JP2014067180A - Information recommendation device, method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To recommend a variety of items for prompting users to utilize a content providing service without recommending only similar items.SOLUTION: When a user utilizes an item, information on the user and information on the item are acquired as a combination so as to store the combination in service utilization log storage means. A log is acquired from the service utilization log storage means, and a graph structure is generated where items and users appearing in the log are defined as nodes and relations between the items recorded in the log and the log are regarded as edges. Importance levels of the items are evaluated on the basis of the generated graph structure, and the graph structure is divided into several partial graphs on the basis of graph associations. Priorities of the items are evaluated on the basis of information on the partial graphs, and items to be recommended are selected. An item is acquired from the selected items in response to a request from the user so as to present the item.

Description

  The present invention relates to an information recommendation apparatus, method, and program, and more particularly, to a content providing service that provides an electronic or physical item to a user. The present invention relates to an information recommendation apparatus, method, and program for recommending an item to be performed.

  Conventionally, there is a method using collaborative filtering as a technique for recommending an item using a user's past item usage history. In this method, the recommendation is performed based on the relationship between the user and the item. For example, when considering item recommendation for a certain user A, a set of users whose past item usage history is similar to the user A is specified, and among these items that are frequently used by these sets of users The user A recommends an item that has not been used so far (see, for example, Non-Patent Document 1).

  In addition, as another method using a user's past item usage history, there is a method in which the relationship between a user and an item included in the usage history is regarded as a graph structure, and a recommendation is made based on the relationship in the graph structure. . For example, Kontas et al. Have proposed a method of recommending by expressing the relationship between a user and an item included in the usage history in a graph structure and considering random walk with restart in the graph structure. By considering random walk with restart, it is possible to recommend items that are highly relevant to the user to whom the item is recommended and generally popular items. Further, in addition to items and users as nodes on the graph structure, metadata indicating the relationship between items and between users can be considered, so that even when the history is small, effective recommendation is possible. (For example, refer nonpatent literature 2).

Paul Resnic, et.a .: GroupLens: an open architecture for collaborative filtering of netnews, Proc.CSCW '94. Konstas, I., Stathopoulos, V. and Jose, J. M .: On social networks and collaborative recommendation Proc. SIGIR'09.

  In the first method described above, a user group having a tendency to use an item similar to the recommendation target is searched for, and an item that the user group uses frequently is presented. However, the recommended target user uses the item sufficiently in the first place. If not, it is difficult to present an appropriate recommended item.

  On the other hand, in the second method shown above, because of the nature of the random walk algorithm, it is possible to evaluate popular items that are highly relevant to other nodes on the graph as important items. Recommendations considering both factors of the degree of popularity are possible. For this reason, it is possible to recommend an item with a high degree of popularity even when the user to be recommended does not use the item sufficiently, and it can be used in a wider situation.

  However, in the recommendation algorithm based on the random walk algorithm, nodes existing at similar positions on the graph tend to be assigned with the same importance. In addition, introducing metadata increases the possibility that nodes indicating similar items are arranged at similar positions in the graph structure. Due to these reasons, there are problems that similar items are arranged in the list of recommended items, the diversity of the recommendation list is impaired, and effective recommendation cannot be performed.

  The present invention has been made in view of the above points, and in the method of recommending by expressing the relationship between the user and the item in a graph structure, the user is encouraged to use the content providing service without recommending only similar items. It is an object of the present invention to provide an information recommendation apparatus, method, and program capable of realizing recommendation of various items.

In order to solve the above problems, the present invention (Claim 1) is an information recommendation device for recommending an item that a user will use in a content providing service that provides an electronic or physical item to the user. Because
An information recommendation device for recommending an item that a user will use in a content providing service that provides an electronic or physical item to a user,
When the user uses the item, log acquisition means for acquiring the user information and the item information in combination and storing the information in the service use log storage means;
A graph construction unit that obtains a log from the service use log storage unit, generates a graph structure that regards an item appearing in the log and a user as each node, and regards a relationship between the item recorded in the log and the log as an edge; ,
Importance evaluation means for calculating the importance of an item based on the graph structure constructed by the graph construction means;
Graph dividing means for dividing the graph structure constructed by the graph constructing means into several subgraphs based on the connection of graphs;
Based on the importance obtained by the importance evaluation means and the information of the subgraph obtained by the graph dividing means, the priority of the item is evaluated, the recommended item is selected, and the recommended item storage means is stored. A recommendation target specifying means to perform,
Recommended item presenting means for acquiring and presenting an item to a specific user in response to a request from the user from among the items in the recommended item storage means.

In the present invention (Claim 2), in the recommendation target specifying means,
Means for adjusting the priority according to whether there is an item belonging to the same subgraph.

In the present invention (Claim 3), in the recommendation target specifying means,
Means for adjusting the priority according to a distance between a node indicating the user and a shortest path to the subgraph;

  As described above, according to the present invention, when an item is recommended to a user based on the past item use history of the user, the item to be recommended is divided into a plurality of groups based on the relationship between the user and the item. Using item affiliation to the group and the relationship between the group and the user, prioritizing and presenting various recommended items based on the user's interests, the item recommendation was similar. While avoiding duplication of items, it is possible to recommend an item that suits the user's interest and support the user's use of the item.

It is a block diagram of the information recommendation apparatus in one embodiment of this invention. It is an example of the data of the service utilization log in one embodiment of this invention. It is an example of the item database in one embodiment of this invention. It is an example of the graph structure in one embodiment of this invention. It is an example of the graph structure division | segmentation in one embodiment of this invention. It is an example of the recommendation item database in one embodiment of this invention. It is a flowchart of the online process in one embodiment of this invention.

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

  FIG. 1 shows a configuration of an information recommendation device according to an embodiment of the present invention.

  The information recommendation apparatus shown in FIG. 1 includes a content providing application 1, an item database 2, a browser 3, a log acquisition unit 101, a graph construction unit 102, an importance evaluation unit 103, a graph division unit 104, a recommendation target specifying unit 105, a recommended item. A presentation unit 106, a service use log database 201, and a recommended item database 202 are included.

  The content providing application 1 receives information for identifying the user such as a user ID from the user via the browser 3, accesses the item database 2 based on the internal content providing logic, and displays the obtained item information in the browser 3. To the user. Further, the recommended item presenting unit 106 is accessed based on the user ID information, and the obtained recommended item is presented to the user via the browser 3. Furthermore, when the user uses the item in a predetermined form (viewing or purchasing), the user ID and the item ID used by the user are passed to the log acquisition unit 101.

  The log acquisition unit 101 receives a combination of a user ID and an item ID from the content providing application 1 and stores it in the service usage log database 201 together with time stamp information.

  The service usage log database 201 is a database that records a user's item usage history. An example is shown in FIG. The service usage log database 201 includes items of user ID, item ID, and date / time.

  The graph construction unit 102 has a graph structure based on a pair of user ID and item ID stored in the service usage log database 201, with each user and item as a node, and the relationship in the service usage log between them as an edge. Build up. In addition, it is conceivable that the metadata elements of items stored in the item database 2 are also handled as nodes by a predetermined setting. For example, when a keyword indicating a field of an item is described in metadata, the keyword is recognized as a node in the graph structure and is connected to the node indicating the item by an edge. An example of the item database 2 is shown in FIG. The item database 2 includes an item ID, an author, and a plurality of keywords. Further, it is conceivable that the weights of the edges connecting the nodes in the graph are all equal or values normalized by the out-degrees. FIG. 4 shows an example of the graph structure. The constructed graph is transferred to the importance evaluation unit 103 and the graph division unit 104.

  The importance evaluation unit 103 evaluates the relevance between the item to be recommended for the item and the item on the graph structure acquired from the graph construction unit 102 as the importance. Then, the combination of the user, the item, and its importance is passed to the recommendation target specifying unit 105. As a method of calculating importance, random walk with restart that evaluates the importance between nodes considering the minimum number of steps between the target user and each item and a random walk on the graph (Non-patent Document 3: Gori, M and Pucci, A .: ItemRank A Random-Walk Based Scoring Algorithm for Recommender Engines. Proc. IJCAI '07).

  The graph division unit 104 divides the graph structure acquired from the graph construction unit 102 into a plurality of subgraphs based on information such as edge density. The acquired division information is passed to the recommendation target specifying unit 105. An example of division into subgraphs is shown in FIG. As a division method, a method using an MDL (Minimum Description Length) principle (for example, Non-Patent Document 4: Deepayan Chakrabarti, Spiros Papadimitriou, Dharmendra S. Modha, and Christos Faloutsos: "Fully automatic cross-associations", Proc. of KDD 2004.).

  Based on the importance of the item based on the relevance to the recommendation target user acquired from the importance evaluation unit 103 and the division information of the graph structure acquired from the graph division unit 104, the estimation target specifying unit 105 Priorities are evaluated, items to be recommended are specified, and stored in the recommended item database 202.

  The recommended item selection method is as follows. First, items are grouped from the division information of the graph, and items with high importance are selected from the respective groups and set as recommended items. Further, prioritization of the recommended item is considered using the distance between the user and the partial graph in the graph structure (for example, a value calculated by the minimum number of hops). That is, in determining the priority of a recommended item, not only the so-called importance level but also the following indices are taken into consideration to identify a recommended item that is close to the user's interest while having diversity.

-Priority adjustment when similar items are included in the recommendation list:
For example, when there is an item belonging to the same subgraph, it is conceivable to discount the importance of the item.

-Priority adjustment by the distance between the subgraph belonging to the graph structure and the user:
For example, it is conceivable to discount the importance of an item belonging to a subgraph that is far away by using a node indicating a user and the shortest path to the subgraph.

  Referring to the example of FIG. 5, the recommended items include an item group with a metadata keyword “NGN”, an item group with “power saving”, and an item group with “Kyoto”. Select a recommended item from each group, In addition, an item group that is close to the user is preferentially presented.

  The recommended item database 202 is a database that stores the recommended items specified by the recommendation target specifying unit 105 together with the target user and the priority. FIG. 6 shows an example of the recommended item database 202.

  The recommended item presentation unit 106 accesses the recommended item database 202 in response to an inquiry from the content providing application 1, acquires a recommended item for the target user, and transmits it to the content providing application 1. Specifically, based on the user ID of the inquiry source, the recommended item database 202 is searched, N items are extracted in descending order of priority, and output to the content providing application 1.

  The browser is an interface with the user, receives information for identifying the user such as a user ID, displays information generated by the content providing application 1, and receives a content use request.

  Next, a processing procedure in the above configuration will be described.

  The procedure is roughly divided into online processing and offline processing.

  In the online processing, the content providing application 1 plays a central role, and the log is stored in the service usage log database 201 and the recommended items stored in the recommended item database 202 are presented to the user according to the user's content usage.

  As offline processing, the recommended item database 202 is constructed using the logs and metadata stored in the service usage log database 201 and the recommended item database 202.

  The procedure for online processing is shown below.

  FIG. 7 is a flowchart of online processing according to an embodiment of the present invention.

  Step 101) The graph construction unit 102 obtains the relationship between the user and the item from the service usage log database 201, obtains the item metadata from the item database 2, and constructs a graph structure having the user, item, and metadata as nodes. And output to the importance evaluation unit 103 and the graph division unit 104.

  Step 102) The graph division unit 104 divides the graph structure acquired from the graph construction unit 102 into a plurality of partial graphs based on the graph structure such as edge density, and outputs the division information to the recommendation target specifying unit 105.

  Step 103) The importance evaluation unit 103 calculates the importance of each item for the determined user based on the graph structure acquired from the graph construction unit 102, and calculates the importance of each user and the importance of each item. Information is transmitted to the recommendation target specifying unit 105.

  Step 104) The recommendation target specifying unit 105 evaluates the priority of each item based on the importance of each node acquired from the importance evaluation unit 103 and the division information acquired from the graph division unit 104, and the item and priority. The combination information of the degree is stored in the recommended item database 202 together with the user identification information.

  Step 105) If the process for all users is completed, the process is terminated. If not completed, the process proceeds to the subsequent stage of Step 101.

  As described above, in the present invention, the graph structure in which the relationship between the item used for recommendation and the user is expressed is divided, the belonging state of the item to the subgraph, and the distance on the graph structure of the recommended user and the subgraph. The recommendation of similar items is suppressed by determining items to be recommended based on. Thereby, when recommending an item in the content providing service, it is possible to recommend a variety of items and to support the user's use of the item.

  In the present invention, the operations of the components shown in FIG. 1 described above can be constructed as a program, installed in a computer used as an information recommendation device, executed, or distributed via a network.

  The present invention is not limited to the above-described embodiment, and various modifications and applications can be made within the scope of the claims.

1 Content provision application (AP)
2 Item database (DB)
3 Browser 101 Log acquisition unit 102 Graph construction unit 103 Importance evaluation unit 104 Graph division unit 105 Recommendation target specification unit 106 Recommended item presentation unit 201 Service use log database (DB)
202 Recommended Item Database (DB)

Claims (7)

An information recommendation device for recommending an item that a user will use in a content providing service that provides an electronic or physical item to a user,
When the user uses the item, log acquisition means for acquiring the user information and the item information in combination and storing the information in the service use log storage means;
A graph construction unit that obtains a log from the service use log storage unit, generates a graph structure that regards an item appearing in the log and a user as each node, and regards a relationship between the item recorded in the log and the log as an edge; ,
Importance evaluation means for calculating the importance of an item based on the graph structure constructed by the graph construction means;
Graph dividing means for dividing the graph structure constructed by the graph constructing means into several subgraphs based on the connection of graphs;
Based on the importance obtained by the importance evaluation means and the information of the partial graph obtained by the graph dividing means, the priority of the item is evaluated, the recommended item is selected, and the recommended item storage means is selected. Means for identifying recommended objects to be stored;
Recommended item presenting means for acquiring and presenting an item in response to a request from the user among items of the recommended item storage means for a specific user;
An information recommendation device comprising: The recommendation target specifying means includes:
The information recommendation device according to claim 1, further comprising means for adjusting the priority according to whether or not an item belonging to the same subgraph exists. The recommendation target specifying means includes:
The information recommendation device according to claim 1, further comprising a unit that adjusts the priority according to a distance of a shortest path from the node indicating the user to the subgraph. An information recommendation method for recommending an item that a user will use in a content providing service that provides an electronic or physical item to a user,
In an apparatus having service usage log storage means, log acquisition means, graph construction means, importance evaluation means, graph division means, recommended item storage means, recommendation target specifying means, recommended item presentation means,
When the user uses an item, the log acquisition unit acquires the user information and the item information as a combination, and acquires the log acquisition step for storing the information in the service usage log storage unit;
The graph construction means obtains a log from the service usage log storage means, an item appearing in the log and a user as each node, and a graph structure in which the relationship between the item recorded in the log and the log is regarded as an edge A graph construction step to generate,
The importance evaluation unit calculates the importance of an item based on the graph structure constructed in the graph construction step, and
A graph dividing step in which the graph dividing means divides the graph structure constructed in the graph construction step into several subgraphs based on the connection of the graphs;
The recommendation target specifying means evaluates the priority of an item based on the importance obtained in the importance evaluation step and the information of the subgraph obtained in the graph division step, and selects an item to be recommended And a recommendation target specifying step stored in the recommended item storage means,
The recommended item presenting means obtains and presents an item in response to a request from the user from among the items of the recommended item storage means for a specific user,
Recommending information. In the recommendation target specifying step,
The information recommendation method according to claim 4, wherein the priority is adjusted depending on whether or not an item belonging to the same subgraph exists. In the recommendation target specifying step,
The information recommendation method according to claim 4, wherein the priority is adjusted according to a distance between a node indicating the user and a shortest path to the subgraph. Computer
The information recommendation program for functioning as each means of the information recommendation apparatus of any one of Claims 1 thru | or 3.

Images