JP2002366838A - Recommendation method, server system, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a server system which makes it possible to recommend new data. SOLUTION: A filtering part 13 gives a static recommendation by using the user profile (16) of an object user. A data mining part 14 collates a relation rule (18) extracted from an access log (17) accumulated as to many users with the access history (17) of the object user to give a dynamic recommendation. A promotion part 15 collates a page relation table (9) showing the relation between a conventional page and a new page with conventional pages extracted by the static and dynamic recommendations to give a recommendation regarding the new page. An integration part 11 integrates functions regarding three recommendations to generate a total recommendation. A WWW page including a recommendation 1st is sent to a client system 2.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a recommendation method and a server system having a recommendation function for recommending presentation of a product or the like in an electronic store or the like according to a user's preference.

[0002]

2. Description of the Related Art With the spread of WWW, computerization of commercial transactions is progressing. For example, WWW sites where books, CDs, and the like can be purchased on the WWW have become widespread and widely used. In such an electronic store, unlike the conventional store, the products to be handled are not affected by the volume of the store, so that a huge number of products can be handled. This makes it difficult for users to find products that fit their tastes,
In order to reduce this, a function of recommending a product according to the user's preference is required.

As conventional techniques for such recommendation in WWW, two techniques of collaborative filtering and data mining are known.

[0004] Collaborative filtering is described in Document 1 (Social Information Filtering: Algorithms for Au).
tomationg “Word of Mouth”, ACM CHI Conference, 199
As seen in 5), the user registers preferences in advance, and the recommendation system finds users who have similar preferences from the registered user preferences (user profile), and uses this to make recommendations. It is. For example, in Literature 1, music recommendation is performed. The user evaluates the preference of the initially presented musician on a seven-point scale and registers it as his profile. The system extracts users with similar profiles and recommends musicians that the user has not yet rated based on them.

On the other hand, data mining is a technique for finding hidden knowledge from a large amount of data. WW
In data mining for W, knowledge hidden in a WWW access log stored in a WWW server is found. As a technique for this, correlation rule discovery and clustering are known. In the correlation rule discovery, a set of pages having a strong co-occurrence relationship can be found based on the frequency of appearance. Reference 2 (Fast Algorithms for mining asso
ciation rules, 20 ^th VLDB 1994), an efficient algorithm for finding association rules has been proposed and widely used. In the clustering, clusters of closely related users and clusters of pages can be found. For example, reference 3 (Matrix Clustering: CRM (Custo
In a new data mining method for mer Relationship Management), an algorithm that can simultaneously find clusters of related users and pages has been proposed. In a recommendation system utilizing such data mining technology, all access logs stored in a WWW server are mined and collated with the history of pages accessed by the current user.
Relevant pages can be found and used to make recommendations.

As described above, the recommendation method using collaborative filtering and the recommendation method using data mining use different data and apply different scenes. Recommendations using collaborative filtering can be used only in specific situations, and data representing the characteristics of the user is also only a profile, so it is considered a static recommendation, but recommendations using data mining are: The present invention can be applied to any scene of navigation on the WWW, and the data representing the characteristics of the user uses the access history at that time, so it is considered as a dynamic recommendation.

[0007]

The above-mentioned recommendation technique has the following problems.

[0008] From the user's standpoint, it would be desirable to use static and dynamic recommendations without distinction. However, in the prior art,
Static and dynamic recommendations are separate features and are not integrated. For example, W
Even if the recommendation function is desired during navigation on the WW, only the dynamic recommendation is executed, and the static recommendation referring to the user profile is not executed. Also, even if the user returns to the static recommendation function page and desires the recommendation function, the dynamic recommendation referring to the access history up to that time is not executed.

Further, the conventional recommendation system has the following problems.

In a recommendation system using collaborative filtering, it is necessary to register a user profile in advance. In an electronic store using WWW, an assortment of products is frequently changed frequently. However, in a method using a user profile, it is difficult to recommend a new product because preferences about a new product are not expressed in the user profile. It is.

In a recommendation system using data mining, a recommendation is made based on the access frequency of a WWW page. However, since the access frequency of a new page is low, it is not included in the rules generated by a normal data mining method. . Therefore, it is equally difficult to recommend a new page.

As described above, the conventional recommendation method has a problem that it is not possible to recommend a new page dealing with a new product. On the other hand, promotion of new products is an important marketing issue for providers, and a recommendation system that meets such requirements of providers is needed.

The present invention has been made in consideration of the above circumstances, and has as its object to provide a recommendation method and a server system that can integrate a plurality of types of recommendations without the user being particularly conscious. I do.

[0014] Another object of the present invention is to provide a recommendation method and a server system capable of recommending new data.

[0015]

SUMMARY OF THE INVENTION The present invention is a recommendation method in a server system for providing a predetermined service to a user of a client system by transmitting predetermined data to a requesting client system. For each of the users of the plurality of client systems, information indicating the preference of the user specified by the user with respect to all or a part of the plurality of predetermined existing data is used as a user profile. When the recommendation process for the user of the client system in which the user profile is registered is started, the user profile registered for the target user and the user profile registered for another user are registered. Compared to the created user profile, By comparing the user profile of the user preferences in the user extracts the user profile of another user is determined to be similar, to the user profile and the subject of the extracted said other users of interest,
Existing data to be recommended for the target user is extracted and related to the extracted existing data to be recommended based on information indicating a relationship between the existing data and the new data. It is characterized in that new data is extracted as data to be further recommended to the target user.

Further, the present invention provides a client / client device for a request source.
A recommendation method in a server system for providing a predetermined service to a user of the client system by transmitting predetermined data to the system, wherein a recommendation process for the user of the client system is started. In this case, based on the information indicating the relationship between the existing data derived based on the access histories of the plurality of client systems, the data accessed by the client system corresponding to the target user is Related data,
It is extracted as existing data to be recommended to the target user, and is related to the extracted existing data to be recommended based on information indicating a relationship between existing data and new data. It is characterized in that new data is extracted as data to be further recommended to the target user.

The present invention also relates to a client / client device that issues a request.
A recommendation method in a server system for providing a predetermined service to a user of the client system by transmitting predetermined data to the system, wherein a plurality of predetermined methods are provided for each user of the plurality of client systems. The information indicating the user's preference specified by the user for all or a part of the existing data is registered as a user profile, and the recommendation processing for the user of the client system is performed. Is started, if a user profile is registered for the target user, a user profile registered for the target user is compared with a user profile registered for another user. , Preferred by the target user A user profile of another user determined to be similar is extracted, and the extracted user profile of the other user is compared with the user profile of the target user, and recommended for the target user. If the existing data to be extracted and the access history exists for the target user,
Based on information indicating the relationship between existing data derived based on the access histories of a plurality of client systems, a client
Data related to the data accessed by the system is extracted as existing data to be recommended to the target user, and extracted based on information indicating a relationship between the existing data and the new data. New data related to the specified existing data to be recommended is extracted as data to be further recommended to the target user.

Preferably, in extracting the existing data to be recommended based on the information indicating the relationship between the existing data and the new data, furthermore, the user profile of the target user is further extracted based on the user profile of the target user. New data related to existing data determined to be favorite may be extracted as new data to be recommended to the target user.

Preferably, the server system is a WWW server on the Internet, and the data is
It may be a WWW page.

It should be noted that the present invention relating to the apparatus is also realized as an invention relating to a method, and the present invention relating to a method is also realized as an invention relating to an apparatus. Further, the present invention according to an apparatus or a method has a function of causing a computer to execute a procedure corresponding to the present invention (or causing a computer to function as means corresponding to the present invention, or a computer having a function corresponding to the present invention). The present invention is also realized as a program (for realizing), and is also realized as a computer-readable recording medium on which the program is recorded.

According to the present invention, a plurality of types of recommendation processes can be executed as appropriate according to the situation at the time when the recommendation processes are activated, and the results can be integrated.

Therefore, according to the present invention, for example, by combining a static recommendation using conventional collaborative filtering with a dynamic recommendation using data mining, it is possible to meet the user's preference and at present. It is possible to generate a recommendation in any situation in any situation.

Further, according to the present invention, a recommendation result of new data can be obtained from a recommendation result of existing data and relation information between existing data and new data.

Therefore, according to the present invention, for example, it is possible to construct a WWW site that is effective for marketing an electronic store and that meets the user's preference.

[0025]

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

FIG. 1 shows a configuration example of a network system including a server system equipped with a recommendation system according to an embodiment of the present invention.

As shown in FIG. 1, the network system of the present embodiment includes a server system 1 and a plurality of client systems 2 which can communicate via a predetermined network 3. Although FIG. 1 shows only one server system 1, a plurality of server systems 1 may be provided.

The network 3 can be applied to a wide area network, a public network, or a LAN. In the present embodiment, the Internet 3 will be described as an example of the network 3.

The server system 1 and the client system 2 can be constituted by a computer and software. In the present embodiment, as an example, the server system 1 will be described as a WWW server on the Internet, and the client system 2 will be described as a computer or terminal equipped with a user interface unit such as a browser. The WWW server can be applied to any service that provides any service, such as a service of an electronic store that sells products and the like and a service that provides information for a fee or free of charge.

In the system shown in FIG. 1, for example, a predetermined request message is transmitted from the client system 2 to the server system 1, and the server system 1 responds to the request message.
Performs predetermined processing as necessary, and executes the server system 1
Returning a predetermined response message from the client system 2 to the client system 2 is defined as one unit, and this is repeated as appropriate, whereby one session is performed. In the WWW, page requests and page responses are typical.

Now, the server system 1 of the present embodiment
Sends the client system 2 to the client system 2.
A recommendation system for providing information on a predetermined form of data to be recommended to a user of the system 2 is provided. In the present embodiment, a WWW page is taken as an example of the data in the predetermined form, and therefore, in this recommendation system, a process for providing information on the recommended page to the client system 2 is performed.

As shown in FIG. 1, the server system 1 of the present embodiment is a recommendation system, which includes an integrating unit 11, a collaborative filtering unit 1
3, data mining unit 14, promotion unit 15,
User profile storage unit 16, access log storage unit 1
7, a correlation rule storage unit 18 and a page relation table storage unit 19. In FIG. 1, the server system 1
2 shows only the components related to the recommendation system, and processes for a communication unit for communication via a network and general functions as a WWW server (including a login function and a user profile registration function). Parts and the like are omitted.

The recommendation system of this embodiment has three recommendation means: a collaborative filtering unit 13, a data mining unit 14, and a promotion unit 15. The integration unit 11 is a means for integrating them.

In the present embodiment, the collaborative filtering unit 13, the data mining unit 14, and the promotion unit 15 obtain the results of the processing for the delivery of information (for example, page identification information) on the recommended pages. Information about the recommended page is added to the recommendation list 12, and the promotion unit 15 and the integration unit 11 respectively refer to the recommendation list 12 as needed. Of course, it is not limited to such a form.

In the following, a page recommended by the promotion process (a newly created page) is referred to as a new page, and a page other than a page treated as a new page (existing page) is referred to as a conventional page. Shall be.

The collaborative filtering unit 13
Has a function of performing a static recommendation using a user profile (16) registered for a target user (determining a recommended page from conventional pages). Note that the registration of the user profile (new registration, update registration, etc.) may be performed, for example, together with the user registration procedure for the site, or in response to a request from a user accessing the site. Or another method.

In the data mining section 14, roughly, the relation rules (18) extracted from the access logs (17) accumulated for a plurality (or a large number) of users.
Is compared with the access history (17) of the target user to perform dynamic recommendation (find a recommended page from the conventional pages). Note that the relation rule (8) is, for example, the data mining unit 14
May be extracted, or may be requested to an external computer for processing.

The promotion unit 15 roughly checks the page relation table (9) indicating the relationship between the conventional page and the new page with the conventional page obtained by the collaborative filtering unit 13 and the data mining unit 14. By doing so, it has a function of making a recommendation for a new page (ie, finding a recommended page from among the new pages). In addition, page relation table (9)
Is set in advance from the outside.

The integrating unit 11 generally integrates the three recommendations-related functions of the collaborative filtering unit 13, the data mining unit 14, and the promotion unit 15 to generate the overall recommendation (the collaborative filtering unit 13).
And generating a recommendation including information on the recommended page obtained by the promotion unit 15 and the integration unit 11).

Information including the recommendation list is:
For example, it is transmitted to the client system 2 as the content of a WWW page dedicated to recommendations or as a part of another WWW page (conventional or new WWW page). When the client system 2 receives this and displays the page, the result of the recommendation is presented to the client user. In addition, if the recommendation result is displayed in the form of a link to the recommended page in the client system 2, the client user can easily click the link to the displayed recommended page. You will be able to jump to the page.

Hereinafter, a collaborative filtering unit 13, a data mining unit 14, and a promotion unit 15
Will be described in detail.

First, the collaborative filtering unit 13 will be described.

FIG. 2 shows an example of the processing procedure of the collaborative filtering of the present embodiment. That is, a user profile is read (step S1), a profile of another user similar to the profile of the target user is extracted (step S2), the extracted profile has an evaluation value, and the target user profile has an evaluation value. Is extracted as a page to be recommended (step S3), and information indicating the extracted conventional page is added to the recommendation list (step S4). In step S3, there are various methods such as a method in which only pages having an evaluation value equal to or more than a predetermined value in the extracted profile are to be extracted.

Here, a specific example will be described. FIG.
FIG. 1 shows an example of a user profile used in collaborative filtering. In the example of FIG. 3, A, B,
C, D, and E indicate users whose profiles are registered, respectively, and a, b, c, d, e, f, and g indicate W
Refers to the WW page. The value of the matrix element is an evaluation value indicating the preference of the corresponding user for the WWW page from 1 to 7
Corresponds to what was entered at the stage. For example, user A:
The evaluation value is designated as 7 for pages a and b. This indicates that the user A likes the pages a and b very much. On the other hand, the user A designates the evaluation value as 2 for the page g. This indicates that they do not like it much. Note that the user A sets c, d, e, and f as follows:
You have not entered an evaluation value. Thus, not all elements of this matrix have values.

In the collaborative filtering, first, the profile of the target logged-in user is fetched and compared with the profiles of other users, whereby the profile is close to the preference (similar in profile).
Find other users.

The closeness of preference is determined, for example, by obtaining a correlation coefficient between profiles. In this case, various types of correlation coefficients have been proposed. Here, a method for obtaining a mean square error will be described as the simplest.

For example, when the user A is targeted, only the profile of the user B and the profile of the user E have items common to the profile of the user A in FIG. For, the correlation coefficient can be defined.

First, the profile of the user A and the user B
Since the pages that are commonly evaluated by both are b and g, the correlation coefficient with the profile of (7−7) ² + (2-1) ² = It becomes 1. Next, when the correlation coefficient with the profile of the user E is similarly obtained, (7-7) ² + (2-3) ² = 1. In this way, users having similar preferences to user A can be determined to be B and E (and their similarities can be determined to be the same).

Next, a recommendation candidate is found using the profiles of the user B and the user E. User B
Since the pages evaluated by user E and not evaluated by user A are c and d, information indicating pages c and d is added to the recommendation list.

In the extraction of the profile of another user used for the recommendation, a condition for selecting that the closeness of preference (for example, the correlation coefficient) is equal to or larger than a predetermined value may be used. In addition, it is also possible to select from the one with the closest preference (for example, correlation coefficient) to the one with a predetermined order. In addition, various methods are conceivable.

As the processing of the collaborative filtering, in addition to these, various variations are possible.

Next, the data mining section 14 will be described.

FIG. 4 shows an example of a data mining processing procedure according to this embodiment. That is, the correlation rule and the access history of the target user are read (step S11),
Based on the access history and the correlation rule, a page (conventional page) predicted to be likely to be accessed within the same session is extracted as a page to be recommended (step S).
12) Add information indicating the extracted conventional page to the recommendation list (step S13).

Here, a description will be given using a specific example. FIG.
Represents seven WWW pages (a, b, c, d, e, f,
15 shows an example of a correlation rule obtained based on the access log regarding g). For example, the rule “a → c” on the first line in FIG. 5 indicates that a user accessing page a is likely to access page c in the same session. In the example of FIG. 5, a support value and a confidence value are used as statistical meanings of the correlation rule. The supported values of this rule are page a and page c during all sessions
And the confidence value represents the ratio of the session accessing the page c to the session accessing the page a. For example, in the rule “a → c” on the first line in FIG. 5, pages a and c are accessed in 30% of all sessions, and 70% of sessions in which page a is accessed.
Indicates that page c is being accessed. Therefore,
In the recommendation using this correlation rule, for example, if page a is included in the access history, it is determined that page c should be recommended. The recommendation in the data mining unit 14 adds the page included in the right side of the matched rule to the recommendation list by comparing the left side of the correlation rule and the access history of the target user in this way.

Here, if the page included in the access history of the target user is only “a”, the rules on the first and second lines in FIG. 5 match, so that the page c and the page c are included in the recommendation list. Information indicating f is added.

In the present embodiment, the access history of the target user is the access after the user logs in the session, but a method of using the past access history of the target user is also possible. is there.

Various other variations are possible for the data mining process.

Next, the promotion section 15 will be described.

The promotion unit 15 includes a collaborative filtering unit 13 and a data mining unit 14.
Unlike this, it is a function for recommending a newly created page. The promotion unit 15 uses a page relation table that describes the relation between a newly created page (new page) and a conventionally existing page (conventional page).

FIG. 6 shows an example of the processing procedure of the promotion according to the present embodiment. That is, a recommendation list (in which information about recommended pages is described by the collaborative filtering unit 13 and the data mining unit 14) and a page relation table are read (step S2).
1) Based on the recommendation list and the page relation table, a page to be recommended (new page) is extracted (step S22), and information indicating the extracted new page is added to the recommendation list (step S23).

Here, a specific example will be described. FIG.
Shows an example of the page relation table. In the example of FIG. 7, the newly created page x is related to the conventional pages a and b, and the new page y is related to the conventional page c. Using this information, the promotion unit 15 extracts a new page related to the conventional page described in the recommendation list and adds this to the recommendation list.

In the above example, a new page is obtained based on the page described in the recommendation list. However, a page having an evaluation value in the profile of the target user (or a page having a predetermined value) is obtained. Page that has an evaluation value greater than or equal to
This may also be added to the recommendation list.

FIG. 8 shows an example of the processing procedure of the promotion in this case. That is, the recommendation list, the profile of the target user, and the page relation table are read (step S31), and based on the recommendation list and the page relation table, a page to be recommended (new page) is extracted, and the target page is extracted. A page to be recommended (new page) is extracted based on the user profile and the page relation table (step S32), and information indicating the extracted new page is added to the recommendation list (step S33).

Various other variations are possible for the promotion process.

In the above-described processing, a page having a high evaluation value (evaluation value equal to or more than a predetermined value) in the profile of the target user is not necessarily added to the recommendation list. Therefore, there is a method in which the integrating unit 11 further describes information indicating a page having a high evaluation value in the profile of the target user in the recommendation list.

In the processing so far, a page having a low evaluation value (evaluation value less than a predetermined value) in the profile of the target user may be described in the recommendation list. Therefore, there is a method in which the integration unit 11 also deletes information indicating a page having a low evaluation value in the profile of the target user from the recommendation list.

Next, the operation of the entire recommendation system of this embodiment will be described.

FIG. 9 shows an example of a processing procedure of the recommendation system of the present embodiment.

First, the user logs in to the WWW site (server system 1) (step S41).

At this point, a user profile can be registered (newly created or updated) (step S).
42). That is, when a request for registering a user profile is received from the client system 2 side, predetermined items for registration are presented from the server system 1 side to the client system 2 side, and presented on the client system 2 side. The client user performs input such as specifying the preference for
The server system 1 notifies the server system 1 of an input for a predetermined item, so that the server system 1
The side creates and registers a profile for the user. After the profile is created, the static recommendations (collaborative filtering unit 1)
3) becomes possible.

In the above description, a procedure for registering a user profile is performed in response to a request. However, a method for registering a user profile immediately after the user logs in is also possible. It is.

On the other hand, access to a page (page request)
If the access request is received, the corresponding page is responded and the access history is recorded (step S47). After a page (in this example, a page other than the first page) is accessed, dynamic recommendation (data mining unit 14) becomes possible. Here, as an example, when a page is accessed, a procedure for activating a recommendation is used.

When receiving a recommendation request from the client system 2 side,
When a page request is received from the system 2 side (for example, when a recommendation button or a link to the page is clicked on the page screen of the client system 2 and the request message is received), the following is received. Perform a series of processes for recommendation. That is, when at least one of the profile registration and the page access is performed, the process of the static recommendation (collaborative filtering unit 13) is performed after the profile registration (step S4).
3) If after page access, dynamic recommendation processing (data mining unit 14) is performed (step S).
44) Then, a promotion process (promotion unit 15) is performed (step S45). If there is information to be recommended as a result of the processing so far, a page including the information of the recommendation is sent to the client system 2 side. Is transmitted (step S46).

After performing the processing in step S43 and the processing in step S45 for the result, step S44 and the processing in step S45 for the result are performed.
May be performed. Step S43 and step S44 may be performed in reverse order or simultaneously.

If neither profile registration nor page access has been performed, the process proceeds to steps S43 to S43.
The processing of S46 is not performed. In this case, for example, a page for prompting a procedure for profile registration may be transmitted.

When a profile is created immediately after logging in, for example, a static recommendation can be made from the first page by using this profile. At this point, since there is no access history for the user, the dynamic recommendation is not activated, but the promotion is activated and if there is a new page corresponding to the recommendation list output by the static recommendation, the And provide it to the client user as a recommendation result. For example, assume that the user A logs in the user profile in FIG. As described above, the collaborative filtering unit 13 performs static recommendation, and outputs page c and page d to the recommendation list. Next, the promotion unit 15 determines from FIG.
Since the page c is related to the new page y by referring to the page relation table, the page y is added to the recommendation list. As a result, three pages c, d, and y are output as a recommendation list. Thereafter, the user can display the page by, for example, clicking the recommendation result (link to the page).

On the other hand, the dynamic recommendation is used from an arbitrary page other than the first page, for example. The client user may click on a recommendation button arranged in the page, for example. here,
As before, assume that user A is logged in and accessing. When the recommendation button is clicked, a static recommendation is made. The result of the static recommendation is the same as before, page c and page d. Subsequently, a dynamic recommendation is performed. As before, using the correlation rule of FIG. 5 and assuming that the access history is only page a, page c and page f are output by dynamic recommendation. The two recommendation lists are combined, and pages c, d, and f are passed to the promotion unit 15. The promotion unit 15 adds the page y using the page relation table shown in FIG. As a result, the recommendation list becomes pages c, d, f, and y. These are combined and displayed as a WWW page.

The processing procedure shown in FIG. 9 is an example, and the processing procedure of the recommendation system is as follows.
Various other variations are also possible.

For example, in FIG. 9, the recommendation is performed at the time of page access, but the recommendation may be performed only in response to an explicit request for the recommendation. FIG. 10 shows an example of the processing procedure in this case.

Further, for example, the client system 2
Without requesting recommendations from the side,
The recommendation may be activated when a predetermined event (other than the reception of a request for recommendation) occurs in the server system 1. The predetermined event is, for example, when a page is accessed (when a page request message is received).
For example, when at least one of the user profile, the access log, and the correlation rule is updated. The recommendation may be activated when a request for a recommendation is received from the client system 2 or when a predetermined event occurs in the server system 1.

Further, from the client system 2 side,
It is possible to specify whether each of the static recommendation, the dynamic recommendation, and the promotion is activated, and the server system 1 can perform only the designated process. Alternatively, on the server system 1 side, the content of the event that has occurred and whether or not each of the static recommendation, the dynamic recommendation, and the promotion are activated when the event of the content occurs are set in advance. Alternatively, it is also possible to perform only the set processing according to the content of the event that has occurred.

Next, another configuration example of the data mining unit will be described.

The configuration example of the data mining unit described with reference to FIGS. 4 and 5 is referred to as a first configuration example, and the other configuration examples described here are referred to as a second configuration example.

In the first configuration example, an example in which the association rule discovery is used as the dynamic recommendation has been described.
Other data mining techniques can also be used as dynamic recommendations. Here, as an example, matrix clustering (Matrix C
An example using the technique of L.L.

FIG. 11 shows the process of matrix clustering.

In FIG. 11A, S1 to S5 indicate sessions, respectively, and a, b, c, d, e, and f indicate
Each indicates a WWW page. When the value of the matrix element is 1, it indicates that the corresponding WWW page has been accessed in the corresponding session, and when it is 0, it indicates that the corresponding WWW page has not been accessed in the corresponding session.

In matrix clustering, marker propagation and pruning are repeated from a specified row or column to find a dense matrix.

Here, it is assumed that the access history is only page a.

The example of FIG. 11 shows a process of obtaining a dense matrix starting from column a (that is, obtaining a page having a relationship with page a).

In FIG. 11B, the marker is propagated from column a to rows S2, S3, and S5 where the element is 1. In FIG. 11C, the markers are propagated to the columns a, c, d, e, and f whose elements are 1 from the rows S2, S3, and S5. In FIG. 11D, pruning is performed on the number of markers received in the columns, and columns d and e have the number of received markers of one.
Therefore, they are erased and a, c, and f survive. FIG. 11 (e)
Then, rows S2, S3, in which the element is 1 from columns a, c, f
Markers are propagated toward S4 and S5. FIG.
In (e), since the number of markers in each row is 2 or more,
No pruning is performed. The situation does not change even if marker propagation is repeated any more.
(E). That is, starting from column a, columns c and f
Are obtained as a related column, and the pages c and f are output as a dynamic recommendation list based on the page a.

Such a procedure is performed for each page described in the access history.

Next, a third configuration example of the data mining unit will be described.

In the first configuration example, the correlation rule discovery is used as the dynamic recommendation, and in the second configuration example, the matrix clustering is used. In the correlation rule discovery, since a rule with a high frequency is generated, a page with a low frequency cannot be handled even if the relation is strong. In matrix clustering, it is possible to extract clusters of infrequent pages, but if there are large clusters at the same time, clusters of infrequent pages cannot be extracted due to convergence to the larger one. . Here, an example will be described in which by combining these two, clusters of pages with low frequency are extracted and dynamic recommendation is performed.

FIG. 12 shows an example of the processing procedure of the dynamic recommendation in this case.

First, correlation rule discovery is applied (step S61). Here, it is assumed that the access history is only “a” as in the first configuration example. Pages c and f are output as in the first configuration example by the correlation rule discovery. At this time, there are a method of adding the extracted pages c and f to the recommendation list, and a method of not adding them.

Next, firing of the rows of the matrix corresponding to pages c and f is suppressed in marker propagation (step S6).
2). That is, it is considered that the rows c and f are deleted in the matrix clustering.

Next, matrix clustering is applied (step S63).

FIG. 13 shows a process of matrix clustering in this state. The row in FIG.
The meanings of columns and matrix elements are the same as in FIG.

In this case, it can be seen that columns c and f are excluded in the processing steps of FIGS. In this way, by executing the matrix clustering while suppressing the example output in the correlation rule discovery, pages d and g corresponding to the infrequently found rules not found in the correlation rule discovery are output by the matrix clustering. You.

Then, pages d and g are added to the recommendation list.

Such a procedure is performed for each page described in the access history.

Of course, it is also possible to combine two or more of the processing in the first to third configuration examples of the data mining unit (for example, the logical sum of the result of the processing in the first to third configuration examples is obtained). Take).

In the above description, a system having all the functions of the static recommendation, the function of the dynamic recommendation, and the function of the promotion has been described. However, the system having the function of the static recommendation and the function of the promotion is provided. A system that does not have the function of dynamic recommendation) or a system that does have the function of dynamic recommendation and the function of promotion (does not have the function of static recommendation) is also possible.

Note that each of the above functions can be realized as software. In addition, the present embodiment can be implemented as a program for causing a computer to execute predetermined means (or for causing a computer to function as predetermined means, or for causing a computer to realize predetermined functions), The present invention can also be implemented as a computer-readable recording medium on which the program is recorded.

The configuration illustrated in the embodiment of the present invention is an example, and is not intended to exclude other configurations, and some of the illustrated configurations may be replaced with other configurations or illustrated. Another configuration obtained by omitting a part of the configuration, adding another function or element to the illustrated configuration, or combining them is also possible. Also, another configuration that is logically equivalent to the illustrated configuration,
Another configuration including a portion logically equivalent to the illustrated configuration, another configuration logically equivalent to a main part of the illustrated configuration, and the like are also possible. Further, another configuration that achieves the same or similar purpose as the illustrated configuration, another configuration that achieves the same or similar effect as the illustrated configuration, and the like are also possible. Various variations of the various components exemplified in the embodiments of the present invention can be implemented in appropriate combinations. Further, the embodiments of the present invention relate to an invention as an individual device, an invention about two or more related devices, an invention as a whole system, an invention about components inside an individual device, or a method corresponding thereto. It encompasses and includes inventions related to various aspects, stages, concepts or categories, such as inventions. Therefore, the present invention can be extracted from the contents disclosed in the embodiments of the present invention without being limited to the illustrated configuration.

The present invention is not limited to the above-described embodiments, but can be implemented with various modifications within the technical scope thereof.

[0107]

According to the present invention, a plurality of types of recommendations can be integrated without the user being particularly conscious. According to the present invention, new data can be recommended.

[Brief description of the drawings]

FIG. 1 is a diagram showing a configuration example of a network system including a server system equipped with a recommendation system according to an embodiment of the present invention.

FIG. 2 is a flowchart illustrating an example of a processing procedure of collaborative filtering.

FIG. 3 is a diagram showing an example of a user profile.

FIG. 4 is a flowchart illustrating an example of a data mining processing procedure;

FIG. 5 is a diagram showing an example of a correlation rule;

FIG. 6 is a flowchart illustrating an example of a processing procedure of a promotion.

FIG. 7 is a diagram showing an example of a page relation table.

FIG. 8 is a flowchart showing another example of the promotion processing procedure.

FIG. 9 is a flowchart illustrating an example of an integrated recommendation processing procedure;

FIG. 10 is a flowchart showing another example of the processing procedure of integrated recommendation.

FIG. 11 is a diagram for explaining a process of matrix clustering.

FIG. 12 is a flowchart illustrating an example of a processing procedure of dynamic recommendation using matrix clustering.

FIG. 13 is a diagram for explaining a process of matrix clustering;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Server system 2 ... Client system 3 ... Network 11 ... Integration part 12 ... Recommendation list 13 ... Collaborative filtering part 14 ... Data mining part 15 ... Promotion part 16 ... User profile storage part 17 ... Access log storage part 18 ... Correlation rule storage unit 19: Page relation table storage unit

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification FI FI Theme Court ゛ (Reference) G06F 17/30 ZEC G06F 17/30 ZEC (72) Inventor Kazuto Kubota Toshiba Komukai, Yuki-ku, Kawasaki City, Kanagawa Prefecture No. 1 town Toshiba R & D Center F-term (reference) 5B075 KK07 KK35 ND20 NR02 NR20 PR08 UU40

Claims (14)

[Claims] 1. A recommendation method in a server system for providing a predetermined service to a user of a client system by transmitting predetermined data to a requesting client system, the method comprising: For each of the users, information indicating the preference of the user specified by the user for all or a part of a plurality of predetermined existing data is registered as a user profile, and the user profile When the recommendation processing for the user of the client system for which the user is registered is started, the user profile registered for the target user is compared with the user profiles registered for the other users. To the target user A user profile of another user determined to be similar is extracted, and the extracted user profile of the other user is compared with the user profile of the target user, and recommended for the target user. The user who extracts the existing data to be recommended and, based on the information indicating the relationship between the existing data and the new data, sets the new data related to the extracted existing data to be recommended as the target user. A recommendation method characterized by extracting the data as data that should be further recommended. 2. A recommendation method in a server system for providing a predetermined service to a user of a client system by transmitting predetermined data to a client system of a requester, the method comprising: When the recommendation process for the target is activated, the user corresponding to the target user is handled based on the information indicating the relationship between the existing data derived based on the access histories of the plurality of client systems. Data related to the data accessed by the client system to be extracted as existing data to be recommended to the target user, and based on information indicating a relationship between the existing data and the new data. New data related to the extracted existing data to be recommended , Recommendation method characterized by extracting the data to be further recommended for the user to be the subject. 3. A recommendation method in a server system for providing a predetermined service to a user of a client system by transmitting predetermined data to a requesting client system, the method comprising: For each of the users, information indicating the preference of the user specified by the user for all or a part of a plurality of predetermined existing data is registered as a user profile, and the client When the recommendation process for the user of the system is started, if a user profile is registered for the target user, the user profile registered for the target user and the other user are registered. Registered user profile and In comparison, a user profile of another user whose preference is determined to be similar to the target user is extracted, and the extracted user profile of the other user is compared with the user profile of the target user. Then, existing data to be recommended for the target user is extracted, and if an access history exists for the target user, the data is derived based on the access histories of the plurality of client systems. Existing data to be recommended to the target user based on the information indicating the relationship between the existing data and the data related to the data accessed by the client system corresponding to the target user. Extracted based on information indicating a relationship between existing data and new data. A new recommendation method characterized by extracting new data related to existing data to be recommended as data to be further recommended to the target user. 4. Extracting existing data to be recommended based on the information indicating the relationship between the existing data, a correlation rule finding method or matrix clustering based on access histories of a plurality of client systems. Recommending, to the target user, data related to the data accessed by the client system corresponding to the target user based on the information indicating the relationship between the existing data derived by the method. 4. The recommendation method according to claim 2, wherein the data is extracted as existing data to be obtained. 5. Extracting existing data to be recommended based on the information indicating the relationship between the existing data, first finds a correlation rule based on access histories of the plurality of client systems. Based on the first information indicating the relationship between the existing data derived by the method, the data related to the data accessed by the client system corresponding to the target user is obtained.
It is extracted as existing data to be recommended to the target user, and then the extracted existing data is excluded and matrix clustering is performed based on the access histories of the plurality of client systems. Based on the derived second information indicating a relationship between existing data, recommending data related to data accessed by a client system corresponding to the target user to the target user. 4. The recommendation method according to claim 2, wherein the data is further extracted as existing data to be obtained. 6. Extracting existing data to be recommended based on information indicating a relationship between the existing data and new data, and further extracting the user's preference based on the user profile of the target user. 4. The recommendation according to claim 1 or 3, wherein new data related to existing data determined to be also extracted as new data to be recommended to the target user. Method. 7. The method according to claim 1, wherein the recommendation process is started in response to a recommendation request from the client system.
7. The recommendation method according to any one of items 6 to 6. 8. The recommendation method according to claim 1, wherein said recommendation process is started in response to a page request from said client system. . 9. A server system for providing a predetermined service to a user of a client system by transmitting predetermined data to a requesting client system, wherein the server system provides a predetermined service to a user of the client system. Means for registering, as a user profile, information indicating the user's preference specified by the user for all or some of a plurality of predetermined existing data, and the user profile is registered. When the recommendation process for the user of the client system that is being performed is started, a user profile registered for the target user is compared with a user profile registered for another user, It is determined that the preference is similar to the target user The user profile of the user is extracted, and the extracted user profile of the other user is compared with the user profile of the target user to extract existing data to be recommended for the target user. Means for performing, based on the information indicating the relationship between the existing data and the new data, the extracted new data related to the existing data to be recommended, to the target user Means for extracting data as further recommended data. 10. A server system for providing a predetermined service to a user of a client system by transmitting predetermined data to a client system of a requester, wherein the server system is intended for a user of the client system. When the recommendation process is activated, the client system corresponding to the target user is based on information indicating the relationship between existing data derived based on the access histories of the plurality of client systems. Means for extracting data related to the data accessed by the user as existing data to be recommended to the target user; and information based on information indicating a relationship between the existing data and the new data. New data related to the extracted existing data to be recommended is targeted. Server system, characterized by comprising means for extracting a further be recommended data to a user. 11. A server system for providing a predetermined service to a user of a client system by transmitting predetermined data to a client system of a requester, wherein the server system provides a predetermined service to a user of the client system. Means for registering, as a user profile, information indicating the user's preference specified by the user for each of all or a part of a plurality of predetermined existing data; and When the recommendation process for a user is started, if a user profile is registered for the target user, a user profile registered for the target user and a user profile registered for another user are registered. Compared to the target user profile The user profile of another user whose preference is determined to be similar to the user is extracted, and the extracted user profile of the other user is compared with the user profile of the target user. Means for extracting existing data to be recommended for, and, if recommendation processing for the user of the client system is started, if an access history exists for the target user For example, based on information indicating the relationship between existing data derived based on the access histories of a plurality of client systems, the information related to the data accessed by the client system corresponding to the target user is obtained. Data
Means for extracting as existing data to be recommended to the target user, and the extracted existing data to be recommended based on information indicating a relationship between existing data and new data. Means for extracting new data related to the data as data to be further recommended to the target user. 12. A program for causing a computer to function as a server system for providing a predetermined service to a user of a client system by transmitting predetermined data to a client system of a request source, comprising: For registering, as a user profile, information indicating the user's preference specified by the user with respect to all or some of a plurality of predetermined existing data for each of the users of the client system. When a recommendation process for a user of the client system in which the user profile is registered is activated, a user profile registered for the target user and a user profile registered for another user are registered. Compare with user profile The user profile of another user whose preference is determined to be similar to the target user is extracted, and the extracted user profile of the other user is compared with the user profile of the target user. Based on information indicating the relationship between existing data and new data, a function for extracting existing data to be recommended to the user, A program for causing a computer to implement a function of extracting related new data as data to be further recommended to the target user. 13. A program for causing a computer to function as a server system that provides a predetermined service to a user of a client system by transmitting predetermined data to a client system of a request source, the program comprising: When the recommendation process for the user of the system is started, based on the information indicating the relationship between the existing data derived based on the access histories of the plurality of client systems, A function for extracting data related to the data accessed by the client system corresponding to the user as the existing data to be recommended to the target user; and a function for extracting the existing data from the new data. Based on the information indicating the relationship between Program for the new data related to the data, to realize the functions and the computer for extracting as data to be further recommended for the user to be the subject. 14. A program for causing a computer to function as a server system for providing a predetermined service to a user of a client system by transmitting predetermined data to a client system of a request source, comprising: For registering, as a user profile, information indicating the user's preference specified by the user with respect to all or some of a plurality of predetermined existing data for each of the users of the client system. If the recommendation processing for the user of the client system is activated, if a user profile is registered for the target user, the registered user profile for the target user And register for other users By comparing the extracted user profile with the target user, a user profile of another user whose preference is determined to be similar to the target user is extracted, and the extracted user profile of the other user and the target user A function for comparing with a user profile and extracting existing data to be recommended for the target user; and a case where recommendation processing for the user of the client system is started. If there is an access history for the target user, based on information indicating the relationship between existing data derived based on the access histories of the plurality of client systems, Data related to the data accessed by the corresponding client system,
A function for extracting as existing data to be recommended to the target user, and the extracted existing recommended data based on information indicating a relationship between the existing data and the new data. A program for causing a computer to realize a function for extracting new data related to data as data to be further recommended to a target user.