JP5938006B2 - Item recommendation device, item recommendation method, and item recommendation program - Google Patents

Description

  The present invention relates to an item recommendation device, an item recommendation method, and an item recommendation program for determining an item to be recommended based on Random Walk with Restart (RWR) in item recommendation.

  There is a method in which a graph composed of users and items is aggregated and constructed in consideration of domains such as a plurality of services, and an item to be recommended is determined by applying RWR (see, for example, Non-Patent Document 1).

  In addition, when calculating RWR, there is a technique for obtaining an inverse matrix using LU decomposition, which is a matrix decomposition technique suitable for obtaining an inverse matrix, and speeding up the calculation of RWR (see, for example, Patent Document 1). .

JP 2012-160015 A

Junta Tsutsuda, Makoto Nakajo, Toshiro Uchiyama, Hiroyuki Toda, Kei Uchiyama: Information recommendation in a cross-domain environment using access logs, 154th DBS / 107th IFAT Joint Research Meeting, August 2012

  However, the calculation of RWR used in the technique of Non-Patent Document 1 described above is new when the number of recommended users receiving recommendations is large, or whenever the usage history of the recommended users or the information on the assigned scores is updated. Therefore, there is a problem that a huge amount of calculation is required.

  Moreover, although the said patent document 1 is the method of calculating RWR at high speed, since LU decomposition | disassembly of the transition matrix showing a graph is required as preprocessing, depending on the structure and processing capability of the computer used, a matrix is required. Insufficient memory to read and calculation may not be possible.

  The present invention has been made in view of the above circumstances, and an object of the present invention is to provide an item recommendation device, an item recommendation method, and an item recommendation method that can apply item recommendation using RWR even to a smaller computer. To provide an item recommendation program.

  In order to solve the above problems, the first present invention is an item recommendation device using RWR (Random Walk with Restart), and includes a user ID, an item ID, a metadata ID, and an evaluation value for an item. Using the item history information acquired from the item history storage means that stores the item history information including the above, a transition matrix representing the relationship between the user and item usage history and the relationship between the item and metadata is generated and generated from the transition matrix. A pseudo inverse matrix calculating means for calculating a pseudo inverse matrix of the matrix to be processed, and a recommended item determining means for determining a recommended item using the pseudo inverse matrix, wherein the pseudo inverse matrix calculating means includes the transition matrix If the pseudo-inverse matrix cannot be calculated using the matrix generated as is, the matrix is converted into a matrix smaller than the matrix and the pseudo-inverse matrix is calculated.

  In the item recommendation device, the pseudo inverse matrix calculation unit determines whether or not the pseudo inverse matrix can be calculated using the size of the transition matrix and the resources provided in the item recommendation device.

  In the item recommendation device, the pseudo inverse matrix calculation unit cannot calculate the pseudo inverse matrix when the memory capacity of the item recommendation device is less than a size obtained by multiplying the size of the transition matrix by a predetermined number. Is determined.

  The second aspect of the present invention is an item recommendation method using RWR performed by a computer, in which item history information including a user ID, an item ID, a metadata ID, and an evaluation value for an item is stored. Using the item history information acquired from the history storage unit, a transition matrix representing a user-item usage history relationship and an item-metadata relationship is generated, and a pseudo inverse matrix of the matrix generated from the transition matrix is calculated. A pseudo-inverse matrix calculating step and a recommended item determining step of determining a recommended item using the pseudo-inverse matrix, wherein the pseudo-inverse matrix calculating step uses the matrix generated from the transition matrix as it is to When the inverse matrix cannot be calculated, the matrix is converted into a matrix smaller than the matrix, and the pseudo inverse matrix is calculated.

  In the item recommendation method, the pseudo inverse matrix calculation step determines whether or not the pseudo inverse matrix can be calculated using the size of the transition matrix and the resources provided in the computer.

  In the item recommendation method, the pseudo inverse matrix calculation step determines that the pseudo inverse matrix cannot be calculated when a memory capacity of the computer is less than a size obtained by multiplying a size of the transition matrix by a predetermined number. To do.

  The third aspect of the present invention is an item recommendation program for causing a computer to function as the item recommendation device.

  According to the present invention, it is possible to provide an item recommendation device, an item recommendation method, and an item recommendation program that can apply item recommendation using RWR even to a smaller computer.

It is a block diagram of the information recommendation apparatus in one embodiment of this invention. It is an image figure of the table of the item history database in one embodiment of this 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 pseudo | simulation inverse matrix construction | assembly process in one embodiment of this invention. It is an image figure of the transition matrix in one embodiment of the present invention.

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

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

  The information recommendation apparatus shown in the figure is an apparatus that recommends an item using RWR (Random Walk with Restart), and includes a recommended user ID acquisition unit 110, an item history database 120, a pseudo inverse matrix construction unit 130, and a recommended item prediction process. Unit 140 and item presentation unit 150.

  The recommended user ID acquisition unit 110 acquires a recommended user ID for presenting an item to be recommended from a predetermined external device or external system.

  The item history database 120 (item history storage means) stores item history information including a user ID, an item ID, a metadata ID, and an evaluation value for the item.

  The pseudo-inverse matrix construction unit 130 (pseudo-matrix calculation unit) uses the item history information acquired from the item history database 120 to generate a transition matrix that represents a user-item usage history relationship and an item-metadata relationship, A pseudo inverse matrix of a matrix generated from the transition matrix is calculated. Further, when the pseudo inverse matrix construction unit 130 cannot calculate the pseudo inverse matrix using the matrix generated from the transition matrix as it is, the pseudo inverse matrix construction unit 130 converts the matrix into a matrix smaller than the matrix and calculates the pseudo inverse matrix.

  The recommended item prediction processing unit 140 (recommended item determination means) determines a recommended item using the pseudo inverse matrix calculated by the pseudo inverse matrix construction unit 130.

  The item presentation unit 150 presents the recommended item determined by the recommended item prediction processing unit 140 to the terminal 1 of the recommended user.

  FIG. 2 is an image of a table of the item history database 120 in the present embodiment. The item history database 120 is a database in which history information of each item used by each user in the past is accumulated. The item history database 120 illustrated is a database in which a user ID, an item ID, a metadata ID, and an evaluation value for an item are stored in association with each other and can be referred to.

  Items indicate items of various genres such as contents on a computer, programs, and products sold at supermarkets. The metadata is information related to the corresponding item. For example, when an item is an application program, it is a characteristic keyword included in the description, and when the item is a unique name of a product, a corresponding category name (for example, ice cream) can be considered. The evaluation value is an index indicating the evaluation of the item corresponding to each user. For example, in the case of a reputation site of a movie, there are five-level evaluations given by the user, 0 or 1 for whether or not a page is viewed, and whether or not a purchase is made. Note that a value estimated by another external system may be used as the evaluation value.

  Note that the item history database 120 is not limited to being realized by one table as shown in FIG. 2, and may be composed of a plurality of tables. For example, it may be realized in such a manner that it has a database stored about the usage history of the user's items and a database stored about the metadata, and can be referred to individually.

  Further, when applied to a service in which data in each column of the item history database 120 does not exist, a transition matrix shown in FIG. 5 described later can be constructed, and a pseudo inverse matrix of S220 in FIG. 3 described later is constructed. As long as the process and the process of determining the recommended item in S230 can be performed, some of the rows and columns may be omitted. For example, when applied to a service in which metadata does not exist, there is no need to have metadata ID information and rows or columns on the transition matrix.

  Note that the information recommendation apparatus according to the present embodiment may be a general-purpose computer system including a CPU, a memory, an external storage device such as an HDD, and the like. In this computer system, each function of the information recommendation device is realized by the CPU executing a program for the information recommendation device loaded on the memory. The program for the information recommendation apparatus can be stored in a computer-readable recording medium such as a hard disk, a flexible disk, a CD-ROM, an MO, a DVD-ROM, or can be distributed via a network.

  Next, the processing of this embodiment will be described.

  FIG. 3 is a flowchart of the process of the information recommendation device in this embodiment. The process of presenting recommended items to the recommended user will be described with reference to the configuration diagram of FIG. 1 along the flowchart of FIG.

  Step 210) First, the recommended user ID acquisition unit 110 acquires the ID of the recommended user and outputs it to the recommended item prediction processing unit 140. For example, the recommended user ID is input from a computer such as an external device or system connected to the information recommendation device.

  Step 220) The pseudo inverse matrix construction unit 130 acquires the user ID, item ID, metadata ID, and evaluation value from the item history database 120, and constructs a pseudo inverse matrix by the process of FIG. Note that the pseudo inverse matrix construction process in step 220 is performed in advance by batch processing or the like asynchronously with step 210. Or it is good also as performing at the timing which acquired ID of the recommended user of S210.

  Step 230) The recommended item prediction processing unit 140 receives the recommended user ID from the recommended user ID acquisition unit 110 and the pseudo inverse matrix from the pseudo inverse matrix construction unit 130, and recommends the recommended user to the recommended user using RWR. To decide.

  Step 240) The item presentation unit 150 presents the recommended item determined by the recommended item prediction processing unit 140 to the recommended user. Specifically, the information of the recommended item is transmitted to the terminal of the recommended user via the network and displayed.

  In the present embodiment of FIG. 1, the information recommendation device has been described as including the item history database 120, but the item history database 120 is a database external to the information recommendation device, without being limited to this embodiment. Thus, the information recommendation device may access the item history database 120, which is an external database, and acquire data.

  In addition, for example, when a part of processing such as a calculation part of the recommended item prediction processing unit 140 is performed in a batch process, the recommended item prediction processing unit 140 temporarily stores the calculated recommended item in an external database. The recommended item may be acquired by accessing an external database as necessary and output to the item presenting unit 150.

  Next, a process in which the pseudo inverse matrix construction unit 130 constructs a pseudo inverse matrix in S220 of FIG. 3 will be described.

  FIG. 4 is a flowchart of the pseudo inverse matrix construction process. An example of the process of the pseudo inverse matrix construction unit 130 will be described with reference to FIG.

  Step 510) First, a user ID, an item ID, a metadata ID, and an evaluation value are acquired for each row from the item history database 120, and all two combinations of user ID, item ID, and metadata ID (permutation) ("User ID and item ID and evaluation value", "user ID and metadata ID and evaluation value", "item ID and user ID and evaluation value", "item ID and metadata ID and evaluation Value "," metadata ID and user ID and evaluation value "," metadata ID and item ID and evaluation value ").

  For example, in the case of a row in which the user ID is u1, the item ID is i1, the metadata ID is m1, and the evaluation value is 5, for example, “e1 and i1 have an evaluation value of 5”, “u1 and m1 Associating evaluation values such as “5”. In the case of a row for which no evaluation value is set, a fixed value such as 1 is used as the evaluation value.

Step 520) An evaluation value matrix R is constructed using an arbitrary combination of users, items and metadata obtained in Step 510 and their evaluation values. The j-th row and k-th column element of the evaluation value matrix R is defined as r _{j, k} . For example, when j is u1 and k is i1, evaluation value 5 is filled in the corresponding r _{j, k} (r _{ju1, i1} ).

Step 530) Normalization processing is performed on the evaluation value matrix R constructed in Step 520 so that the sum of the values of the respective columns becomes 1, and the transition matrix A is obtained. When the j-th row and k-th column element of the transition matrix A is expressed as a _{j, k} , the element a _{j, k} is obtained as in Expression (1) using the evaluation value r _{j, k} . FIG. 5 is an image of a transition matrix used in the present embodiment. The transition matrix is a matrix representing the usage history relationship between the user and the item and the relationship between the item and the metadata.

Step 540) Here, p is a column vector storing the relevance of the item, α is a constant satisfying 0 <α <1, A is the transition matrix obtained in Step 530, and q is only the origin node representing the recommended user Assuming that the column vector is 1, RWR is expressed as equation (2).

Furthermore, the RWR equation (2) is transformed as shown in the equation (3), and the {I- (1-α) A} portion of the obtained RWR equation (3) is generated from the matrix W (the transition matrix A Matrix). I in Equation (3) is a unit matrix.

Step 550) Then, it is determined whether or not W in the expression (3) can be processed by the information recommendation device. That is, it is determined whether or not the information recommendation apparatus can calculate the pseudo inverse matrix W- ¹ . The conditional branch of whether or not calculation is possible is determined using the size of the transition matrix A generated in step 530 and the resources implemented by the information recommendation device (computer or computer).

  For example, if the memory capacity of the information recommendation device is less than the size of the transition matrix A multiplied by a predetermined number, it may be determined that the pseudo inverse matrix cannot be calculated. That is, it is determined based on whether W can be secured in the memory of the information recommendation device. Specifically, if the memory capacity of the information recommendation device is 3 times or more the data size of the transition matrix A, it is determined that it can be calculated, and if it is smaller than 3 times the data size of the transition matrix A, the calculation is not possible. It may be possible to determine that this is possible.

  If the process is possible (S550: YES), the process proceeds to step 551. If the process is not possible (S550: NO), the process proceeds to step 552.

Step 551) A pseudo inverse matrix W ⁻¹ of W in the equation (3) is obtained. When W is subjected to singular value decomposition into Equation (4), the pseudo inverse matrix W ⁻¹ is obtained as Equation (5). Component sigma ^-1 _ii of sigma ^-1 of formula (5), when the components of the sigma and sigma _ii, obtained as _{^{σ -1 ii = 1 / σ ii}} .

Step 552) The pseudo inverse matrix W ⁻¹ is calculated while multiplying the W of the equation (3) by the matrix to be converted into a compact. W is an n-by-n matrix, and an n-by-r matrix Ω that is randomly generated is multiplied by the transpose matrix W ^T of W to obtain Equation (6). r is set to a value smaller than n in order to convert the pseudo inverse matrix W ⁻¹ into a compact form. When using a smaller computer as an information recommendation device, or when expecting faster processing, set a smaller value to r, and if a more accurate calculation result is required, set a value closer to n. Set and use. For example, when n = 10000, it may be set according to the resource of the information recommendation device, such as r = 100. The value of r is not limited to 100, and can be set within a natural number range smaller than n.

Next, the matrix B is obtained by multiplying W by Y in the equation (6) according to the equation (7). Furthermore, Z is obtained by multiplying B in equation (7) by a matrix Ψ of r rows and r columns in which elements are randomly generated, according to equation (8). According to Equation (9), C is obtained by multiplying the transpose matrix Z ^T of Z in Equation (8) by B. Using equation (10), singular value decomposition is performed on C, which has been reduced to r rows and r columns, and Σ ⁻¹ is obtained using equations (10) and (11) as in step 551, and then Z and Y ^T are ^calculated . Multiply to get W ^-1 .

Step 560) The pseudo inverse matrix W ⁻¹ obtained in step 551 or step 552 is output to the recommended item prediction processing unit 140, and the process of the pseudo inverse matrix construction unit 130 is terminated.

Finally, the process of the recommended item prediction processing unit 140 that determines an item to be recommended using the pseudo inverse matrix W ⁻¹ acquired from the pseudo inverse matrix construction unit 130 will be described. In other words, the degree of association between the acquired recommended user ID and each item ID is obtained, and the RWR (the above equation (3)) using the item with a high value as the recommended item is used.

As described above, the element of the vector representing the recommended user of q in Equation (3) is set to 1, and the node representing the item from the vector p obtained by multiplying W ^-1 by q is the recommended item that is highly related. To the item presenting unit 150.

  In this embodiment, the procedure for presenting an item to a recommended user is described. However, if an item representing q is an item (recommended item) and an output is an item, an item strongly related to the recommended item is displayed. Can be presented. Further, when q is metadata and a recommendation target is an item, it is possible to present an item strongly related to the metadata.

  In the present embodiment described above, when the pseudo inverse matrix cannot be calculated using the matrix generated from the transition matrix as it is, the matrix is converted into a matrix smaller than the matrix, and the pseudo inverse matrix is calculated. Thereby, in this embodiment, information recommendation using RWR can be performed even for a small-scale computer. That is, it is possible to avoid a situation in which calculation is not possible due to a lack of memory for reading a matrix depending on the configuration and processing capability of the computer to be used.

  In addition, this invention is not limited to the said embodiment, A various change and application are possible within a claim.

110: Recommended user ID acquisition unit 120: Item history database 130: Pseudo inverse matrix construction unit 140: Recommended item prediction processing unit 150: Item presentation unit

Claims (7)

An item recommendation device using RWR (Random Walk with Restart),
Using the item history information acquired from the item history storage means storing the item history information including the user ID, item ID, metadata ID, and evaluation value for the item, the usage history relationship between the user and the item and the item A pseudo-inverse matrix calculating means for generating a transition matrix that represents the relationship between the metadata and the metadata, and calculating a pseudo-inverse matrix of the matrix generated from the transition matrix;
Recommended item determination means for determining a recommended item using the pseudo inverse matrix,
The pseudo inverse matrix calculation means calculates the pseudo inverse matrix by converting the matrix into a matrix smaller than the matrix when the pseudo inverse matrix cannot be calculated using the matrix generated from the transition matrix as it is. Item recommendation device. The item recommendation device according to claim 1,
The item recommendation device, wherein the pseudo inverse matrix calculation means determines whether or not the pseudo inverse matrix can be calculated using a size of the transition matrix and a resource included in the item recommendation device. The item recommendation device according to claim 1 or 2,
The pseudo inverse matrix calculating means determines that the pseudo inverse matrix cannot be calculated when the memory capacity of the item recommendation device is less than a size obtained by multiplying the size of the transition matrix by a predetermined number. Item recommendation device. An item recommendation method using RWR (Random Walk with Restart) performed by a computer,
Using the item history information acquired from the item history storage unit storing the item history information including the user ID, the item ID, the metadata ID, and the evaluation value for the item, the usage history relationship between the user and the item and the item A pseudo-inverse matrix calculating step of generating a transition matrix representing a relationship between the meta-data and the metadata, and calculating a pseudo-inverse matrix of the matrix generated from the transition matrix;
Performing a recommended item determining step for determining a recommended item using the pseudo inverse matrix;
The pseudo inverse matrix calculation step calculates the pseudo inverse matrix by converting the matrix to a matrix smaller than the matrix when the pseudo inverse matrix cannot be calculated using the matrix generated from the transition matrix as it is. Item recommendation method characterized by The item recommendation method according to claim 4,
The item recommendation method, wherein the pseudo inverse matrix calculation step determines whether or not the pseudo inverse matrix can be calculated by using the size of the transition matrix and resources provided in the computer. The item recommendation method according to claim 4, wherein:
The pseudo-inverse matrix calculating step determines that the pseudo-inverse matrix cannot be calculated when the memory capacity of the computer is less than a size obtained by multiplying the size of the transition matrix by a predetermined number. Item recommendation method.   An item recommendation program for causing a computer to function as the item recommendation device according to any one of claims 1 to 3.

Images