JP2012221315A - Information recommendation device, method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To select meta data more strongly associated only with a user who is to receive recommendation from meta data imparted to an item, and to present it simultaneously with a recommendation item.SOLUTION: An information recommendation device comprises an item history database storing information on a user ID, an item ID, the meta data of the item, and an evaluation value for the item by a user, and an adjacency matrix construction processing part for a recommendation system, which indicates the information in the item history database as an adjacency matrix. A recommendation item is predicted for a user using the adjacency matrix, and for the recommendation item and the meta data imparted to the item as associated meta data, respective degrees of association with the user are outputted, personalized meta data to be presented simultaneously with the item is extracted using the recommendation item, the associated meta data and the adjacency matrix, and the recommendation item and the personalized meta data are presented to the user.

Description

  The present invention relates to an information recommendation device, method, and program, and more particularly, to an information recommendation device, method, and program for simultaneously presenting personalized metadata of an item in information recommendation.

  As a conventional technique for realizing the present invention, there are a method of selecting metadata having the highest appearance frequency and a method of selecting based on statistical criteria such as a Jaccard coefficient and a chi-square value. The method of using the appearance frequency is the most assigned metadata by counting the number of metadata assigned to all items from among the metadata assigned to items to be recommended (hereinafter referred to as recommended items). It is a method to choose. In addition, methods based on statistical indicators such as the Jaccard coefficient and chi-square value can calculate the degree of association between the metadata of items included in the user's history and the metadata of recommended items, and the user's history is large. In some cases, personalized metadata can be extracted.

  In addition, a method of selecting metadata highly relevant to the user from among the metadata assigned to the recommended items by a graph theory approach called Random Walk with Restart (RWR) (for example, non-patent literature) 1). RWR is a method for calculating the degree of association between users and metadata in a graph that has user, item, and metadata as nodes in the graph and the relationship as an edge. It can be extracted as metadata.

Tong, H., Faloutsos, C. and Pan, J.-Y .: Fast random walk with restart and its applications, in Proc.ICDM '06, pp.613-622 (2006).

  However, the method of selecting the most frequently occurring metadata or the method of selecting based on statistical criteria such as the Jaccard coefficient and chi-square value tends to make it difficult to extract metadata when the user has a small number of items as history. In addition, there is a problem that selection of metadata given hierarchically is not successful.

  Further, by applying the technique of Non-Patent Document 1, the same problem as the statistical standard can be avoided, but it is associated with various items such as metadata that is a large amount of item metadata. There is a problem that the metadata that is being used is easily selected.

  The present invention has been made in view of the above points, and selects metadata (personalized metadata) more strongly related only to the user receiving the recommendation from the metadata assigned to the item, and the recommended item It is an object to provide an information recommendation device, method and program that can be presented simultaneously.

In order to solve the above problems, the present invention provides a recommendation device capable of presenting and recommending personalized metadata of an item to a service user in addition to the item recommendation function in the service. There,
An item history database that stores information about user ID, item ID, item metadata, and user evaluation values,
An adjacency matrix construction processing unit for a recommendation system that represents the information of the item history database as an adjacency matrix;
A recommended item prediction processing unit that predicts a recommended item for a user using the adjacency matrix, and outputs a degree of association with each user, using the recommended item and metadata attached to the item as related metadata,
Using the recommended item predicted by the recommended item prediction processing unit and related metadata, and an adjacency matrix obtained from the adjacency matrix providing unit for the recommendation system, to extract personalized metadata to be presented simultaneously with the item; A presentation metadata selection processing unit;
An item to recommend and a personalized metadata presentation processor.

The presentation metadata selection processing unit
The degree of association between the recommended item and the related metadata obtained from the recommended item prediction processing unit;
Present the item with the highest ratio of the relationship between the PageRank value obtained from the adjacency matrix obtained from the adjacency matrix construction processing unit for the recommendation system and the related metadata at the same time as the item from the related metadata. Select as presentation metadata.

  As described above, according to the present invention, in information recommendation, personalized metadata having a higher degree of association with a user can be extracted from metadata assigned to recommended items.

  Further, it is possible to realize information recommendation for making a recommendation by simultaneously presenting a recommended item and the personalized metadata.

It is a block diagram of the information recommendation apparatus in one embodiment of this invention. It is a flowchart of the adjacency matrix production | generation for the information recommendation system in one embodiment of this invention. It is an image of the table of the item history database in one embodiment of the present invention. It is an example of expression by the adjacency matrix of the item history database in one embodiment of the present invention. It is a flowchart of the information recommendation apparatus in one embodiment of this invention.

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

  FIG. 1 is a configuration diagram of an information recommendation device according to an embodiment of the present invention. The information recommendation apparatus shown in FIG. 1 includes a user ID acquisition unit 110, an item history database 120, an adjacency matrix providing unit 130 for a recommendation system, a recommended item prediction processing unit 140, a presentation metadata selection processing unit 150, and a presentation unit 160. Is done.

  The user ID acquisition unit 110 acquires the user ID that receives the recommendation, and outputs it to the recommended item prediction processing unit 140.

  The item history database 120 stores information on user IDs, item IDs, item metadata, and user evaluation values for items.

  FIG. 3 is an image of a table of the item history database 120 according to an embodiment of the present invention. The database is such that user IDs, item IDs, metadata IDs, and evaluation values for items can be referred to.

  The adjacency matrix providing unit 130 for the recommendation system performs processing for representing information in the item history database 120 as an adjacency matrix by a method described later.

  The recommended item prediction processing unit 140 predicts a recommended item for the user using the adjacency matrix acquired from the adjacency matrix providing unit 130 for the recommendation system, outputs the prediction item to the presentation processing unit 160, and assigns the recommended item to the recommendation system and the item. The related metadata is output to the presentation metadata selection processing unit 150 using the related metadata as the related metadata.

  The presentation metadata selection processing unit 150 uses the relevance of the related metadata acquired from the recommended item prediction processing unit 140 and the adjacency matrix acquired from the adjacency matrix providing unit 130 for the recommendation system, for each user who presents at the same time as the item. Select metadata for.

  The presentation processing unit 160 presents the recommended item acquired from the recommended item prediction processing unit 140 and the presentation metadata acquired from the presentation metadata selection processing unit 150 to the user.

  Next, using the flowchart of FIG. 2, in the adjacency matrix providing unit 130 for the recommendation system, as shown in FIG. 3, it is adjacent from the item history database 120, which is a database including user ID, item ID, metadata ID, and evaluation value. A method for constructing a matrix will be described.

The adjacency matrix here is a representation of the relationship between items and users and the relationship between items and metadata obtained by referring to the item history database 120 as a single matrix as shown in FIG. When the adjacency matrix is A and the relationship between the matrix and the element is A = (a _{i, j} ), the following procedure is used.

  Step 210) The adjacency matrix providing unit 130 for the recommendation system collects all user sets from the item history database 120.

全アイテム集合

All item sets

全メタデータ集合

Full metadata set

を収集する。

To collect.

Step 220) A matrix A = (a _{j, k} ) having a size of | U | + | I | + | M | is prepared. j represents a user when 1 ≦ j ≦ | U |, j represents an item when | U | + 1 ≦ j ≦ | U | + | I |, and j represents metadata Is the case when | U | + | I | + 1 ≦ j ≦ | U | + | I | + | M |, and k is the same. For example, a _{j, k} = 1 (| U | + 1 ≦ j ≦ | U | + | I |, | U | + | I | + 1 ≦ k ≦ | U | + | I | + | M |) , The value representing the item j and the metadata k in the matrix is 1.

Steps 231 to 235) For all items and all metadata, 1 is substituted into matrix values a _{j, k} representing the associated combinations on the item history database 120 (S233). If there is no association, 0 is substituted (S234). In the case of the database structure as shown in FIG. 4, the association on the item history database 120 indicates that each ID exists in the same entry with reference to the item ID column and the metadata ID column.

Steps 236 to 242) For all items and all users, values are substituted into matrix values a _{j, k} representing combinations associated with each other in the item history database 120. If the value is data that the user gives an evaluation value to the item in the service, the evaluation value of the item i of a certain user u is divided by the maximum value of all the evaluation values that the user u gave to the item A value is substituted (S241). In the case where there is no evaluation value on the database, for example, data such as POS data in which whether or not the user has purchased an item is recorded, 1 is assigned to the corresponding value (S240). If there is no association, 0 is substituted (S239).

  Steps 250 to 270) The matrix A is normalized so that the sum of the matrix column values becomes 1. Specifically, as shown in Expression (1), the value obtained by dividing the value by the sum of the column values is updated, and what is performed for all the columns is set as an output adjacency matrix A.

次に、図５のフローチャートに沿って、ユーザにアイテムの履歴をもとにアイテムを推薦し、また、そのアイテムに付与されたメタデータの中から同時に提示するメタデータを選択して同時に提示する処理について、図１の構成図を用いて説明する。

Next, according to the flowchart of FIG. 5, the item is recommended to the user based on the history of the item, and the metadata to be simultaneously presented is selected from the metadata attached to the item and presented simultaneously. The processing will be described with reference to the configuration diagram of FIG.

  Step 510) First, as a user ID acquisition process of a user who receives a recommendation, the user ID acquisition unit 110 acquires the ID of the user who receives a recommendation and outputs the ID to the item history database 120. This process is performed at the same time as the login process to the service or system used by the user, for example.

  Step 520) As the recommended item prediction processing unit 140, the user ID of the user who receives the recommendation input from the user ID acquisition unit 110 and the adjacency matrix obtained from the adjacency matrix providing unit 130 for the recommendation system are received.

  Step 530) The recommended item prediction processing unit 140 calculates the degree of association between the user who has acquired the user ID and each item by Random Walk with Restart (RWR). Items with a high degree of association are recommended items.

The formula for RWR is as follows: p ^(t) is a vector of item relevance at step t of calculation, constant α (0 <α <1) is the probability of returning to the starting node, A is the adjacency matrix, and q is recommended If the vector is such that only the starting point node representing the user is 1, it is expressed as in equation (2).

p ^{(t + 1)} = (1−α) Ap ^(t) + αq Equation (2)
The calculation of RWR is repeated t times on the adjacency matrix A, p ^(t) is updated, and a node having a high degree of relevance of a node representing an item from the vector of p ^(t) is output as a recommended item. Here, t is 100 times. The number of implementations is not limited to 100, and may be a sufficiently large value.

  Step 540) Further, after the recommended item prediction processing unit 140 determines the recommended item by calculating the RWR, the recommended item prediction processing unit 140 refers to the item history database 120 using the item ID as a key and sets a set of metadata associated with the item. get.

Step 550) For each of the metadata sets, the corresponding value of p ^{(t) in} the RWR calculation as the relevance of the related metadata m

を取得し、メタデータと共に提示メタデータ選択処理部１５０へ出力する。

Is output to the presentation metadata selection processing unit 150 together with the metadata.

  Step 560) As the presentation metadata selection process, the presentation metadata selection processing unit 150 receives the adjacency matrix output from the adjacency matrix providing unit 130 for the recommendation system, and calculates the PageRank of each node. The calculation of PageRank is obtained by obtaining the eigenvector r for the maximum eigenvalue of the adjacency matrix A. In general, the calculation is performed by introducing a matrix G called the Google matrix of the equation (3) and solving the eigen equation (4) for the matrix G. When the adjacency matrix A is an n-order square matrix and β is a constant satisfying 0 <β <1, the following equations are obtained. The value of β is usually 0.85 in the calculation of PageRank, but another value may be used.

ステップ５６０） また、提示メタデータ選択処理部１５０は、推薦アイテム予測処理部１４０から関連メタデータmの関連度

Step 560) Also, the presentation metadata selection processing unit 150 receives the relevance level of the related metadata m from the recommended item prediction processing unit 140.

を受け取り、その値とその関連メタデータmのPageRankの値をｒ_mとするとき、その比である

Receive, when their values the value of PageRank its associated metadata m and r _m, is the ratio

を最大にする関連メタデータmを、提示メタデータとして出力する。

The related metadata m that maximizes is output as presentation metadata.

  Step 580) As the recommended item presentation process, the presentation unit 160 presents the recommended item output from the recommended item prediction processing unit 140 and the presentation metadata output from the presentation metadata selection processing unit 150 at the same time, and the recommendation is completed. To do.

  The operation of each component of the information recommendation apparatus shown in FIG. 1 can be constructed as a program, installed on a computer used as the information recommendation apparatus, executed, 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.

110 User ID acquisition unit 120 Item history database 130 Adjacency matrix providing unit 140 for recommended system Recommended item prediction processing unit 150 Presentation metadata selection processing unit 160 Presentation unit

Claims (5)

An information recommendation device that presents personalized metadata of an item to a service user in addition to the item recommendation function in the service,
An item history database that stores information about user ID, item ID, item metadata, and user evaluation values,
An adjacency matrix construction processing unit for a recommendation system that represents the information of the item history database as an adjacency matrix;
A recommended item prediction processing unit that predicts a recommended item for a user using the adjacency matrix, and outputs a degree of association with each user, using the recommended item and metadata attached to the item as related metadata,
Using the recommended item predicted by the recommended item prediction processing unit, related metadata, and the adjacency matrix obtained from the adjacency matrix providing unit for the recommendation system, personalized metadata to be presented simultaneously with the item is extracted. A presentation metadata selection processing unit;
A presentation processing unit for presenting the recommended item and the personalized metadata to the user;
An information recommendation device comprising: The presentation metadata selection processing unit
Relevance of the recommended item and related metadata obtained from the recommended item prediction processing unit,
Present the item with the highest ratio of the relation between the PageRank value obtained from the adjacency matrix obtained from the adjacency matrix construction processing unit for the recommendation system and the related metadata at the same time as the item from the related metadata. Including means for selecting as presentation metadata to
The information recommendation device according to claim 1. An information recommendation method for presenting personalized metadata of an item to a service user in addition to the item recommendation function in the service,
Adjacency matrix construction means, an adjacency matrix construction processing step for a recommendation system, which represents information of an item history database storing information about user ID, item ID, item metadata, and evaluation values for user items as an adjacency matrix The recommended item prediction processing means predicts a recommended item to the user using the adjacency matrix, and outputs the degree of association with each user using the recommended item and metadata attached to the item as related metadata. Recommended item prediction processing step to perform,
The presentation metadata selection means presents the item at the same time using the recommended item predicted in the recommended item prediction processing step, the related metadata, and the adjacency matrix obtained in the adjacency matrix construction processing step for the recommendation system. Presenting metadata selection processing step for extracting personalized metadata to be performed;
A presentation processing step in which a presentation processing means presents the recommended item and the personalized metadata to the user;
Recommending information. In the presenting metadata selection processing step,
The degree of association between the recommended item and related metadata obtained in the recommended item prediction processing step, the PageRank value obtained using the adjacency matrix obtained in the adjacency matrix construction processing step for the recommendation system, and the related meta data The information recommendation method according to claim 3, wherein the data with the highest ratio of the degree of relevance is selected as presentation metadata to be presented simultaneously with the item from the related metadata. Computer
The information recommendation program for functioning as each means of the information recommendation apparatus of Claim 1 or 2.

Images