probability, the sign of the return value of the inverse function of the standard normal distribution cumulative distribution function is set to be a negative characteristic score. If both H1 and H2 are 0.5 or more, the feature score is 0.
From the above, the characteristic score of the degree of influence on “Mr. A's recommendation regarding baseball” is calculated.

  In addition, for a specific genre such as “sports”, which is a superordinate concept of “baseball”, the detailed browsing content is a using the number (= q) where “Mr. A's recommendation” and “sports” appear simultaneously. When browsing randomly selected items, select the cumulative probability (H1) that the number of detailed browsing contents that “Recommended by Mr. A” and “Sports” appear simultaneously is q or more, and a detailed browsing content. The cumulative probability (H2) that the number of detailed browsing contents in which “A-san's recommendation” and “sports” appear at the same time becomes q or less is obtained. Then, in the same manner as described above, the characteristic score of the degree of influence on “Mr. A's recommendation regarding sports” is obtained.

  Similarly, for “soccer”, which is a subordinate concept of “sports”, for example, using the number (= q) of “Mr. A's recommendation” and “Sports” appearing simultaneously, the impact on “Mr. A ’s recommendation about soccer” Also find the degree feature score.

  After the update of the impact score for “Recommended by Mr. A” is completed, the characteristics related to the impact level are also applied in the same way as above for those who have been given a “recommended tag” other than Mr. A included in the same list browsing content. The score is calculated, and the influence score is updated using the above formula (1).

  According to the second embodiment, “recommended tag” is assigned to the content to which “recommended tag” is assigned, and the “recommended tag” is assigned at the same time as the concept is added. The degree of influence of each concept on other users can be estimated. In other words, the influence on the person is not only expressed by one score, but also the difference in the degree of influence depending on the genre (for example, Mr. AA is affected by the genre X but not the genre Y). It is also possible to grasp.

(4-3) Third embodiment
FIG. 11 is a diagram illustrating an influence analysis process according to the third embodiment. In the third embodiment, a method for estimating the degree of influence from group F on Mr. AA will be described. In other words, when other users who have given recommendation information have group attributes and organizational systems, and can create a hierarchical system for these relationships (hereinafter, the system consisting of people mentioned above is referred to as a human system), browse the list. The upper group attributes of other users who have added recommendation tags that appear in the contents of the content list and the detailed browsing content list are extracted from the human system, and the upper group attributes are considered to appear in the content. Further, the present invention is characterized in that the user's influence degree score for is updated. Note that the human system relationship information can have a graph structure when a number of other users or group attributes are related to each other. In the case of having a graph structure, the direct or indirect superior attribute of each other or group attribute is referred to as a “parent group”, and the direct or indirect superior attribute is referred to as a “child group” Considered a group).

The human system database 220 receives a list of recommended tag granters included in the list browsing content list (in this example, “A-san”, “C-san”, “D-san”, and “E-san”) and received history information. Received from the unit 210 and stored. The history information receiving unit 210 reads the human system (group or attribute, etc.) higher than the recommended tag grantor from the human system database 220 (in this example, “group F”, “group G”, and “group H”). ). The history information receiving unit 210 adds the parent group read from the human system database 220 and passes the recommended tag information to the influence characteristic score calculating unit 230.

  The influence characteristic score calculator 230 receives the recommended tag information to which the group information is added from the history information receiver 210. The characteristic score calculation unit 230 of the influence degree uses the same method as in the first embodiment, and the characteristic of the influence degree regarding “Recommended by Mr. A”, “Recommended by Mr. C”, “Recommended by Mr. D” and “Recommended by Mr. E”. Update the score.

  Further, an influence score for the parent group is obtained. As a specific example, a method of obtaining the influence score of “Group F” is shown. First, contents including recommended tags of individuals “Mr. A”, “Mr. B”, and “Mr. C” belonging to “Group F” (in this example, “Mr. A” and “Mr. C” are recommended) Is regarded as “Group F recommendation”, and the characteristic score of the influence degree is obtained in the same manner as described above using the total number of contents including recommended tags of individuals belonging to “Group F”, and the above equation (1) is obtained. Use to update the impact score.

  When the number of “Group F recommendation” that appears in the list browsing content (S) is R, the detailed browsing content in which “Group F recommendation” appears when the detailed browsing content (a) is selected is r. It shall be a piece. At this time, the influence characteristic score calculator 230 calculates the influence characteristic score Z ′ for “group F” using the analysis parameters S, a, R, and r. The characteristic score Z ′ of the influence degree in this case can be calculated by replacing N and n with R and r in the calculation formula shown in FIG.

  In addition, regarding child groups that belong to the lower level of the parent group to be learned (in “Group F”, “Mr. A”, “Mr. B”, “Mr. C”), the child groups that are not included in the list browsing content list (For “Group F” in this example, “Mr. B”), the influence score is updated using the characteristic score of “Group F”.

  The updated impact score is stored in the root concept of the impact concept system of each person and group in the impact concept system database 250 having the impact score for each person / group. And the score of the root concept is given to the subordinate concepts corresponding to the descendants with the same value.

(4-4) Fourth embodiment
FIG. 12 is a diagram illustrating an influence analysis process according to the fourth embodiment. The human system database 220 is the same as that in the third embodiment. The feature score calculation unit 230 of the influence degree uses the same method as in the second embodiment for the recommended tag received from the human system database 220, and for each genre of the person (or group) to which the recommended tag is assigned. Update impact score.

(5) Reliability Analysis Device The reliability analysis device 600 includes a history information receiving unit 610, a reliability calculation source database 620, a reliability coefficient calculation unit 630, and a reliability coefficient database 640.
The history information receiving unit 610 receives the total number N of information corresponding to the reliability coefficient calculation target from the content server 500, and also displays the list browsing content list and the detailed browsing content list generated by the client terminals 400-1 and 400-2. Is received from the history information transfer unit 550 of the content server 500. The total number N of the information corresponding to the received reliability coefficient calculation target and the total number of information displayed to the target user corresponding to the reliability coefficient calculation target obtained from the list browsing content list and the detailed browsing content list. The reliability value is obtained by using the value n and the cumulative value m of the action history (for example, detailed page browsing history and bookmark registration history) related to the feature of the user's selection action corresponding to the calculation target of the reliability coefficient as reliability calculation source data. Store in the calculation source database 620.

  The reliability coefficient calculation unit 630 reads the reliability calculation source data N, n, m from the reliability calculation source database 620 for each user and for each reliability detection target (for example, a gourmet genre such as curry or beef bowl). Then, using some or all of the read reliability calculation source data N, n, m, the reliability coefficient T for each reliability detection target (for example, gourmet genre such as curry or beef bowl) for each user. Is calculated. The calculation method will be described in detail later.

  The reliability coefficient database 640 is used to store the reliability coefficient T calculated by the reliability coefficient calculation unit 630 for each reliability detection target (for example, gourmet genre such as curry and beef bowl) for each user.

Next, the operation of the reliability analysis apparatus 600 will be described.
First, in the history information receiving unit 610, the total number N of information corresponding to the reliability coefficient calculation target is acquired from the history information transfer unit 550 via the communication network. Here, the information corresponding to the reliability coefficient calculation target is, for example, a gourmet genre such as curry or gyudon, a news category or conceptual term such as economy, appreciation of the yen, or baseball, or a recommended tag such as A, B, or X group. The total number N includes topics such as the total number of stores, the total number of applicable news, the total number of recommended tags, and the like. The acquired total number N is stored in the reliability calculation source database 620.

  In the history information receiving unit 610, the list browsing content list and the detailed browsing content list generated by the client terminals 400-1 and 400-2 are acquired from the history information transfer unit 550 of the content server 500. Then, based on these histories, the cumulative value n of the number of information displayed to the target user corresponding to the reliability coefficient calculation target, and the behavior related to the feature of the user's selection behavior corresponding to the reliability coefficient calculation target A cumulative value m of the history (for example, detailed page browsing history, bookmark registration history) is calculated, and the calculated cumulative value n of the display number and the cumulative value m of the action history are associated with the user ID and the reliability calculation source database 620. Stored in

As shown in FIG. 13, the reliability coefficient calculation unit 630 first reads the N, n, m from the reliability calculation source database 620 in step S11, and uses the read N, n, m to read the reliability coefficient. T is calculated (step S12). In short, the reliability coefficient T indicates how much the response rate p is different from the response rate p ₁ obtained from the population N when the response rate p is obtained from the sample. It is the reciprocal of the index (referred to as sampling error) to be expressed, and has a characteristic that the value increases as the error decreases.

The following three embodiments can be considered as a method of calculating the reliability coefficient.
(5-1) First Example A step function defined by a predetermined threshold value k is prepared. The formula of this step function is shown below.

  Then, n is compared with the threshold value k, and the reliability coefficient is T = 0 until n reaches the threshold value k, and the reliability coefficient T = 1 when n is equal to or greater than the threshold value k. . FIG. 14 shows the step function as a graph.

ただし、ｐ＝０のとき、ｐ＝１／（ｎ＋１）とし、ｐ＝１のとき、ｐ＝ｎ／（ｎ＋１）とする。これにより、ｎが０の場合はｐ=０．５とし、誤差最大となり、ｎが大きくなるに従い、ｐ＝０あるいはｐ＝１に近づくため、妥当な結果を得る。 (5-2) Second Example FIG. 15 is a diagram for explaining a second example of the reliability coefficient calculation method. For example, let N be the total number of curry store registrations, n be the total number of curry store information displayed to the user, m be the total number of visits to the detailed page related to the curry store by the user, Let p be the percentage of browsing. In this case, the reliability coefficient calculation unit 630 calculates the reliability coefficient T by the following equations (3-1) and (3-2).

However, when p = 0, p = 1 / (n + 1), and when p = 1, p = n / (n + 1). As a result, when n is 0, p = 0.5, the error is maximized, and as n increases, p = 0 or p = 1 is approached, so a reasonable result is obtained.

These equations (3-1) and (3-2) are used in the case of estimating the population N when extracting the sample number n when the response rate p is known in the random sampling used in the sample survey or the like. This is based on a statistical method for obtaining a sampling error. E is a sampling error and represents how much the response rate p is different from the response rate p ₁ obtained from the population N when the response rate p is obtained from the sample.

  Using equation (3-1), the reliability considering the size of the population N can be obtained. At this time, the response rate p may be calculated as p = m / n as shown in the equation (3-2), or may be 0.5 which maximizes the sampling error. The value of may be used. The coefficient 1.96 in equation (3-1) is 95% confidence interval, that is, a coefficient assuming a standard normal distribution with a probability of 95% in the population N. This coefficient is changed based on the confidence interval. Is possible.

FIG. 16 shows a graph when the horizontal axis is the number of samples n, the vertical axis is the sampling error E or 1-E, and N = 100 and p = 0.5. As shown in FIG. 16, the reliability coefficient 1-E becomes a trend similar to the case where the response rate p ₁ the larger the value of the sample number n to obtain a response rate p ₂ from a population N, response rate p ₁ trust It turns out that it is a possible value.

(5-3) Third Example This third example uses the reliability coefficient calculation method described in the first and second examples together.
That is, the reliability coefficient until 1-E reaches the threshold value k is T = 0, and when it is equal to or higher than the threshold value k, the reliability coefficient T = 1-E. This is expressed by the following formula (4).

  FIG. 17 shows a graph represented by the formula (4). However, this graph illustrates the case where the population N = 100 and the response rate p = 0.5. As shown in this graph, the reliability coefficient T becomes 0 until a sampling error E below a certain standard is secured.

  When the second example was used, even when the interest score with low reliability was kept low, the learning progressed based on the list displayed for the first time, so it was not displayed first in the list As for things, even if the user “likes” them, there remains a problem that they are not displayed at the top, but when using the third embodiment, various learning progresses at the initial stage of learning, and Compared with the reliability coefficient calculation method shown in the first embodiment, the learning result with a small number of learning times is less likely to be reflected in the recommendation, so the reliability is more accurate than the methods shown in the first and second embodiments. A degree coefficient T can be calculated.

  The calculation of the reliability coefficient T is performed for each user and for each reliability coefficient calculation target (for example, gourmet genre, news category or conceptual term, recommendation tag, or topic). The calculated reliability coefficient T is stored in the reliability coefficient database 640 in association with the user ID and the reliability coefficient calculation target.

(6) Content Scoring Device The content scoring device 300 includes a sorted content list transmission unit 310, a content evaluation processing unit 320, a content database 330, and a content / recommendation tag reception unit 340.

  The content scoring apparatus 300 receives the content to be presented from the content list transmission unit 530, scores the content, sorts the content based on the score, and sends the sorted content list to the sorted content list reception unit 520. hand over.

Specifically, the following processing is performed.
(6-1) When content evaluation is performed based on the interest score and the influence score
FIG. 18 is a diagram showing an outline of processing when content evaluation is performed based on the interest score and the influence score.
The content / recommended tag receiving unit 340 receives the presentation target content, the concept indicating the content of the presentation target content, and the “recommended tag” given to the presentation target content from the content list transmission unit 530 by the content database. Stored in 330.

Next, the content evaluation processing unit 320 tags each presentation target content (for example, evaluation content (evaluation content 1 has “concept H”, “concept E”, “recommended A”), and “recommended B”). )), The following formula using the “concept H” score from the influence concept system (“Mr. A” and “Mr. B” influence concept system) from the person who gave the recommended tag of the content (5) The evaluation impact score is obtained in the same manner as for “concept E”.

Thereafter, using the scores of “concept H” and “concept E” in the concept system having the interest score, the interest score for evaluation is obtained by the following equation (6).

Then, using the evaluation impact score and the evaluation interest score described above, the evaluation impact score and the evaluation interest score are superimposed by the following equation (7) to obtain the evaluation score of the evaluation content 1.

  Similarly, evaluation scores are obtained for all the presentation target contents, and the contents are sorted based on the calculated evaluation scores. Then, the sorted content is passed to the sorted content list transmission unit 310. Sorted content list transmission unit 310 transmits the sorted content list to sorted content list reception unit 520.

  The sorted content list receiving unit 520 sends the sorted content to the content transmission processing unit 510 based on the sorted content list, and transmits it to the content presentation unit 430 of the client terminal 400-1.

  As described above, by taking into account the user's own interests and the influence of other users on the user, it is possible to estimate the degree of influence from other users when browsing the content with high accuracy. This makes it possible to realize an information recommendation service that is close to the actual social life and selection behavior for the user.

(6-2) When content evaluation is performed by adding a reliability coefficient to the interest score or influence score
In step S13 of FIG. 13, the content scoring device 300 first acquires an interest score or an influence score from the interest analysis device 100 or the influence analysis device 200 based on the ID of the user to be evaluated. In addition, a display target list for the user is acquired from the content database 330, and a reliability coefficient corresponding to an evaluation target item such as a gourmet genre or a conceptual term included in the display target list is acquired from the reliability coefficient database 640. Then, the acquired interest score or influence score, display target list, and reliability coefficient of the evaluation target item are sent to the content evaluation processing unit 320.

The content evaluation processing unit 320 scores the content by multiplying the user interest score or the influence score by the reliability coefficient for all the content included in the display target list. Equation (8) shows a formula for calculating the content evaluation score when the user interest score is multiplied by the reliability coefficient.

  In Equation (8), Ti represents a reliability coefficient, totalZi represents a user interest score of each genre calculated for each user, and wi represents a weight of the genre indicating each content. For example, when the content is represented by a plurality of genres (or topics or concepts), the weight w i is determined by the strength of the relevance between each genre i (or topic or concept) and the content. The weight wi may be arbitrarily set by a service provider or the like.

  Equation (8) shows a calculation formula when the user interest score is used, but an influence score may be used instead of the user interest score. In this case, the content may be scored by the above equation (8) using the influence score and the reliability coefficient Ti shown on the right side of the number (5) described above.

(7) Sorting contents and outputting the result The sorted content list transmission unit 310 performs content sorting and sorted content list output processing by one of the following two methods.

(7-1) First Example The sorted content list transmission unit 310 sorts the content information based on the content score received from the content evaluation processing unit 320, and sorts the sorted result in the order of the content score. It transmits to the content server 500 as a content list.

  The content server 500 receives the sorted content list transmitted from the content scoring device 300 by the sorted content list receiving unit 520, and receives the received sorted content list from the content transmission processing unit 510 to the client of the corresponding user. Transfer to terminals 400-1 and 400-2. As a result, the sorted content list is displayed on the content presentation unit 430 of the client terminals 400-1 and 400-2.

(7-2) Second Example As shown in FIG. 20, the sorted content list transmission unit 310 has a service user out of the display list or the display screen area of the client terminals 400-1 and 400-2 of the user. Arbitrarily determined portion S is assigned as a sorted content display region, and the remaining 1-S portion is assigned as a content display region for which interest estimation has not been performed. And it arrange | positions so that the result sorted based on the content score received from the content evaluation process part 320 may be displayed on the said S part. Further, in the 1-S portion, content including evaluation item targets for which interest estimation is not performed is arranged to be displayed at random.

(8) Specific Examples of Reliability Coefficient Calculation Processing and Content Evaluation Processing Here, the gourmet genre is handled as the reliability coefficient calculation target, and the third implementation of the reliability coefficient calculation method described in (5-3) above. A case where the reliability coefficient T is calculated using an example will be described as an example.

  Now, for example, as shown in FIGS. 19A, 19B, and 19C in the reliability calculation source database 620, the total number N of registered stores for each gourmet genre and the number of stores displayed for each gourmet genre displayed to an arbitrary user. , And the total number m of the number of viewings of the store detailed information for each gourmet genre by the user is stored. Further, the conceptual system / user interest score database 140 stores the user interest score shown in FIG. 19 (e), and the content database 330 shows the correspondence between the store name shown in FIG. 19 (f) and its gourmet genre. It is assumed that data to represent is stored.

  The reliability coefficient calculation unit 630 of the reliability analysis apparatus 600 includes the total number N of stores corresponding to each gourmet genre shown in FIG. 19A from the reliability calculation source database 620 and each of the items shown in FIG. The total number n of stores displayed for the user of interest in the store corresponding to the gourmet genre and the data m regarding the total detailed browsing history of stores corresponding to each gourmet genre shown in FIG. Then, the reliability coefficient T is calculated using the reliability coefficient calculation formula (4) described in (5-3) above.

If k = 5, the reliability coefficient T is calculated by the following equation.

For example, when calculating the reliability coefficient for the user A for each gourmet genre, the total number of applicable stores N for each gourmet genre (“curry”, “gyudon”, “Okinawa cuisine”, “sushi”, “ramen”) By substituting into the formula (9) the total number of stores displayed for each gourmet genre, the total number of stores displayed, and the total detailed browsing history of stores corresponding to each gourmet genre, the following calculation is performed: Is done.

  The calculated reliability coefficient T for each gourmet genre related to the user A is stored in the reliability coefficient database 640. If the previously calculated value is already stored at this time, the past value is updated to the newly calculated value.

  The content evaluation processing unit 320 of the content scoring device 300, as shown in FIG. 19 (f), from the content database 330, the gourmet genres “curry”, “gyudon”, “gyudon”, “Okinawa cuisine”, Read out “sushi”, “curry”, “ramen”, “ramen”. Then, the content evaluation processing unit 320 inquires the reliability coefficient database 640 for the reliability coefficient corresponding to the read gourmet genre, and T (curry, Mr. A) = 0.63 calculated by the above equation (10). T (beef bowl, Mr. A) = 0.73, T (Okinawa cuisine, Mr. A) = 0, T (sushi, Mr. A) = 0, T (ramen, Mr. A) = 0.70 are read out.

なお、この実施形態では各コンテンツは１つのジャンルからなるため、i=1であり、また各重みは１とする。 Then, an evaluation score is calculated for each content. In this embodiment, the following calculation is performed using the previously described equation (8).

In this embodiment, since each content consists of one genre, i = 1 and each weight is 1.

  The sorted content list transmission unit 310 sorts the content based on the calculated content score, and the sorted content list is sent to the client terminal of the user A via the content server 500. It is displayed on the presentation unit 430. FIG. 21 shows an example of the display result. As shown in the figure, content with a higher score is displayed above the list, and content with a lower score is displayed below the list. In FIG. 21, the gourmet genre is shown in association with the content.

  As described above in detail, in this embodiment, a reliability coefficient for the estimation result of the user's selection behavior is calculated, and a content score is calculated by multiplying the reliability coefficient by the user interest score. The display order of contents is sorted based on. For this reason, until the reliability of an estimation result is ensured, the learning opportunity of another estimation object can be ensured, and mislearning can be reduced. In addition, it is possible to reduce errors in list display due to erroneous evaluation of contents.

  Furthermore, in this embodiment, a display area for displaying a randomly selected list without using the calculated content score is added to the sorted content list. For this reason, even when learning has progressed to some extent, it is possible to continuously present unlearned evaluation item targets to the user and learn a wide range of interests.

  The present invention is not limited to the above embodiment. For example, in the embodiment, when calculating the reliability coefficient, a threshold value k is set for the total value n of the number of information displayed to the target user corresponding to the reliability coefficient calculation target. The reliability coefficient was calculated using the step function defined by However, the present invention is not limited to this. A threshold is set for the reliability (1-E) shown on the vertical axis in FIG. 16, and the reliability is determined using a step function defined by this threshold. A coefficient may be calculated. For example, when the reliability (1-E) is less than 0.5, it is set to 0, and when it is 0.5 or more, the reliability is calculated using the curve shown in FIG. That is, an estimation result whose reliability is not guaranteed is regarded as zero, and when a certain level of reliability is obtained, the reliability is considered according to the value.

  In FIG. 2, the concept system / user interest score database 140 is included in the interest analysis apparatus 100, the human system database 200 and the influence concept system database 250 are included in the impact analysis apparatus 200, and the content database is stored in the content scoring apparatus 300. 330 has been described as an example in which the reliability calculation source database 620 and the reliability coefficient database 640 are provided in the reliability analysis apparatus 600. These databases are not included in each analysis apparatus and the content scoring apparatus 300. You may make it provide.

  Further, the information sender is not limited to a user of the same system, and may be a blog author, for example.

  In short, the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. Further, various inventions can be formed by appropriately combining a plurality of constituent elements disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, you may combine suitably the component covering different embodiment.

  DESCRIPTION OF SYMBOLS 100 ... Interest analysis apparatus, 110 ... History information receiving part, 120 ... Interest characteristic score calculation part, 130 ... Interest concept system update process part, 140 ... Concept system / user interest score database, 200 ... Influence analysis apparatus, 210 ... History information receiving unit, 220 ... Human system database, 230 ... Feature score calculation unit of influence degree, 240 ... Concept system update processing unit of influence degree for each person / group, 250 ... Influence having an influence degree score for each person / group Degree conceptual system database, 300 ... content scoring device, 310 ... sorted content list transmission unit, 320 ... content evaluation processing unit, 330 ... content database, 340 ... content / recommended tag reception unit, 400 ... client terminal, 410 ... history Collection unit, 420 ... history information transmission unit, 430 ... con Content presentation unit, 440 ... content request transmission unit, 500 ... content server, 510 ... content transmission processing unit, 520 ... sorted content list reception unit, 530 ... content list transmission unit, 540 ... content list input unit, 550 ... history information Transfer unit, 560 ... Content request transfer unit, 600 ... Reliability analysis device, 610 ... History information reception unit, 620 ... Reliability calculation source database, 630 ... Reliability coefficient calculation unit, 640 ... Reliability coefficient database, 700 ... display Target database

Claims (9)

By the computer, an information recommendation method of presenting select the information to be recommended in hot water over The,
Calculating information indicating the degree of interest of the user with respect to the information to be presented from the list browsing content list and the detailed browsing content list of the user;
Information indicating the degree of influence from other users when the user takes action based on the information to be presented from the list browsing content list, detailed browsing content list, and recommended information from other users. The process of calculating,
Based on the presentation history of the information to be presented to the user or the behavior history of the user based on the information to be presented for each of the information to be presented or the set thereof, the estimation result of the feature of the user's behavior Calculating information representing the reliability of
Based with both information representing the degree of information and the degree of influence representing the degree of interest of the user to the presentation target information, and information indicating the calculated reliability, the information or the presentation object A process of calculating evaluation points for the set;
And a step of generating a list of information to be recommended to the user based on the calculated evaluation score. The process of calculating the information indicating the reliability includes:
The total number N of information corresponding to the reliability calculation target, the cumulative value n of the information presented to the user among the information corresponding to the reliability calculation target, and the information corresponding to the reliability calculation target A process of obtaining a cumulative value m of user behavior;
The method according to claim 1, further comprising: calculating a reliability coefficient based on at least one of the total number N of the acquired information, the cumulative value n of the presentation information, and the cumulative value m of the action. Information recommendation method.   The process of generating the list includes a list having a first region presenting recommended information selected based on the calculated evaluation score and a second region presenting arbitrarily selected recommendation information. The information recommendation method according to claim 1, wherein the information recommendation method is generated. An information recommendation device to select the information to be recommended in hot water over The presentation,
Means for calculating information indicating a degree of interest of the user with respect to information to be presented from the list browsing content list and the detailed browsing content list of the user;
Information indicating the degree of influence from other users when the user takes action based on the information to be presented from the list browsing content list, detailed browsing content list, and recommended information from other users. Means for calculating;
Based on the presentation history of the information to be presented to the user or the behavior history of the user based on the information to be presented for each of the information to be presented or the set thereof, the estimation result of the feature of the user's behavior Means for calculating information representing the reliability of
Based on the information or its set of the presented objects and both information representing the degree of information and the degree of influence representing the degree of interest of the user, and information representing the calculated reliability for the presentation target information Means for calculating an evaluation score for
An information recommendation apparatus comprising: means for generating a list of information to be recommended to the user based on the calculated evaluation score. The means for calculating the information indicating the reliability is:
The total number N of information corresponding to the reliability calculation target, the cumulative value n of the information presented to the user among the information corresponding to the reliability calculation target, and the information corresponding to the reliability calculation target Means for obtaining a cumulative value m of user behavior;
5. The apparatus according to claim 4, further comprising means for calculating a reliability coefficient based on at least one of the total number N of the acquired information, the total value n of the presentation information, and the total value m of the action. Information recommendation device.   The means for generating the list includes a list having a first area for presenting recommended information selected based on the calculated evaluation score, and a second area for presenting arbitrarily selected recommendation information. 6. The information recommendation device according to claim 4, wherein the information recommendation device is generated.   An information recommendation method for estimating characteristics of a user's behavior by a computer and selecting and presenting information to be recommended to the user based on the estimation result,
  Calculating information indicating the degree of interest of the user with respect to the information to be presented from the list browsing content list and the detailed browsing content list of the user;
  Information indicating the degree of influence from other users when the user takes action based on the information to be presented from the list browsing content list and the detailed browsing content list of the user and recommended information from other users. The process of calculating,
  Based on the presentation history of the information to be presented to the user or the behavior history of the user based on the information to be presented for each of the information to be presented or the set thereof, the estimation result of the feature of the user's behavior Calculating information representing the reliability of
  Based on information representing the degree of influence from the other user and information representing the calculated reliability, a process of calculating an evaluation score for the information to be presented or a set thereof,
  Generating a list of information to be recommended to the user based on the calculated evaluation score;
An information recommendation method comprising:   An information recommendation device for selecting and presenting information to be recommended to a user,
  Means for calculating information indicating a degree of interest of the user with respect to information to be presented from the list browsing content list and the detailed browsing content list of the user;
  Information indicating the degree of influence from other users when the user takes action based on the information to be presented from the list browsing content list and the detailed browsing content list of the user and recommended information from other users. Means for calculating;
  Based on the presentation history of the information to be presented to the user or the behavior history of the user based on the information to be presented for each of the information to be presented or the set thereof, the estimation result of the feature of the user's behavior Means for calculating information representing the reliability of
  Means for calculating an evaluation score for the information to be presented or a set thereof based on information representing the degree of influence from the other user and information representing the calculated reliability;
  Means for generating a list of information to be recommended to the user based on the calculated evaluation score;
An information recommendation device comprising:   The program which makes the said computer perform the process in each process with which the information recommendation method in any one of Claims 1, 2, 3, and 7 is provided.

Images