JP5467062B2 - Information recommendation apparatus, method and program - Google Patents

Description

  The present invention relates to an information recommendation apparatus, method, and program, and more particularly to an information recommendation apparatus, method, and program for enabling a plurality of service content items (hereinafter referred to as “items”) to be recommended across information. About.

  Conventionally, there is a technique for improving accuracy in a single service using metadata assigned to an item as a recommendation system. In the technology, the recommendation system can recommend an item with higher accuracy, particularly in a situation where the user's item usage history is small (for example, see Non-Patent Document 1).

  In addition, in order to recommend items handled by other services to users of the own service, there is a method of using a user's usage history common to the other services and the own service (for example, see Non-Patent Document 2).

Ziegler, C. N., Lausen, G. and Thieme, L. S .: Taxonomy-driven computation of product recommendations, in Proc.CIKM'04, pp. 406-415 (2004). Yanagihara, T., Hoashi, K., Ono, C., Matsumoto, K. and Takishima, Y .: Proposal and Evaluation of Cross Media Recommendation, DBSJ Journal, Vol. 8, No. 2, pp. 13-18 ( 2009).

  However, in the technique of Non-Patent Document 1, when items in a plurality of services are considered as targets for recommendation, the item metadata is often not common, and when items handled by other services are recommended to users of the own service. Has a problem that it is difficult to use.

  Further, the technique of Non-Patent Document 2 may recommend items for other services at all when there is no common user among multiple services or when user ID matching between services is not possible. There is a problem that you can not.

  The present invention has been made in view of the above points, and an object of the present invention is to provide an information recommendation device, method, and program that can be recommended even for other services that do not have common information (such as user history). And

In order to solve the above-described problems, the present invention is an information recommendation apparatus that enables a user of a service to highly accurately recommend items of other services in addition to the item recommendation function of the service. And
A history database that stores service usage history information for each of multiple services and information about items and item metadata;
Metadata associating means for associating the metadata assigned to the items handled by the respective services acquired from the history database with each other;
An adjacency matrix creating means for generating an adjacency matrix using information representing an association between the service usage history and the metadata, and storing the adjacency matrix in an adjacency matrix database;
Predicted value calculation means for calculating the degree of association between each element in the matrix using the adjacency matrix stored in the adjacency matrix database;
Recommended item presenting means for presenting the highly relevant items to the user.

The metadata association means includes :
For the previous SL metadata object for calculating the degree of association from the historical database set includes means for calculating the statistics based on the co-occurrence frequency information in the text data including the items of the item associated with each corresponding metadata .

  As described above, according to the present invention, recommendations for items handled in other services can be realized with higher accuracy than conventional methods using user usage history information in single or multiple services. it can.

  Further, according to the present invention, the use history information of a user of a service in a single service or a plurality of services is used to recommend an item handled in another service, and the use history information of a user common between these services. Even when there is no common information such as common metadata, it is possible to present recommended items.

It is a block diagram of the information recommendation apparatus in one embodiment of this invention. It is an example of the adjacency matrix for the information recommendation apparatus in one embodiment of the present invention. It is a flowchart of a process of the information recommendation apparatus in one embodiment of this invention. It is a flowchart of the process of the conventional information recommendation apparatus.

  Embodiments of the present invention will be described below 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 device shown in the figure includes a user ID acquisition unit 101, an item history DB (database) 110 for service A, an item history DB 120 for service B, a metadata association unit 130, an adjacency matrix creation unit 140 for recommendation devices, It consists of an adjacency matrix DB 150 for recommendation devices, a predicted value calculation unit 160 and a presentation unit 170.

  The item history DB 110 in the service A and the item history DB 120 in the service B include a user ID in each service, an item used by the user (service usage history), an evaluation value for the item, and metadata about the item.

  In the figure, the feature of the present invention is that it has a metadata associating unit 130 in particular. A database in which the adjacency matrix for recommendation device 140 constructs an adjacency matrix for recommendation device as shown in FIG. 2 and manages them using the combination of contents and the combination of metadata output from the metadata association unit 130 Is stored in the adjacency matrix DB 150 for recommended devices.

  FIG. 3 is a flowchart of the process of the recommendation system in one embodiment of the present invention. For comparison, FIG. 4 shows a flow of a conventional recommendation system.

  The difference between the present embodiment shown in FIG. 2 and the conventional example shown in FIG. 4 is that related item metadata obtained by the processing of S220 in FIG. An adjacency matrix generation process S330 is added. By this process, the adjacency matrix for the recommendation device according to the present invention shown in FIG. 2 can be constructed in an expanded form of the adjacency matrix constructed and used by the conventional recommendation system.

  A process for recommending an item based on the item history will be briefly described with reference to FIG.

  Step 310) First, as the recommendation target user ID acquisition process, the user ID acquisition unit 101 acquires the ID of the user who receives the recommendation, and outputs it to the metadata association unit 130 and the recommended device adjacency matrix creation unit 140. This process is performed at the same time as the login process to the service or system used by the user, for example.

  Step 320) Next, as the service usage history information acquisition process, the metadata association unit 130 refers to the item usage history of the user in the item history DB 110 of the service A using the ID of the user receiving the recommendation as a key, and sets the content name and the metadata. Get the data. Further, the adjacency matrix creation unit for recommendation device 140 acquires the user ID, content name, evaluation value, and metadata from the item history DB 120 of the service B using the ID of the user receiving the recommendation as a key.

  Step 330) As the related metadata adjacency matrix generation process, the metadata association unit 130 includes the item metadata in the item usage history of the user acquired from the item history DB 110 of the service A and the item in the item history DB 120 of the service B. The association process with the metadata is performed, and the result is output to the adjacency matrix creation unit 140 for the recommendation device.

  Step 340) The recommendation device adjacency matrix creation unit 140 substitutes a numerical value indicating the degree of association for the user, item, and metadata obtained in Steps 320 and 330, and sets the adjacency for each pair. A matrix is created and stored in the adjacency matrix DB 150 for recommended devices.

  Step 350) As a predicted value calculation process, the predicted value calculation unit 160 calculates the degree of association between each element in the matrix using the adjacent matrix stored in the recommended device adjacent matrix DB 150. The presentation unit 170 presents items highly relevant to the items in the user's history to the user who receives the recommendation. For calculating the similarity between items in the adjacency matrix, a network with each item (user, item, metadata) as a node can be calculated using Random Walk with Restart (RWR), which is a network mining technique. .

In RWR calculation, p ^(c) is the vector of item relevance at step t of calculation, α is the probability of returning to the starting node, A is the adjacency matrix of the recommendation system, and q is only the starting node representing the user receiving the recommendation Assuming that the vector is 1, the expression is as shown in Equation (1).

p ^{(c + 1)} = αAp ^(c) + (1−α) q Equation (1)
Step 360) As the recommended item presentation processing, the presentation unit 170 presents a vector having a high degree of relevance from the vector p ^(c) representing the degree of relevance to the user corresponding to the starting node, and the item recommendation is completed.

Next, the adjacency matrix A = (α _{i, j} ) will be described.

  First, the user, item, and item metadata are represented by nodes, and are represented by a single weighted graph G composed of edges indicating the respective relationships. If the node set is V and the edge set is E, G = {V, E}.

Further, U is a set of users, I is a set of items, and M is a set of item metadata. For each of these, services A and B are distinguished from each other by V = {U _A , U _B , I _A , I _B , M _A , M _B }
A vector representing the user set of service A

サービスBのユーザ集合を表すベクトルを

A vector representing the user set of service B

サービスAのアイテム集合を表すベクトルを

A vector representing the item set of service A

サービスBのユーザ集合を表すベクトルを

A vector representing the user set of service B

サービスAのメタデータ集合を表すベクトルを

A vector representing the metadata set of service A

サービスBのメタデータ集合を表すベクトルを

A vector representing the metadata set of service B

と定義される。

It is defined as

U _A I _A ^T represented by the matrix of Figure 2 has a user U _{a, i} item I _a, the value r _i representing the presence or absence of evaluation values or the history for _j is 0 or 1 _{and j} as elements . The matrix represented by I _A M _A ^T is a matrix indicating whether or not the item I _{a, i} has metadata M _{a, j,} and if it has metadata _{, the} r _{i, j} value is 1. Otherwise it is 0. I _B M _B ^T is similarly defined. I _A M _B ^T is a matrix representing the association of items, and the degree of association s _{i, j as} an element is determined by a function g (I _{a, i} , I _{b, j} ) determined by the procedure for calculating the degree of association between items. Is defined, and g (I _{a, i} , I _{b, j} ) = g (I _{b, j} , I _{a, i} ) if the function g is symmetric about the argument. I _B I _A ^T is similarly defined. M _A M _B ^T is a matrix having metadata relevance m _{i, j} as an element, and is defined by m (M _{a, i} , M _{b, j} ) in equation (3) described later. The calculation procedure for each degree of association is as follows.

  Next, the specific calculation process in the above step 330 and the process procedure up to the step 360 recommended item presentation process will be described.

(1) First, the metadata association unit 130 collects the item set I _A of the service A from the item history DB 110 (service A). Similarly, the item set IB of the service _B is collected from the item history DB 120 (service B).

(2) For each of these pairs, the degree of association s _{i, j} between the items I _{a, i} and I _{b, j} is calculated using the function g (I _{a, i} , I _{b, j} ). Various methods can be used for g (I _{a, i} , I _{b, j} ). For example, a Jaccard coefficient obtained using the appearance frequency in large-scale text data such as a blog can be used. The Jaccard coefficient is as follows when a document set including an item I _{a, i} with _a service A is _a vector d _{a, i} and _a vector of the document set including an item I _{b, j} of a service B is d _{b, j} It is expressed as equation (2).

（３）メタデータ間の関連度ｍ_i,jを、以下の関数m(M_a,i，M_b,j)を用い計算する。

(3) The degree of association m _{i, j} between metadata is calculated using the following function m (M _{a, i} , M _{b, j} ).

The degree of association m _{i, j} between metadata M _{a, i} and M _{b , j} is a vector of item sets having M _{a, i}

を持つアイテム集合のベクトル

Vector of item sets with

とすると、以下の式３で表される。

Then, it is expressed by the following formula 3.

ただし、iとｊの組合せ<i,j>はアイテム集合における積集合の添え字の組合せであり、

However, the combination <i, j> of i and j is a combination of the subscripts of the product set in the item set,

である。

It is.

(4) The recommendation device adjacency matrix creation unit 140 creates the adjacency matrix A shown in FIG. 2 according to the following equation (step 340). a _{i, j} represents an element of the adjacency matrix A. r _{i, j} is a value indicating the presence or absence of association in the content history DB 110, s _{i, j} is the similarity between items calculated using the equation (2), and _{mi, j} is a metadata equation ( 3) the degree of association between metadata calculated using

として与えられ、隣接行列Aが定義される。a_i,jは、グラフを用いたアプローチで用いられる正規化処理として、必要に応じて添え字jの表すノードの入次数によって除して用いてもよい。上記のように求められた行列は推薦装置用隣接行列ＤＢ１５０に格納される。

And an adjacency matrix A is defined. a _{i, j} may be used as a normalization process used in a graph-based approach by dividing by a degree of the node represented by the subscript j as necessary. The matrix calculated | required as mentioned above is stored in the adjacency matrix DB150 for recommendation apparatuses.

  (5) The predicted value calculation unit 160 performs RWR calculation on the adjacency matrix A t times. t is 100 times (step 350). The number of implementations is not limited to 100, and may be a sufficiently large value.

(6) The presentation unit 170 rearranges the vector p ^(c) of the degree of association of items after t times in descending order of similarity. Then, for users who have only domain A items, only domain B items are extracted in descending order from the vector p ^(c), and are presented to the user as rankings of recommended items, and the process ends (step S3). 360).

  In the above embodiment, the item history DBs 110 and 120 have been described as being included in the information recommendation device. However, the present invention is not limited to this example, and may be connected to the outside of the information recommendation device.

  Further, each operation of the constituent elements of the information recommendation apparatus shown in FIG. 1 can be constructed as a program and installed in a computer used as the information recommendation apparatus for execution, or distributed via a network. is there.

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

101 User ID acquisition unit 110 Item history DB (service A)
120 Item history DB (Service B)
130 Metadata Association 140 Recommendation Device Adjacency Matrix Creation Unit 150 Recommendation Device Adjacency Matrix DB
160 Predicted value calculation unit 170 Presentation unit

Claims (5)

In addition to the item recommendation function for the service, it is an information recommendation device that can highly accurately recommend items of different other services to users of the service,
A history database that stores service usage history information for each of multiple services and information about items and item metadata;
Metadata associating means for associating the metadata assigned to the items handled by the respective services acquired from the history database with each other;
An adjacency matrix creating means for generating an adjacency matrix using information representing an association between the service usage history and the metadata, and storing the adjacency matrix in an adjacency matrix database;
Predicted value calculation means for calculating the degree of association between each element in the matrix using the adjacency matrix stored in the adjacency matrix database;
Recommended item presentation means for presenting the user with a high degree of association;
An information recommendation device comprising: The metadata association means includes :
For the previous SL metadata object for calculating the degree of association from the historical database set includes means for calculating the statistics based on the co-occurrence frequency information in the text data including the items of the item associated with each corresponding metadata The information recommendation device according to claim 1. In addition to the item recommendation function for the service, it is an information recommendation method that can highly accurately recommend items of different other services to users of the service,
A history database that stores service usage history information for each of multiple services and information about items and item metadata;
In an apparatus having an adjacency matrix database storing adjacency matrices,
The metadata association means
A metadata association step for correlating the metadata given to the items handled by the respective services acquired from the history database;
An adjacency matrix creating means that generates an adjacency matrix using information representing an association between the service usage history and the metadata, and stores the adjacency matrix in the adjacency matrix database;
A predicted value calculation means for calculating a relevance between each element in the matrix using the adjacent matrix stored in the adjacent matrix database;
A recommended item presenting means for presenting a highly relevant item to the user, a recommended item presenting step;
Recommending information. In the metadata association step ,
For the previous SL metadata object for calculating the degree of association from the historical database set, claim 3 for calculating the statistics based on the co-occurrence frequency information in the text data including the items of the item associated with each corresponding metadata The information recommendation method described.   The information recommendation program for functioning a computer as each means which comprises the information recommendation apparatus of Claim 1 or 2.

Images