JP2001331516A - Data analyzing method, information retrieving method, and information recommending method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To solve the problem that arrangement representing latent features of data well can not be obtained as samples and categories become more since all samples and categories are arranged in one space of small dimensions for conventional analysis by class-3 quantification. SOLUTION: While samples having similar trends of interest in categories are grouped into several feature spaces of small dimensions, samples and categories having high relativity are arranged in the respective feature spaces closely in euclidean distance.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for analyzing data relating to personal hobbies, preferences, purchasing behavior, and the like.

[0002]

2. Description of the Related Art Conventionally, as a method of analyzing data on personal hobbies, preferences, and purchasing behavior obtained through a questionnaire or the like, three types of quantification (eg, Keigaku Publishing, Introduction to Multivariate Analysis, Chapter 8) are known. Have been. In quantification type 3, for example, in order to analyze what kind of person performs what kind of purchasing behavior, a category related to various purchasing behaviors of an individual (for example, “place” is a supermarket or a specialty store, Product knowledge ”is rich or poor,“ brand ”oriented or not)
Response patterns, and place individuals and categories in a multidimensional space so as to maximize the correlation between individuals (samples) and categories. Preference for stores. "

[0003]

However, in quantification class 3, the analysis of the arranged samples and categories is performed by first using the first eigenvector corresponding to the maximum eigenvalue.
The analysis is performed by the axis, and then the analysis is performed by the second axis. That is, since the proximity of a sample or category in a multidimensional space is not represented as a Euclidean distance, it is difficult to intuitively understand the relationship.
In addition, as the number of samples and categories increases, it becomes difficult not only to perform eigenvalue calculations for the matrices needed to determine their placement, but also to place all samples and categories in one multidimensional space, There has been a problem that an arrangement that well represents a latent feature may not be obtained.

[0004] In view of the above, the present invention arranges highly relevant samples and categories in the sense of the Euclidean distance, and at the same time, clusters the samples and categories into a plurality of groups. It is an object of the present invention to provide a data analysis method which can express a plurality of multidimensional spaces, which can express intuitively and facilitate intuitive understanding of closeness.

[0005]

According to a first aspect of the present invention, there is provided an individual / category response pattern storage means for storing a response to a category of a sample (sample / category response pattern), and a distance between the sample and the category. A sample-to-category distance storage means for storing a sample-to-category distance, a sample-to-sample distance storage means for storing the distance between the samples (inter-sample distance), and a distance between the categories (inter-category distance). Inter-category distance storage means, feature space membership degree storage means for storing the degree of membership of the sample or the category in a feature space representing the correlation between the sample and the category, and the feature space of the sample and the category Means for storing coordinates in a feature space for storing coordinates in the sample space; Learning means for learning the degree of belonging of the sample or category to the feature space based on the distance between the sample categories, the distance between the samples, and the distance between the categories; a sample / category reaction pattern; A feature space coordinate learning means for learning coordinates of the sample and the category in the feature space based on the inter-category distance, the inter-sample distance, and the inter-category distance; A data analysis method comprising learning control means for controlling the start and end of the operation of the coordinate learning means.

According to the present invention, there is provided a user attribute-corresponding content information managing means for managing a user attribute and content related information used by a user in association with each other, and a response pattern of the content (sample) to the user attribute (category). Sample / category reaction pattern storage means for storing (content / user reaction pattern); sample / category distance storage means for storing a distance between the content and the user attribute (content / user attribute distance); An inter-sample distance storage means for storing a distance between contents (inter-content distance), an inter-category distance storage means for storing a distance between the user attributes (distance between user attributes), and a correlation between the content and the user attributes. Storing the degree of belonging of the content or the user attribute to the feature space to be represented Feature space belonging degree storage means, feature space coordinate storage means for storing coordinates of the content and the user attribute in the feature space, the content / user response pattern, the content / user attribute distance, and the content A feature space belonging degree learning means for learning the degree of belonging of the content or the user attribute to the feature space based on the inter-user distance and the inter-user attribute distance, the content / user response pattern, the content / user attribute distance, A feature space coordinate learning unit for learning coordinates of the content and the user attribute in the feature space based on the content distance and the user attribute distance, the feature space belonging degree learning unit, and the feature space coordinate learning unit Learning control means for controlling the start and end of the operation of the feature space; The correlation between the content and the user attribute is displayed by referring to the content or user attribute belonging degree stored in the coordinate space of the content and the user attribute stored in the feature space coordinate storage unit, and displaying the correlation between the content and the user attribute. An information search method characterized by comprising a user attribute / content correlation display type information search interface providing means for causing a requester to search for content.

According to the present invention, there is provided a user-specific use content information managing means for managing content-related information used for each user in association with each other, and a reaction pattern (user / user) for a user (sample) to content (category). Sample / category reaction pattern storage means for storing a content reaction pattern), sample / category distance storage means for storing a distance between the user and the content attribute (distance between user and content), and a distance between the users ( Inter-sample distance storage means for storing the distance between the users (inter-user distance), inter-category distance storage means for storing the distance between the contents (inter-content distance), and the user to a feature space representing the correlation between the user and the content. Or a feature space belonging degree storage means for storing the belonging degree of the content attribute; A coordinate storage unit in a feature space for storing coordinates of the user and the content attribute in the feature space; and a user content response pattern, the user-content attribute distance, the user distance, and the content distance. Feature space belonging degree learning means for learning the degree of belonging of the user or the content to the feature space based on the user / content response pattern, the user / content distance, the user distance, and the content distance. Learning means for learning coordinates in a feature space of a user and content in the feature space, learning control means for controlling start / end of the operation of the feature space belonging degree learning means and the coordinate learning means in the feature space, The user or content belonging degree stored in the feature space belonging degree storage means and Use content similar user selection means for selecting the users located within a predetermined distance range from each other in the feature space with reference to the coordinate values of the user and the content attribute stored in the feature space coordinate storage means And a recommendation information selection providing means for recommending each other's use contents to the users whose use contents are similar to each other.

[0008]

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

(Embodiment 1) FIG. 1 is a configuration diagram of an information recommendation apparatus showing a first embodiment. In FIG.
101 is a response to a sample category (sample
An individual / category response pattern storage means for storing a category reaction pattern), a sample / category distance storage means for storing a distance between the sample and the category (a sample / category distance), and 103 a distance between the samples. (Inter-sample distance) storing means, inter-sample distance storing means for storing the inter-category distance (inter-category distance), and 105 to a feature space representing a correlation between the sample and the category. Is a feature space belonging degree storing means for storing the degree of belonging of the sample or the category, 106 is a feature space coordinate storing means for storing coordinates of the sample and the category in the feature space, and 107 is the sample category. Reaction pattern, distance between sample categories, distance between samples, and The feature space belonging degree learning means 108 for learning the degree of belonging of the sample or category to the feature space based on the distance between categories, the sample / category reaction pattern, the distance between sample categories, the distance between samples, and the category A feature space coordinate learning unit 109 for learning the coordinates of the sample and the category in the feature space based on the distance controls the operation start / end of the feature space belonging degree learning unit 107 and the feature space coordinate learning unit 108. This is learning control means.

FIG. 2 shows a hardware configuration of the present embodiment in which the information recommendation device configured as described above is executed. FIG. 2 is basically the same as the configuration of a general-purpose computer system. In FIG. 2, reference numeral 1002 denotes a main storage device composed of a volatile memory for storing programs, 1003 denotes an auxiliary storage device composed of a nonvolatile memory for storing programs and data, and 1001 executes a program stored in the main storage device 1002. CPU Reference numeral 1004 denotes an interface for inputting analysis data such as a sample / category reaction pattern, which includes a monitor, a keyboard, a mouse, and the like. After the program stored in the auxiliary storage device 1003 is loaded into the main storage device 1002,
This is executed by U1001.

The operation of the first embodiment configured as described above will be described. In order to explain the operation of the first embodiment more clearly, here, an example of analyzing what kind of user likes what kind of homepage from the sample category reaction pattern shown in FIG. 3 will be described.
In this example, the sample is I users (user 1 to user I), and the category is J homepages (HP1 to HP
J). The numerical values in FIG. 3 are, for example, a score of the degree of interest of each user on each homepage and the number of times of browsing each homepage. In this case, in the analysis according to the present invention, as shown in FIG. 4, users having similar interests in the homepage are compared with the multidimensional data relating to the user and the homepage shown in FIG. (E.g., two-dimensional or three-dimensional), and at the same time, in each feature space, the homepages of interest and users, users with similar interests in homepages, and homepages with similar user reactions are identified. This is to arrange the objects closer to each other in the sense of the Euclidean distance, and as a result, it is possible to interpret what kind of user likes what kind of homepage from what viewpoint.

The operation of the data analysis method according to the first embodiment will be described below with reference to the flowchart of FIG.
The arrangement of the samples and categories in the K-dimensional feature space using the sample / category reaction pattern according to the present invention is realized by minimizing the following evaluation function Q of (Equation 1).

[0013]

(Equation 1)

Here, xij is the degree of reaction of sample i to category j, vcj is the cluster center vector of sample cluster c, uci is the degree of membership of sample i to cluster c, and dij is the distance between sample i and category j. , pcik is the coordinates in the feature space of sample i belonging to cluster c, qcjk is the coordinates in the feature space of sample j belonging to cluster c, pdis is the distance between sample i and sample s, qdjt is category j and category t Represents the distance. Q1 is an evaluation function to make the distance between sample i and category j closer to the sample-cluster distance dij in the feature space, and Q2 is an evaluation function to make the distance between sample i and s closer to the inter-sample distance pdis in the feature space. , Q3 are evaluation functions for bringing the distance between categories j and t closer to the inter-category distance qdjt in the feature space.

By minimizing Q based on the gradient method by the following procedure, the coordinates Pcik and qcjk of the sample and the category in the C K-dimensional feature spaces can be obtained.

[STEP 1a] Initialization C, K, pcik, qcjk, uci, vcj are initialized by the learning control means 109. c specifies the number of samples you want to group. K is the dimension of the feature space and is also specified. FIG.
In the case of, K = 3. pcik and qcjk are initialized by a random number of [0, 1] and stored in the feature space coordinate storage means 106. uci is 1 / C for all i. vcj is xij u
The weighted average by ci is Σuci · xij / Σuci. These are stored in the feature space membership degree storage means 106.

[STEP 2a] Update of feature space belonging degree uci The feature space belonging degree learning unit 107 updates the feature space belonging degree uc.
Update i. Here, in the case of learning by the steepest descent method, uci
Since u may be negative, uci is updated by the interior point method using affine transformation. The value of uci determines to what extent the sample belongs to which feature space.

[STEP 3a] Update of coordinates pcik, qcjk in feature space pcik, qcjk is updated by (Equation 2) by the feature space coordinate learning means 108. The updated coordinates are stored in the feature space coordinate storage unit 106.

[0019]

(Equation 2)

[STEP 4a] Update of the cluster center vcj The learning control means 109 executes xij which is a solution of ∂Q / ∂vcj = 0.
Vcj by uci weighted average iuci · xij / Σuci
To update.

[STEP 5a] Learning end determination The learning control means 109 determines that max | uci (new) -uci (old) | <
Learning is completed if ε is satisfied. Others return to STEP2a.

By minimizing Q based on the gradient method according to the above procedure, the coordinates Pcik and qcjk of the sample and the category in the C K-dimensional feature spaces are obtained. As described above, in the first embodiment, samples having a similar tendency of interest in a category are grouped into several low-dimensional feature spaces. Since the arrangement is close to the meaning, it is possible to more easily intuitively grasp the relationship between the sample and the category.

Here, the samples are grouped into a plurality of feature spaces, but it goes without saying that the categories can be grouped and analyzed by transposing the sample / category reaction pattern.

(Embodiment 2) Next, a second embodiment according to the information recommendation method of the present invention will be described. In the first embodiment, samples (users) having similar interests in a category (homepage) are grouped into several low-dimensional feature spaces, and at the same time, in each feature space, a highly related user and a homepage are grouped. In the second embodiment, the analysis method is described in which the relations are arranged closer to each other in the sense of the Euclidean distance so that their relations can be grasped more intuitively. By displaying the correlation in the feature space to the information searcher, the information searcher can easily search for the content that is used by a person who is very similar to himself as a landmark using the attribute related to oneself in the feature space. It is considered as follows.

FIG. 6 is a block diagram of the information recommendation apparatus according to the second embodiment. The same components as those in the first embodiment shown in FIG. Is omitted. The difference between the second embodiment and the first embodiment is that a user attribute corresponding content information management means 201 for managing a user attribute and content related information used by a user in association with each other and the content and the user attribute And a user attribute / content correlation display type information search interface providing means 202 for displaying the correlation of the information and allowing the requester to search for the content.
That is, the second embodiment is, as shown in FIG. 6, an interface and a system for allowing a user to search for information (for example, a homepage) that meets a user's own attribute by joining a network, for example.

The operation of the second embodiment configured as described above will be described with reference to the flowchart of FIG. However, [STEP1a] to [STEP5a] of the flowchart of FIG. 5 representing the operation of the data analysis method of the first embodiment are the same as [STEP4b] to [STEP8b] of the flowchart of FIG. Only [STEP1b] to [STEP2b] and [STEP9b] will be described. In order to explain the operation of the present invention more clearly, here, an example will be described in which a user accesses a Web server using a Web browser mounted on the user terminal 2001 and searches for information. Usually, a user browses a directory or a recommended link of a homepage in Yahoo or MSN and searches for desired information. In the present embodiment, the user is caused to search for information using his / her attribute as a key.

[STEP 1b] Storage of user attribute-compatible content usage history
As shown in FIG. 6, through a user terminal connected over a network, for example, in a format as shown in FIG.
It collects and accumulates user attributes and information on the homepages that have been viewed (for example, the number of times viewed). For example, the user attribute may be input at the time of user registration when using the information search service in the system according to the present embodiment. The user (browser) can be specified using cookies or the like. The information provider detects cg in the information provider
It can be realized by i or JavaScript. The user attribute corresponding content information management means 201 stores this information for a predetermined time.

[STEP 2b] Determination of Elapse of Predetermined Time The user attribute corresponding content information management means 201
Time management.

If a predetermined time has passed since the information was collected by STEP 1b, the process proceeds to STEP 3b. Otherwise, the information collection by STEP1b is continued.

[STEP 3b] Related Distance Calculation The user attribute corresponding content information management means 201
From the correspondence table between user attributes and contents as shown in FIG. 8, data on the user attributes to be analyzed is summarized, and first, a content / user attribute reaction pattern is created. In this case, for example, when analyzing by the user attribute of occupation, it is performed by summing the numerical values included in the attribute for each content. Based on this content / user attribute response pattern, the distance between content / user attributes, the distance between contents, and the distance between user attributes are calculated. These are calculated as Euclidean distances.

[STEP 9b] Updating of information search interface At predetermined time intervals managed in STEP 2b, mapping of contents and user attributes in the feature space updated by the feature space belonging degree learning means 107 and the feature space coordinate learning means 108 is performed. indicate. This is realized, for example, as a home page stored in a Web server. If the feature space is two-dimensional, it is realized in a chart format, and if it is three-dimensional, it is realized in VRML or the like. The user can browse the map of the content and the user attribute through a browser mounted on the terminal 2001, and searches for information by using the user's own attribute as a key and referring to the content near the attribute. Information that can be searched.

As described above, in the second embodiment, the correspondence data between the user attribute and the content information obtained from many users is analyzed, and the correlation between the content and the user attribute is mapped to the low-dimensional space to provide the information searcher with the information. Since the information searcher presents the information, the information searcher can easily obtain useful information for the person only by selecting the contents arranged in the vicinity using his / her attribute as a key.

Here, the user attributes such as age, gender, occupation, etc., which do not identify individuals to some extent, are used as keys. However, for example, users may be directly arranged in the space using anonymity. In this case, the information of Mr. XX can be searched more directly and more finely, rather than being well suited to his or her preference.

Also, here, the case of browsing the map showing the correlation between the user attribute and the content has been described in connection with the registration system for further data collection.

(Embodiment 3) Next, a third embodiment according to the information recommendation method of the present invention will be described. In the second embodiment, the interrelationship between the user attribute and the used content is presented to the user as a low-dimensional map,
Although the description has been given of the example in which the information search is facilitated by using the attribute as a key, the third embodiment manages the usage content information for each user, and uses the usage information between users having similar usage content in the obtained feature space. Is distributed so that the information suitable for oneself can be automatically obtained without having to search for the information by oneself.

FIG. 9 is a diagram showing the configuration of an information recommendation apparatus according to a third embodiment. The same components as those in the second embodiment shown in FIG. Is omitted. The difference between the third embodiment and the second embodiment is that a user-specific usage content information management unit 301 that manages content-related information used for each user in association with each other, and a feature space membership degree storage. With reference to the degree of belonging of the user or the content stored in the means 107 and the coordinate values of the user and the content attribute stored in the coordinate storage means 108 in the feature space, they are arranged within a predetermined distance range from each other in the feature space. And a recommendation information selection providing unit 303 for mutually recommending each other the users selected by the used content similar user selecting unit 302. . That is, the third
In the embodiment, as shown in FIG. 9, for example, information (for example, a homepage) that meets a user through a network
Is a system for automatically distributing.

The operation of the third embodiment configured as described above will be described with reference to the flowchart of FIG. However, [STEP2b], [STEP4b] to [STEP8b] in the flowchart of FIG. 7 showing the operation of the information search method according to the second embodiment.
Are [STEP2c], [STEP4c] to [SEP2] in the flowchart of FIG.
TEP8c], so the differences are [STEP1c] ~
Only [STEP3c], [STEP9b] and [STEP10c] will be described. To better explain the operation of the present invention,
Here, an example will be described in which when a user accesses a Web server using a Web browser mounted on the user terminal 2001, information recommended to the person is linked to a predetermined homepage.

[STEP 1c] Storage of usage content information for each user The usage content information management means 301 for each user collects and manages the usage history of content for each user in a format as shown in FIG. The information to be collected may store not only the number of times of browsing but also the score of the content evaluated by the user. The collection management method is basically the same as in the second embodiment.

[STEP 3c] Calculation of Related Distance The user-specific used content information management means 301 calculates the user-content distance, the user distance, and the content distance based on the user-content reaction pattern shown in FIG. These are calculated as Euclidean distances.

[STEP 9c] Use Content Similar User Selection
In the mapping of the content and the user attribute in the feature space updated by the feature space belonging degree learning means 107 and the feature space coordinate learning means 108 at predetermined time intervals managed in 2b, users within a predetermined distance range from each other are identified. , A user whose usage content is similar is selected.

[STEP 10c] Mutual information recommendation of similar users The recommended information selection and provision means 303 mutually exchanges contents to be used between similar users. This recommended content may be provided, for example, in a form in which the user links to a predetermined homepage that can be accessed using a Web browser mounted on the user terminal 2001. Also, UR
Only L may be transmitted by e-mail or the like.

As described above, in the third embodiment, the usage content information is managed for each user, the users whose usage content is similar in the obtained feature space are specified, and the usage information is mutually distributed. You can automatically obtain information that suits you without having to search for the information.

In the present invention, the distance between users and the distance between contents are calculated from the user / content reaction pattern. However, this is not always necessary. When the content includes text information such as a homepage, a vector space model is used. The document similarity of may be converted to distance before use. As for the distance between users, the degree of coincidence such as age, occupation, and gender may be converted.

[0044]

According to the first embodiment, since highly related samples and categories are arranged close in the sense of the Euclidean distance and at the same time, the samples and categories are grouped into a plurality of feature spaces. Than in the case where all samples and categories are arranged in the same feature space as in the above, the potential features of the samples and categories can be well represented, and the intuitive understanding of the closeness can be facilitated. .

According to the second embodiment, the correspondence data between the user attribute and the content information obtained from many users is analyzed, and the correlation between the content and the user attribute is mapped to the low-dimensional space to provide the information searcher with the information. Since the information searcher presents the information, the information searcher can easily obtain useful information for the person only by selecting the contents arranged in the vicinity using his / her attribute as a key.

According to the third embodiment, in mapping of contents and user attributes in the feature space, users located within a predetermined distance from each other are selected as users whose usage contents are similar and information is mutually distributed. By doing so, it is possible to automatically obtain information suitable for oneself without having to search for information by oneself.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a data analysis method according to a first embodiment of the present invention.

FIG. 2 is a hardware configuration diagram according to the first embodiment of this invention;

FIG. 3 shows an example of a sample category reaction pattern.

FIG. 4 is a diagram showing an example of a data analysis result according to the first embodiment of the present invention;

FIG. 5 is a flowchart showing an operation procedure of the data analysis method according to the first embodiment of the present invention.

FIG. 6 is a configuration diagram of an information search method according to a second embodiment of the present invention.

FIG. 7 is a flowchart showing an operation procedure of the information search method according to the second embodiment of the present invention;

FIG. 8 is a diagram showing an example of a user attribute / content correspondence table.

FIG. 9 is a configuration diagram of an information recommendation method according to a third embodiment of the present invention.

FIG. 10 is a flowchart illustrating an operation procedure of the information recommendation method according to the third embodiment of this invention;

[Explanation of symbols]

Reference Signs List 101 sample / category reaction pattern storage means 102 sample / category distance storage means 103 sample-to-sample distance storage means 104 inter-category distance storage means 105 feature space belonging degree storage means 106 feature space coordinate storage means 107 feature space belonging degree learning means 108 Feature space coordinate learning means 109 Learning control means 201 User attribute corresponding content information management means 202 User attribute / content correlation display type information search interface providing means 301 User-specific usage content information management means 302 User content similar user selection means 303 Recommendation information Selection providing means 1001 CPU 1002 Main storage device 1003 Auxiliary storage device 1004 Interface 2001 User terminal 2002 Information providing server 3001 Information distribution server

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Hiroshi Kutsumi 1006 Kazuma Kadoma, Kadoma-shi, Osaka Matsushita Electric Industrial Co., Ltd. F-term (reference)

Claims (3)

[Claims] 1. A sample / category reaction pattern storage means for storing a reaction of a sample to a category (hereinafter, referred to as a sample / category reaction pattern), and a distance between the sample and the category (hereinafter referred to as a sample / category distance). ), And a distance between the samples (hereinafter, referred to as “sample / category distance storage means”).
Inter-sample distance storage means for storing the inter-sample distance), inter-category distance storage means for storing the distance between the categories (hereinafter, referred to as the inter-category distance), and the correlation between the sample and the category A feature space belonging degree storing means for storing the degree of belonging of the sample or the category to the feature space representing the; a feature space coordinate storing means for storing coordinates of the sample and the category in the feature space; Feature space belonging degree learning means for learning the degree of belonging of the sample or category to the feature space based on a sample / category reaction pattern, the distance between sample categories, the distance between samples, and the distance between categories; Reaction pattern, distance between sample categories, distance between samples, and distance between categories Learning means for learning the coordinates of the sample and the category in the feature space based on the information, learning control for controlling the operation start / end of the feature space belonging degree learning means and the coordinate learning means in the feature space A data analysis method characterized by comprising means. 2. A user attribute-corresponding content information managing means for managing a user attribute in association with content-related information used by a user, and a response pattern to a user attribute of the content (hereinafter referred to as a content / user response pattern). A sample / category reaction pattern storage means for storing; a sample / category distance storage means for storing a distance between the content and the user attribute (hereinafter referred to as a content / user attribute distance); and a distance between the contents. (Hereinafter, referred to as inter-content distance), inter-sample distance storage means, which stores a distance between the user attributes (hereinafter, referred to as user inter-attribute distance), The content or the user attribute into a feature space representing a correlation of user attributes Characteristic space belonging degree storage means for storing the degree of belonging of the content, coordinate storage means in the characteristic space for storing the coordinates of the content and the user attribute in the characteristic space, the content / user reaction pattern and the content user Feature space belonging degree learning means for learning the degree of belonging of the content or the user attribute to the feature space based on the attribute distance, the content distance, and the user attribute distance, and the content / user response pattern and the content / user response pattern. A feature space coordinate learning unit for learning coordinates of the content and the user attribute in the feature space based on the user attribute distance, the content distance, and the user attribute distance; Learning control means for controlling operation start / end of the coordinate learning means in the feature space; The correlation between the content and the user attribute with reference to the content or user attribute belonging degree stored in the inter-attribution degree storage means and the content and user attribute coordinate values stored in the feature space coordinate storage means. An information search method characterized by comprising a user attribute / content correlation display type information search interface providing means for displaying a user attribute and a content search by the requester. 3. A user-specific use content information management means for managing content-related information used for each user in association with each other, and a response pattern to a user's content (hereinafter, referred to as a content response pattern).
Sample / category response pattern storage means for storing user / content response patterns, and a distance between the user and the content attributes (hereinafter, user / content response pattern).
Sample-category distance storage means for storing a distance between contents (hereinafter referred to as content-to-content distance); inter-sample distance storage means for storing the distance between the users (hereinafter referred to as a user-to-user distance); (Hereinafter referred to as an inter-content distance), and an inter-category distance storage means for storing the degree of belonging of the user or the content attribute to a feature space representing a correlation between the user and the content. A feature space coordinate storage unit for storing coordinates of the user and the content attribute in the feature space; a user content response pattern, the user-content attribute distance, the user distance, and the content Feature space return for learning the degree of belonging of the user or content to the feature space based on distance Degree learning means for learning coordinates in the feature space of the user and the content based on the user-content response pattern, the user-content distance, the user distance, and the content distance. Means, learning control means for controlling the operation start / end of the feature space belonging degree learning means and the coordinate learning means in the feature space, and the degree of belonging of the user or the content stored in the feature space belonging degree storage means. Use content similar user selection means for selecting the users located within a predetermined distance range from each other in the feature space with reference to the coordinate values of the user and the content attribute stored in the feature space coordinate storage means And recommending each other's usage contents to the selected users similar to the usage contents Information recommendation method characterized by comprising Como information selection providing means.