JP2009289092A - Information processor, recommendation method, program and recording medium - Google Patents

Abstract

It is possible to flexibly combine various algorithms in content recommendation.
An identifier having profile data representing a user attribute in at least one of a row direction and a column direction and a data item of metadata representing a feature of the content, and identifying the user in at least one of the row direction and the column direction Data for acquiring an arbitrary submatrix from the data holding unit using a data holding unit for holding data by a matrix-type data structure having an identifier data item for identifying content and a specifier for specifying a row or a column An operation unit, a basic algorithm unit that performs a typical calculation process on the profile data and the metadata included in the partial matrix acquired by the data operation unit, and the routine calculation performed by the basic algorithm unit. Use the results to create a recommendation list that lists the content to recommend to users And Application algorithm unit which Tokusuru, to provide an information processing apparatus including a.
[Selection] Figure 7

Description

  The present invention relates to an information processing apparatus, a recommendation method, a program, and a recording medium.

  In recent years, a technique called recommendation for differentiating services provided through a network such as the Internet has been widely used. The recommendation is a technique for promoting the use of content by selecting and presenting content suitable for each user based on the user's preference and behavior, for example. Recommendations are used for Internet shopping and content distribution services such as video, music or news articles.

  In order to realize a recommendation service, an arbitrary recommendation algorithm is usually installed in a service providing system. The recommendation algorithm is individually designed in accordance with the purpose of providing the service, the target user, or the content. However, even though the recommendation results and the display formats that are ultimately provided to the user are different, there are few common algorithms among many recommended services for effective recommendation. Absent. For example, Patent Documents 1 and 2 listed below disclose information processing apparatuses that can make recommendations by applying an algorithm called a collaborative filter (Collaborative Filtering). In addition to the collaborative filter, a common algorithm such as a content-based filtering, a neural network, or a Bayesian estimation may be used for the recommendation service.

JP 2006-215867 A JP 2008-077386 A

  However, in the conventional recommendation service, the basic algorithm such as the collaborative filter and content filter described above and the application algorithm that generates the final recommendation result according to the needs of each service are integrated. It was common to do. For this reason, when developing a new recommendation service, it is necessary to redesign most of the recommendation algorithm and change the data model, resulting in an increase in the cost for service development.

  Accordingly, the present invention has been made in view of the above problems, and an object of the present invention is to provide a new and improved information processing apparatus, recommendation method, and program capable of flexibly combining various algorithms. And providing a recording medium.

  In order to solve the above-described problem, according to an aspect of the present invention, profile data representing user attributes and metadata data items representing content characteristics are provided in at least one of a row direction and a column direction, and Using a data holding unit for holding data in a matrix format data structure having an identifier for identifying a user in at least one of the direction and the column direction and an identifier data item for identifying content, and a specifier for designating a row or column A data operation unit that acquires an arbitrary partial matrix from the data holding unit, and a basic algorithm unit that executes a typical calculation process for the profile data and the metadata included in the partial matrix acquired by the data operation unit And using the result of the routine calculation performed by the basic algorithm unit, The information processing apparatus and a application algorithm unit for acquiring a recommendation list that lists the content that is provided.

  According to such a configuration, the data holding unit has profile data representing user attributes and metadata data representing content characteristics in at least one of the row direction and the column direction, and at least in the row direction or the column direction. On the other hand, the profile data and the metadata are held by a data structure in a matrix format having an identifier for identifying the user and a data item for identifying the content. Then, the data operation unit acquires an arbitrary partial matrix from the data structure in the matrix format using a specifier that designates a row or a column. Further, the basic algorithm unit performs a typical calculation process on the profile data and the metadata included in the partial matrix acquired from the matrix format data structure. Then, the application algorithm unit acquires a recommendation list in which contents recommended to the user are listed using the result of the routine calculation.

  The specifier may be a hierarchically expressed character string that designates one of the nodes in a hierarchical structure including nodes corresponding to data and nodes corresponding to a category of data.

  The basic algorithm unit may calculate the degree of correlation of the profile data between a specified user and another user as one of the routine calculations.

  The basic algorithm unit may calculate the degree of correlation between the profile data of the designated user and the metadata of an arbitrary content as one of the routine calculations.

  In addition, the basic algorithm unit may include, as one of the routine calculations, the degree of correlation of the profile data between the specified user and another user, or the profile data of the specified user and any content The recommendation list may be generated based on the degree of correlation with the metadata.

  Further, the basic algorithm unit may calculate a correlation degree of the metadata between the designated content and other content as one of the routine calculations.

  The basic algorithm unit may calculate a correlation degree between the metadata of the designated content and the profile data of an arbitrary user.

  Further, the data holding unit may have a plurality of data values in one field in the matrix format data structure.

  The application algorithm unit further performs the action according to the metadata of the content that is the target of the action included in the action history data when the action history data of the user is received from the terminal device. The profile data of the user may be updated.

  The information processing apparatus may further include a database input / output unit that requests the database to make the data held in the data holding unit permanent based on an instruction from the data holding unit.

  In addition, the data holding unit may acquire data in a range assigned to the information processing apparatus from the database via the database input / output unit among the data items of the profile data and the data items of the metadata. Good.

  In order to solve the above problem, according to another aspect of the present invention, profile data representing user attributes and metadata data items representing content characteristics are provided in at least one of a row direction and a column direction. And holding the profile data and the metadata by a data structure in a matrix format having a data item of an identifier for identifying a user and an identifier for identifying a content in at least one of a row direction and a column direction, and a row or a column Obtaining an arbitrary submatrix from the data structure of the matrix format using a specifier that designates, and the profile data and metadata included in the submatrix acquired from the data structure of the matrix format Using the step of performing a typical calculation process and the result of the routine calculation, The method comprising the steps of: obtaining a recommendation list which shows the content to be recommended to the over The recommendation methods including is provided.

  In order to solve the above-described problem, according to another aspect of the present invention, a computer that controls an information processing apparatus is provided with profile data and content characteristics that represent user attributes in at least one of a row direction and a column direction. A data holding unit for holding data in a matrix format data structure having metadata data items to represent and having an identifier for identifying a user in at least one of a row direction and a column direction and an identifier data item for identifying content A data operation unit that acquires an arbitrary submatrix from the data holding unit using a specifier that specifies a row or a column, and the profile data and the metadata included in the submatrix acquired by the data operation unit A basic algorithm unit that performs a typical calculation process for Using the results of the routine calculations, the application algorithm unit for acquiring a recommendation list that lists the contents to be recommended to the user, the program to function as is provided.

  In order to solve the above-described problem, according to another aspect of the present invention, a computer that controls an information processing apparatus is provided with profile data and content characteristics that represent user attributes in at least one of a row direction and a column direction. A data holding unit for holding data in a matrix format data structure having metadata data items to represent and having an identifier for identifying a user in at least one of a row direction and a column direction and an identifier data item for identifying content A data operation unit that acquires an arbitrary submatrix from the data holding unit using a specifier that specifies a row or a column, and the profile data and the metadata included in the submatrix acquired by the data operation unit A basic algorithm unit that performs a typical calculation process for Provided is a computer-readable recording medium in which a program for functioning as an application algorithm unit that acquires a recommendation list that lists contents recommended to a user by using the result of the routine calculation is recorded. The

  In order to solve the above problem, according to another aspect of the present invention, a terminal device including a user application unit that requests content recommendation from an information processing device, and a user in at least one of a row direction and a column direction In the form of a matrix having profile data representing attributes and metadata data items representing content characteristics, and identifiers identifying users in at least one of a row direction and a column direction, and identifier data items identifying the contents A data holding unit that holds data according to a data structure, a data operation unit that acquires an arbitrary partial matrix from the data holding unit using a specifier that specifies a row or a column, and the partial matrix acquired by the data operation unit Basics for performing routine calculation processing on the profile data and the metadata included Application algorithm unit that acquires a recommendation list that lists contents recommended to the user using a result of the routine calculation by the algorithm unit and the basic algorithm unit, and transmits the acquired recommendation list to the terminal device And an information processing system including the information processing apparatus.

  As described above, according to the information processing apparatus, recommendation method, program, and recording medium according to the present invention, various algorithms can be combined flexibly.

  Exemplary embodiments of the present invention will be described below in detail with reference to the accompanying drawings. In addition, in this specification and drawing, about the component which has the substantially same function structure, duplication description is abbreviate | omitted by attaching | subjecting the same code | symbol.

[1] First Embodiment First, FIG. 1 is a schematic diagram showing a configuration of a recommendation system 1 according to a first embodiment of the present invention. Referring to FIG. 1, the recommendation system 1 includes a terminal device 10, a terminal device 12, a network 20, a Web server 22, a recommendation server 30, and a DB (Database) server 32.

  The terminal device 10 and the terminal device 12 are connected to the recommendation server 30 via the network 20 and the Web server 22. The terminal device 10 and the terminal device 12 transmit a recommendation request to the recommendation server 30 according to a user operation, and present the recommendation result received from the recommendation server 30 via the network 20 to the user. In FIG. 1, a PC (Personal Computer) is shown as an example of the terminal device 10 and the terminal device 12, but the terminal device 10 and the terminal device 12 are not limited to the PC. The terminal device 10 and the terminal device 12 may be information processing devices such as a mobile phone, a PDA (Personal Digital Assistant), a game terminal, a music / video playback device, or a home appliance.

  The network 20 is a network that connects between the terminal device 10 and the terminal device 12 and the Web server 22. For example, when the recommendation system 1 provides a recommendation service on the Internet, the network 20 corresponds to the Internet. However, the network 20 is not limited to the Internet. The network 20 may be any network such as a LAN (Local Area Network), a WAN (Wide Area Network), or a dedicated line.

  The web server 22 is connected to the network 20 and the recommendation server 30. The Web server 22 plays a role of mediating a recommendation request transmitted from the terminal device 10 or the terminal device 12 to the recommendation server 30 via the network 20 and a response to the recommendation request from the recommendation server 30. In addition, when the Web server 22 receives a request for browsing a static Web page, for example, instead of a recommendation request from the terminal device 10 or the terminal device 12, the Web server 22 acquires the content of the requested Web page and obtains the request source. You may respond to the device.

  In the recommendation system 1, a server device for an arbitrary service that mediates a recommendation request may be arranged between the network 20 and the recommendation server 30 instead of the Web server 22. Further, the recommendation system 1 may be configured by omitting the Web server 22.

  The recommendation server 30 is connected to the Web server 22 and the DB server 32. The recommendation server 30 receives a content recommendation request from the terminal device 10 or the terminal device 12, and transmits a recommendation result acquired by performing calculations based on various algorithms described later to the terminal device 10 or the terminal device 12. The recommendation server 30 acquires content metadata and user profile data required for the calculation from the DB server 32.

  For example, when the recommendation system 1 handles music data as content, a list of songs that prompt the user to view or purchase is transmitted as a recommendation result from the recommendation server 30 in response to a request from each terminal device. . When the recommendation system 1 handles television program guide information as content, a list of programs that the user is expected to be interested in is requested from the recommendation server 30 in response to a request from each terminal device. Sent as a recommendation result.

  When the recommendation system 1 provides a recommendation service on the Internet, the recommendation server 30 is realized as a Web application server, for example.

  The DB server 32 is connected to the recommendation server 30. The DB server 32 records content metadata, user profile data, and the like necessary for the recommendation calculation in the storage device. Further, the DB server 32 centrally manages reading and writing of data from the recommendation server 30.

  Here, the content metadata is data representing the characteristics of the content expressed by various data items corresponding to the content. For example, when music data is handled as content, the music genre, target age group, tempo, key, etc. can be handled as music data metadata. In addition, when video data is handled as content, scene information and names of characters included in the video can be handled as metadata.

  The content metadata may be artificially given at the time of data registration, or may be acquired using techniques such as acoustic analysis of music and scene analysis of moving images.

  The user profile data is data representing attributes such as the user's age group, hobbies, and preference trends. In the present embodiment, user profile data is also represented by data items similar to content metadata. For example, the user's hobbies and preferences for music content can be expressed as scores for genres such as “Japanese music” and “Western music” of the music content.

  The user's profile data is acquired or updated by analysis of the user's behavior history, in addition to being registered in advance, such as an age group. The user action history is information that records when, what site or page the user visits, and what action the user performs. As user action history data, for example, actions such as content reference, addition to the attention list, viewing, and purchase can be recorded together with the user identifier and time information.

  FIG. 2 is an explanatory diagram illustrating a state in which the user profile data is updated according to the user action history.

  Referring to FIG. 2, first, a new user represented by an identifier “U001” accesses a Web site provided by the recommendation system 1 (A). At this time, the profile data of the user U001 is “Western music” = 0 points and “Japanese music” = 0 points.

  Thereafter, the user U001 is viewing Western music A on the website (B). Thereby, 10 points are added to the item “Western music” in the profile data of the user U001. Note that the point value added here may be any value.

  Next, the user U001 purchases Japanese music B on the website (C). Thereby, 100 points are added to the item “Japanese music” in the profile data of the user U001. In this way, for example, profile data indicating that the user U001 has a stronger preference for Japanese music than Western music is generated.

  FIG. 3 is an explanatory diagram showing another example of changing the user profile data in accordance with the user action history. In FIG. 3, the recommendation system 1 is premised on handling news articles as contents.

Referring to FIG. 3, first, when a news article represented by an identifier “NW01” is registered, metadata of the article NW01 is acquired by content analysis (A). For example, a news article can be analyzed by TF / IDF (Term Frequency / Inverse Document Frequency), which is known as a technique for analyzing the content of an article represented by text. In the example of FIG. 3, as the metadata of the article NW01, since the terms “Tokyo” and “baseball” are included in the article content, the feature amount is 10 points for “Tokyo” and 20 points for “baseball”. Is given.

  Thereafter, when the user U002 accesses the news article distribution site (B) and refers to the article NW01 (C), the metadata of the article NW01 “Tokyo” = 10 points and “sports” = 20 in the profile data of the user U002. Each point value is added.

  Note that FIG. 3 illustrates an example in which profile data is updated when a user refers to an article. However, for example, the user's behavior history may be accumulated for an arbitrary period, the behavior history may be analyzed at a timing such as when the user leaves the website or once a day, and the profile data may be updated.

  In the recommendation system 1 shown in FIG. 1, content recommendation for a user is performed using the metadata of the content to be recommended and the profile data of the user.

  FIG. 4 is a block diagram showing a logical functional arrangement of the recommendation system 1. Here, for convenience of explanation, only the terminal device 10 is shown among the terminal devices 10 and 12 shown in FIG. The terminal device 12 is configured in the same manner as the terminal device 10. Further, the network 20 and the Web server 22 that do not directly affect the content of the recommendation process are not shown.

  Referring to FIG. 4, the terminal device 10 includes a user application unit 110. The user application unit 110 requests the recommendation server 30 to recommend content, and presents a list of recommended content returned from the recommendation server 30 (hereinafter referred to as a recommendation list) to the user of the terminal device 10. Further, the user application unit 110 may collect user action history data and transmit the collected action history data to the recommendation server 30.

  The user application unit 110 can be realized using, for example, a web browser or a plug-in introduced into the web browser. When the recommendation system 1 is a client-server type system, the user application unit 110 may be realized as dedicated client software.

  Next, referring to FIG. 4, the recommendation server 30 includes an application interface unit 120, an application algorithm unit 130, a basic algorithm unit 140, a data operation unit 150, a data holding unit 160, and a database input / output unit 170.

  The application interface unit 120 is an interface for receiving a recommendation request from the user application unit 110 and returning a recommendation result. The application interface unit 120 can be realized as a function group based on RPC (Remote Procedure Call) based on, for example, SOAP (Simple Object Access Protocol). Further, for example, the application interface unit 120 may be realized by passing parameters based on simpler HTTP (HyperText Transfer Protocol).

  In this embodiment, a recommendation algorithm for recommending content is roughly classified into an application algorithm unit 130 and a basic algorithm unit 140 described below.

  The application algorithm unit 130 processes a recommendation request according to the needs of each recommended service, processes the recommendation result, and obtains a final recommendation list to be returned to the user application unit 110. For example, the application algorithm unit 130 may rearrange the recommendation list or limit the number of displays in accordance with needs or restrictions on screen display. The applied algorithm unit 130 may integrate a plurality of recommendation lists output from the basic algorithm unit 140 or add related information.

  Furthermore, when the user's behavior history data is received from the user application unit 110, the application algorithm unit 130 may update the profile data of the user who performed the behavior by analyzing the behavior history data.

  The basic algorithm unit 140 has a function of executing a standardized algorithm used for recommendation. As described above, among various recommendation services that are currently in practical use, there are not a few common basic algorithms for effective recommendation. For example, a collaborative filter and a content filter described below are examples of algorithms commonly used in many recommended services.

  FIG. 5 is an explanatory diagram showing an example of recommendation processing by the cooperative filter.

  Referring to FIG. 5, the degree of correlation of profile data between the user U011 and the other users U012, U013, and U014 is first calculated after receiving the recommendation request from the user U011 (A). The correlation degree refers to a value indicating the degree of association between a plurality of profile data or metadata. In the example of FIG. 5, the degree of correlation represented by the sum of the difference values for each data item is used (that is, the smaller the value, the higher the degree of association).

  For example, the correlation between the user U011 and the user U012 is calculated as | 500-0 | + | 100-300 | + | 0-0 | = 700. Similarly, the correlation degree between the user U011 and the user U013 is calculated as 300, and the correlation degree between the user U011 and the user U014 is calculated as 900.

  Note that the degree of correlation between data is not limited to this example. For example, the user's profile may be regarded as a vector, and a direction cosine representing a similarity between profile data, a standardized distance, a Hamming distance, or the like may be calculated and used as the degree of correlation.

  Next, the users U012, U013, and U014 are ranked using the calculation result of the correlation degree between users (B). In the example of FIG. 5, the user U013, the user U012, and the user U014 are ranked in order from the smaller calculated correlation degree.

  Then, recommended content is determined based on the profile data of the user positioned higher as a result of ranking (C). In the example of FIG. 5, music A and music C having a high score are selected from the profile data of the user U013 having the smallest degree of correlation (that is, most relevant to the user U011). Here, the recommended content is determined from the profile data of the user most relevant to the user U011, but the method for determining the recommended content is not limited to this example. For example, the recommended content may be determined from past purchase history data of the users with the highest degree of correlation.

  FIG. 6 is an explanatory diagram showing an example of a recommendation process using a content filter as another example of an algorithm commonly used for recommendation. In FIG. 6, the news article is the content to be recommended.

  Referring to FIG. 6, first, the degree of correlation between the profile data of the user U021 and the metadata of the contents (ie, news articles) NW11, NW12, and NW13 is calculated in response to receiving a recommendation request from the user U021. (A). Note that the metadata of each content is given in advance by the above-described content analysis or the like.

  For example, the correlation degree between the user U021 and the news article NW11 is calculated as 190 as in the example of FIG. Further, the correlation degree between the user U021 and the news article NW12 is calculated as 290, and the correlation degree between the user U021 and the news article NW13 is calculated as 20.

  Next, news articles are ranked using the calculation result of the degree of correlation between the user profile data and the content metadata (B). In the example of FIG. 6, the news articles NW13, NW11, and NW12 are ranked in order from the smaller calculated correlation degree.

  Then, news articles NW13 and NW11 that are positioned higher as a result of ranking are determined as recommended contents (C). Here, content positioned higher as a result of ranking is recommended, but the criterion for extracting recommended content is not limited to this example. For example, content extracted using a predetermined threshold for the degree of correlation may be recommended.

  In the example of FIG. 5, the degree of correlation of profile data between the designated user and other users, and in the example of FIG. 6, the degree of correlation between the designated user's profile data and metadata of an arbitrary content However, the calculation of the degree of correlation is not limited to this example. For example, in the recommendation request from the user application unit 110, content may be specified instead of the user, and the correlation degree of metadata between the specified content and other content may be calculated. Further, the degree of correlation between the specified content metadata and the profile data of an arbitrary user may be calculated.

  Returning to the description of FIG. 4, various general-purpose algorithms used for such recommendation are mounted in the basic algorithm unit 140 in a state of being partized as a routine calculation process. As general-purpose algorithms other than those described here, for example, a recommendation algorithm for performing learning based on a neural network, a probabilistic recommendation algorithm based on a Bayesian network, and the like may be implemented in the basic algorithm unit 140. Further, basic calculation processing such as similarity calculation and matrix four arithmetic calculations may be implemented in the basic algorithm unit 140.

  The data operation unit 150 provides an interface for reading and writing data from the application algorithm unit 130 or the basic algorithm unit 140 to the data holding unit 160 described later. For example, the basic algorithm unit 140 can read user profile data from the data holding unit 160 via the data operation unit 150 when recommending content based on the collaborative filter. Further, for example, the application algorithm unit 130 can create profile data based on the user's behavior history data and write the profile data to the data holding unit 160 via the data operation unit 150. Data reading / writing from the data operation unit 150 to the data holding unit 160 is performed by a submatrix operation using a specifier that designates a row or a column, as will be described later.

  The data holding unit 160 centrally holds user profile data and content metadata in the recommendation server 30 using a data structure in a matrix format described in detail below.

  Here, as can be understood from FIG. 2 or FIG. 3, the data items of the user profile data and the data items of the content metadata include items common to each other. For example, a genre (such as “Western music” or “Japanese music”), which is metadata of music content, is used as a data item of user profile data in FIG. Further, for example, a profile data item “Tokyo” corresponding to a user's regional attribute or the like is used as a metadata data item of news article content in FIG. 3. Therefore, the user profile data and the content metadata can be held in a matrix-type data structure having data items of both data in at least one of the row direction and the column direction.

  FIG. 7 is an explanatory diagram for explaining the data structure of the matrix format in the data holding unit 160 according to the present embodiment.

  Referring to FIG. 7, in the row direction of the data structure in the matrix format of the data holding unit 160, the user (“U111”, “U222”), content (“Song A”, “Song B”), regional attribute (“ Eight data items are arranged, “Tokyo”, “Osaka”), and genre (“Western music”, “Japanese music”). Also, the same eight data items are arranged in the column direction of the matrix-type data structure. Each square cell (hereinafter referred to as a field) holds a data value represented by a numerical value.

  In FIG. 7, the meaning of the data value of each field differs depending on the position in the data structure of the field. For example, data held in a region D1 corresponding to a user row, a region attribute, and a genre column corresponds to user profile data. Data held in the region D2 corresponding to the content row, the region attribute, and the genre column corresponds to content metadata.

  Further, in the area D3 corresponding to the user row and the user column, for example, the correlation degree between users obtained as a result of the above-described collaborative filter calculation can be stored. In the area D4 corresponding to the user's row and content column, for example, data obtained by quantifying the action history for the user's content used for the above-described collaborative filter calculation can be stored. Instead, the area D4 may store, for example, the degree of correlation between the user and the content obtained as a result of the above-described content filter calculation. Further, in the area D5 corresponding to the content row and the content column, for example, the correlation degree between the contents used for the calculation of the content filter described above may be stored.

  Furthermore, in a region D6 in which both rows and columns correspond to regional attributes or genres, the weights of links between nodes (associations between attributes) when making a recommendation based on, for example, a neural network may be stored. The region D6 may store a probability value or the like when making a recommendation based on a Bayesian network.

  Here, as described above, the application algorithm unit 130 or the basic algorithm unit 140 reads and writes data held in the data holding unit 160 via the data operation unit 150. Assuming the data structure shown in FIG. 7, for example, a function of acquiring and outputting the regions D1 to D6 which are partial matrices of the data holding unit 160 can be included in the function of the data operation unit 150. More generally, the data operation unit 150 reads / writes data in a submatrix in a range specified by using a row specifier and / or a column specifier described later in the matrix data structure of the data holding unit 160. Can be implemented to provide functionality.

In an information system not according to the present invention, data is usually classified for each entity, normalized, and retained (may be partially denormalized due to performance requirements). For example, user data, user profile data, correlation between users, content data, content metadata, and the like are considered as separate entities. In the information system not according to the present invention, the recommendation processing is performed after the data acquired from these separate entities is reconfigured into a format suitable for each recommendation algorithm.

  However, as can be understood from the description so far, in the processes such as the collaborative filter and the content filter commonly used for the recommendation service, a matrix or a vector in which similar data items are arranged in the row direction or the column direction. The calculation process for is the main. Therefore, as in the present embodiment, by holding data in a matrix format data structure having data items of user profile data and content metadata in at least one of the row direction and the column direction, a lot of calculation processing is performed. It can be simplified. As a result, various algorithms can be flexibly combined.

  7 has the same eight data items in the row direction and the column direction, but the data items in the row direction and the column direction do not necessarily match. Also good. For example, the recommendation system 1 may be implemented by omitting the four rows from the bottom of the matrix format shown in FIG. 7, that is, the rows corresponding to the category of the region attribute and the genre.

  Returning to FIG. 4, the description of the logical functional arrangement of the recommendation system 1 according to the present embodiment will be continued.

  Data held in a matrix format in the data holding unit 160 is data held in the memory of the recommendation server 30. A database input / output unit 170 is provided in order to perpetuate such data in the DB server 32.

  The database input / output unit 170 receives an instruction from the data holding unit 160 and requests the database management unit 180 of the DB server 32 to make the data permanent and read the data. Data persistence refers to recording data in a physical storage device. The database management unit 180 receives a data input / output request from the database input / output unit 170 and centrally manages data perpetuation and data reading using the storage unit 190.

  The database input / output unit 170 or the database management unit 180 may have a cache function. For example, the database management unit 180 maintains the response contents for past inquiries while maintaining the data structure in the matrix format shown in FIG. 7, thereby returning a response to inquiries about data that has not been updated at high speed. be able to.

  The database input / output unit 170 and the database management unit 180 can be realized by applying or extending middleware that provides a general-purpose RDBMS (Relational DataBase Management System) function, for example.

  The storage unit 190 is a storage area for storing various data held by the recommendation system 1. The storage unit 190 can be implemented using, for example, a storage device 916 of a general-purpose computer described later.

  Note that when the data of the recommendation system 1 is made permanent in the DB server 32, the data need not necessarily be recorded in the matrix format described with reference to FIG. For example, the data may be made permanent after being converted into a normalized relational database format or XML (extensible Markup Language) format.

  So far, the logical functional arrangement of the recommendation system 1 has been described. Next, a data access method in the data holding unit 160 of this embodiment will be described.

  As described above, the data holding unit 160 holds various types of data using a matrix-type data structure having data items of user profile data and content metadata in at least one of the row direction and the column direction. In such a matrix, since some fields are not used in actual processing, the matrix held by the data holding unit 160 is a so-called sparse matrix. In response to a request from the application algorithm unit 130 or the basic algorithm unit 140, the data operation unit 150 extracts a partial matrix for data reference or update for such a sparse matrix.

  The extraction of the partial matrix from the data holding unit 160 via the data operation unit 150 can be performed using, for example, a row specifier and a column specifier. A row specifier and a column specifier are identifiers that serve to specify individual rows or columns of a matrix format data structure. In the present embodiment, in order to perform data access to the data holding unit 160 easily and at high speed, character strings expressed hierarchically according to the category of data are used as row specifiers and column specifiers.

  FIG. 8 is an explanatory diagram showing a data access method using a row specifier and a column specifier in the data holding unit 160.

  Referring to FIG. 8, each row of data items held in the data holding unit 160 has a three-level hierarchical structure including a first hierarchy and a second hierarchy corresponding to the data category, and a third hierarchy corresponding to the data item. ((A) to (C)). One node in the hierarchical structure is specified by a row specifier and a column specifier, which are character strings expressed hierarchically.

  For example, the row specifier “1” for designating up to the first layer corresponds to “user”, the row specifier “2” corresponds to “content”, and the row specifier “3” corresponds to “attribute” (A). The row specifier “1-1” for designating up to the second hierarchy is “general user”, “1-2” is “special user”, “2-1” is “music”, and “2-2” is “Movie” and “3-1” correspond to “Regional attribute”, and “3-2” corresponds to “Genre” (B). Further, the row specifier = “1-1-1” for designating up to the third hierarchy corresponds to the user U111, “1-1-2” corresponds to the user U112, etc. (C). Here, the layers are separated by a hyphen symbol “-”, but the hyphen symbol “-” may be omitted.

  Such a row designator “XX-X” identifies, for example, one row in the matrix format of FIG. That is, for example, when the basic algorithm unit 140 according to the present embodiment acquires data for one row of the user U 111, the row specifier “1-1-1” is received from the basic algorithm unit 140 to the data operation unit 150. Passed. Then, the data operation unit 150 acquires data under the node of the user U111 corresponding to the row specifier “1-1-1” as data for one row and outputs the data to the basic algorithm unit 140.

  In addition to the row specifier, one field in the matrix format data structure can be specified using column specifiers (A ′ to C ′) similarly hierarchized into three hierarchies. For example, the field specifier “1-1-1-3-1-1” designates a node corresponding to the data of the regional attribute “Tokyo” in the profile data of the user U111. The field specifier “2-1-1-3-1-1” designates a node corresponding to the data of the regional attribute “Tokyo” in the metadata of the music piece A.

  Furthermore, in terms of program implementation, a pointer that points to an address where data is physically stored is held at the end of the hierarchical structure specified by the field specifier (D). Then, the actual data is stored at the address indicated by the pointer (E). For example, in FIG. 8, the profile data value “100” for the region attribute “Tokyo” of the user U111 is assigned to the address “AF05” indicated by the end pointer of the field specifier “1-1-1-3-1-1”. Is stored.

  Here, the method of accessing data held in the data holding unit 160 using a field specifier in which a row specifier is combined in the first half and a column specifier in the second half has been described. That is, as shown in FIG. 8, on the data model, the start point side of the hierarchical structure corresponds to the row and the end point corresponds to the column. However, in implementing the recommendation algorithm, it may be required to refer to or update data by specifying a column first. Therefore, in addition to the data access method shown in FIG. 8, a hierarchical data access method that can access data in the data holding unit 160 in a form in which a column specifier is combined in the first half and a row specifier in the second half is combined. It is preferable to mount them together.

  The hierarchical data access method described with reference to FIG. 8 can be realized by defining nodes in each layer using classes in an object-oriented programming language, for example.

  One of the merits of realizing the hierarchical data access method as shown in FIG. 8 with respect to the data structure in the matrix format (corresponding to a logical data model) described with reference to FIG. Is to become. It is known that the data access by the hierarchized identifier is faster than the data access by the successive comparison with respect to the simple one-dimensional array data based on the concept of the binary search. By such a data access method, even if the data structure held by the data holding unit 160 becomes a large-scale sparse matrix, high-speed data access that can sufficiently withstand practical use can be performed. In addition, by using numerical values for the hierarchized identifiers, it is also possible to cut out a partial matrix by flexibly specifying a range according to the magnitude relation of the numerical values.

  Furthermore, when nodes in each hierarchy are defined using classes, each node is an object generated from the class. In a normal recommendation system not according to the present invention, for example, when there is data access to one user or one content, the entire data related to the user or content to be accessed is locked while the data is being accessed. . Locking is a process for preventing data from being simultaneously manipulated from a plurality of processes and ensuring data consistency. However, as described above, if the nodes of each hierarchy in FIG. 8 are each one object, the range to be locked can be limited to individual nodes, that is, the submatrix or field of the range to be processed. Thereby, even when a plurality of processes are operated in parallel, processing delay due to waiting for lock release is reduced.

  In FIG. 8, only one actual data is held in each field that is the end of hierarchical data, but each field may hold a plurality of actual data instead of one actual data. For example, by assigning a plurality of pointers to the end of the hierarchical data in FIG. 8D, a plurality of actual data can be held. According to such a configuration, for example, profile data based on a session history or a daily activity history on the Web and profile data based on all past behavior histories can be held in parallel. By doing so, it is possible to construct a service for making a recommendation from a multifaceted viewpoint.

  So far, the data structure and the data access method for the data structure in the recommendation system 1 of the present embodiment have been described. Next, the flow of processing in the recommendation system 1 will be described using FIG. 9 and FIG.

  FIG. 9 is a sequence diagram showing a flow of user profile data generation processing in the present embodiment.

  9, first, user action history data is transmitted from the user application unit 110 of the terminal device 10 to the application algorithm unit 130 of the recommendation server 30 (S1104). As described with reference to FIG. 4, requests and responses between the user application unit 110 and the application algorithm unit 130 are made via the application interface unit 120. However, for convenience of explanation, the application interface unit 120 is omitted here.

  As described above, the behavior history data is transmitted from the user application unit 110 when the user performs an action such as browsing or purchasing content, when the user leaves the website, or once a day. Done at the moment. The behavior history data transmitted here typically includes an identifier that identifies the user, an identifier that identifies the content that is subject to behavior such as reference or purchase, and a code that identifies the type of behavior. .

  Next, the application algorithm unit 130 that has received the action history data requests the data operation unit 150 to acquire content metadata (S1108). For example, if the action history data includes identifiers that specify the music piece A and the music piece B, acquisition of metadata in the area D2 shown in FIG. 7 is requested.

  Upon receiving the request from the application algorithm unit 130, the data operation unit 150 refers to the data holding unit 160 and acquires a partial matrix in the data holding unit 160 corresponding to the metadata of the content (S1112 and S1116). For example, the metadata of the range of the region D2 illustrated in FIG. 7 can be acquired using the row specifier “2-1” and the column specifier “3” illustrated in FIG. The content metadata acquired here is transferred from the data operation unit 150 to the application algorithm unit 130 (S1120).

  Further, the application algorithm unit 130 requests the data operation unit 150 to acquire user profile data (S1124). For example, when the action history data includes an identifier U111 that identifies a user who refers to the content, acquisition of profile data of the user U111 in the area D1 illustrated in FIG. 7 is requested.

  In response to the request from the application algorithm unit 130, the data operation unit 150 refers to the data holding unit 160 and acquires a partial matrix in the data holding unit 160 corresponding to the user profile data (S1128, S1132). For example, the profile data of the user U111 can be acquired using the row specifier “1-1-1” and the column specifier “3” illustrated in FIG. The user profile data acquired here is transferred from the data operation unit 150 to the application algorithm unit 130 (S1136).

  Thereafter, the application algorithm unit 130 generates new profile data of the user by analyzing the action history data (S1140). For example, the application algorithm unit 130 can add the metadata of the music piece A and the purchased music piece B referred to by the user U111 to the profile data of the user U111 after being multiplied by a predetermined coefficient. Such arithmetic processing is suitable for processing using a partial matrix acquired from a data structure in a matrix format.

  Then, the application algorithm unit 130 passes the newly generated user profile data to the data operation unit 150 and instructs the data update (S1144). Then, the sub-matrix corresponding to the user profile data in the data holding unit 160 is updated by the data operation unit 150 (S1148). Further, the data holding unit 160 instructs the database input / output unit 170 to make the updated user profile data permanent at any point after S1148 (S1148).

  In FIG. 9, the action history data analysis process is described as being performed by the application algorithm unit 130, but the action history data analysis process may be standardized and implemented in the basic algorithm unit 140.

  Up to this point, the user profile data generation processing in this embodiment has been described with reference to FIG. Referring to FIG. 9, it can be understood that reading and writing with respect to the user profile data and content metadata held in the matrix format in the data holding unit 160 can be simplified to the operation of the partial matrix via the data operation unit 150.

  Next, FIG. 10 is a sequence diagram illustrating the flow of a recommendation list generation process in the present embodiment.

  Referring to FIG. 10, first, a recommendation request is transmitted from the user application unit 110 to the application algorithm unit 130 via the application interface unit 120 (S1204). The recommendation request may include, for example, a code specifying a range of content to be recommended and a recommendation method (such as a collaborative filter and a content filter) in addition to an identifier for identifying a user who desires recommendation.

  Next, the application algorithm unit 130 that has received the recommendation request requests the basic algorithm unit 140 to calculate the degree of correlation (S1208). The calculation of the degree of correlation is, for example, the processing described in relation to FIG. 5 (A) or FIG. 6 (A). The calculation processing of the correlation degree can be standardized in the basic algorithm unit 140.

  Upon receiving the correlation degree calculation request from the application algorithm part 130, the basic algorithm part 140 acquires data necessary for the correlation degree calculation from the data holding part 160 via the data operation part 150 (S1212 to S1224). . For example, when calculating the degree of correlation described with reference to FIG. 5A, data corresponding to the genre of music in the user profile data is acquired from the data holding unit 160 in the form of a partial matrix. The

  The basic algorithm unit 140 calculates the degree of correlation using the acquired data (S1228), and the calculation result is output from the basic algorithm unit 140 to the application algorithm unit 130 (S1232).

  Further, the application algorithm unit 130 requests the basic algorithm unit 140 to generate a recommendation list based on the correlation degree calculation result (S1236). The generation of the recommendation list is, for example, the processing described in relation to FIGS. 5 (B) to (C) or FIGS. 6 (B) to (C). The recommendation list generation process can also be standardized in the basic algorithm unit 140.

  The basic algorithm unit 140 that has received a recommendation list generation request from the application algorithm unit 130 acquires data necessary for generating a recommendation list from the data holding unit 160 via the data operation unit 150 (S1240 to S1252). . For example, when the recommendation list described with reference to FIGS. 5B to 5C is generated, data representing the degree of correlation between the user and individual music contents in the user profile data is stored in the data holding unit. 160 is obtained in the form of a submatrix.

  Then, the basic algorithm unit 140 generates a recommendation list using the acquired data (S1256), and the generated recommendation list is output from the basic algorithm unit 140 to the application algorithm unit 130 (S1260). Although not shown in FIG. 10, the application algorithm unit 130 may process the recommendation list acquired in accordance with the needs and restrictions of each recommendation service after acquiring the recommendation list.

  Then, the finally acquired recommendation list is output from the application algorithm unit 130 to the user application unit 110 via the application interface unit 120 (S1264).

  So far, the recommendation list generation processing in the present embodiment has been described with reference to FIG. In the present embodiment, the application algorithm unit 130 acquires a recommendation list using a typical calculation process implemented in the basic algorithm unit 140. For example, the correlation degree calculation process and the recommendation list generation process based on the correlation degree calculation result shown in FIG. 10 can be standardized and implemented in the basic algorithm unit 140. The data format handled in the basic algorithm unit 140 is shared by the submatrix format acquired from the data holding unit 160 via the data operation unit 150. Such matrix-format data is suitable for flexibly combining various calculation results output from the basic algorithm unit 140 in the application algorithm unit 130.

  The first embodiment of the present invention has been described above with reference to FIGS. According to the recommendation system 1 according to the first embodiment, the application algorithm unit 130 can flexibly combine the basic algorithm unit 140 that executes a typical calculation process.

  In addition, the user profile data and content metadata held in the matrix format in the data holding unit 160 can be read and written via the data operation unit 150 in a simplified manner in the form of submatrix operations.

  Furthermore, by providing a data access method using row and column specifiers expressed hierarchically for a data structure in a matrix format, the data access speed is increased and the range of data to be processed is flexible. Can be specified.

  That is, according to the recommendation system 1 according to the first embodiment, the design and development flexibility of the recommendation algorithm in the recommendation system 1 is improved, the cost of service development is reduced, and more effective content recommendation is achieved. The result is expected to be provided to the user.

[2] Second Embodiment In the recommendation system 1 according to the first embodiment, the system is configured by using one recommendation server and one DB server. However, when user profile data and content metadata are held in a matrix format, data sharing by a plurality of recommendation servers can be easily performed. Therefore, as a second embodiment of the present invention, an example in which a recommendation system is configured using a plurality of recommendation servers will be described.

  FIG. 11 is a schematic diagram showing the configuration of the recommendation system 2 according to the second embodiment of the present invention. Referring to FIG. 11, the recommendation system 2 includes a terminal device 10, a terminal device 12, a network 20, a Web server 22, a recommendation server 30 a and a recommendation server 30 b, and a DB server 36.

  The recommendation server 30a and the recommendation server 30b are connected to the Web server 22 and the DB server 36, respectively. Similar to the recommendation server 30 according to the first embodiment, the recommendation server 30 a and the recommendation server 30 b receive a content recommendation request from the terminal device 10 or 12, obtain a recommendation list, and receive the recommendation list via the Web server 22. Is transmitted to the requesting terminal device. Further, the recommendation server 30 a and the recommendation server 30 b obtain content metadata and user profile data required for calculation from the DB server 36.

  Here, the recommendation server 30a and the recommendation server 30b may be redundant recommendation servers that handle the same content as a recommendation target. Further, the recommendation server 30a and the recommendation server 30b may be distributed cooperative recommendation servers that handle different ranges of content as recommendation targets.

  The recommendation server 30a or the recommendation server 30b may have a function of receiving a recommendation request from the other recommendation server, generating a recommendation list, and returning the recommendation list to the requesting recommendation server. By doing so, it becomes possible to integrate the recommended lists of different ranges of content processed in a distributed manner and transmit them to the terminal device 10 or the terminal device 20.

  In the present embodiment, the DB server 36 manages users or contents to be processed by the recommendation server 30a or the recommendation server 30b using a data allocation table that defines the data allocation range for each recommendation server.

  FIG. 12 is a block diagram showing a logical functional arrangement of the DB server 36 of the recommendation system 2. Referring to FIG. 12, the DB server 36 includes a database management unit 182, a storage unit 190, and a data allocation table 192.

  The database management unit 182 has the same function as the database management unit 180 according to the first embodiment, and further uses the data allocation table 192 to specify the range of users or contents to be processed by the recommendation server 30a or the recommendation server 30b. to manage.

  The data allocation table 192 is a table for defining a data allocation range for each recommended server. In FIG. 12, for convenience of explanation, the data allocation table 192 is shown separately from the storage unit 190, but the data allocation table 192 may be included in the storage unit 190.

  FIG. 13 is an explanatory diagram for explaining an example of the data configuration of the data allocation table 192. FIG. 13A shows the data configuration of the data allocation table 192. In FIG. 13A, the data allocation table 192 has a recommendation server identifier (30a or 30b) on the vertical axis, and a data allocation range and an upper limit value of the number of data to be read on the horizontal axis. FIG. 13B conceptually shows the content of the recommendation system 2 and the metadata of the content managed in the DB server 36 in a matrix format.

  Referring to FIG. 13A, the recommendation server 30a is assigned data in the range of the region D7 among the content handled by the recommendation system 2 and the metadata of the content. Referring to FIG. 13B, the area D7 is an area for the music content and the metadata (attribute 1) of the music content. Further, the upper limit value of the number of data read to the recommendation server 30a is limited to 10,000.

  On the other hand, among the content handled by the recommendation system 2 and the metadata of the content, data in the range of the region D8 is assigned to the recommendation server 30b. Referring to FIG. 13B, the area D8 is an area for the news article content and the metadata (attributes 1 and 2) of the news article content. The upper limit value of the number of data read to the recommendation server 30a is not defined.

  The database management unit 182 shown in FIG. 12 refers to such a data allocation table 192, and realizes redundancy or distributed cooperative processing of recommended servers by processing only the data in the range allocated to each recommended server. To do. Further, by limiting the upper limit value of the number of data transferred to the recommendation server, it is possible to suppress a load on the recommendation server when the number of data increases.

  Here, when the recommended server is made redundant, even when one of the recommended servers stops functioning due to a failure or the like, the service is continued using the other recommended server. That is, the availability of the recommendation system 2 as a whole is ensured. Further, when the distributed cooperative processing is realized between the recommendation servers, for example, when the recommendation system 2 handles large-scale content data, the response performance of the recommendation system 2 as a whole is improved.

  Although FIG. 13 shows an example in which a predetermined range of content and content metadata is assigned to each recommendation server, instead, a predetermined range of users and user profile data may be assigned to each recommendation server.

  FIG. 14 is a sequence diagram illustrating an example of data reading processing from the database when the recommendation server 30a and the recommendation server 30b are activated in the present embodiment. In FIG. 14, the recommendation server 30a is activated first, and then the recommendation server 30b is activated.

  14, when the recommendation server 30a is activated, the data holding unit 160 of the recommendation server 30a requests the database management unit 182 to read data via the database input / output unit 170 (S2004, S2008).

  Upon receiving the data read request, the database management unit 182 acquires the data allocation range of the recommended server by referring to the data allocation table 192 using the identifier of the requesting recommended server (S2012 and S2016). For example, if the data allocation table 192 holds the contents shown in FIG. 13A, information indicating the area D7 is acquired from the recommendation server 30a.

  Next, the database management unit 182 acquires data included in the acquired data allocation range with reference to the storage unit 190 (S2020, S2024). For example, data corresponding to the music content and attribute 1 included in the region D7 illustrated in FIG.

  Then, the database management unit 182 outputs the data acquired from the storage unit 190 to the data holding unit 160 of the recommendation server 30a via the database input / output unit 170 (S2028, S2032).

  As a result of such a series of processing, data required for the recommendation processing in the recommendation server 30a is read into the data holding unit 160 of the recommendation server 30a.

  Further, as shown in FIG. 14, when the recommendation server 30b is started, a data read request (S2104, S2108), acquisition of a data allocation range (S2112, S2116), data reference (S2120, S2124), and data output are similarly performed. (S2128, S2132) are performed. As a result, data required for recommendation processing in the recommendation server 30b is read into the data holding unit 160 of the recommendation server 30b.

  In the present embodiment, after the data required for the recommendation process is read into each data holding unit 160 of the recommendation server 30a or the recommendation server 30b, each recommendation server shifts to a standby state for a recommendation request. Then, when a recommendation request is transmitted from the terminal device 10 or 12, processing similar to the recommendation list generation processing according to the first embodiment described with reference to FIG. 10 is performed.

  The second embodiment of the present invention has been described above with reference to FIGS. In the present embodiment, the DB server 36 manages the range of data individually allocated to the recommendation servers 30a and 30b using the data allocation table. When the database management unit 182 receives the data read request, the database management unit 182 acquires the data by referring to the data allocation table, and outputs only the data in the defined range to the requesting recommendation server. According to such a configuration, the same data range can be redundantly assigned to a plurality of recommended servers to ensure system availability, or different data ranges can be assigned to a plurality of recommended servers to realize distributed cooperative processing.

  In addition, although the structure using one DB server and a some recommendation server was demonstrated here, you may comprise a recommendation system using a some DB server and a recommendation server. For example, the DB server may be divided into a server that manages users and user profile data, and a server that manages content and content metadata.

  A series of processing according to the first and second embodiments described in this specification is executed using, for example, a general-purpose computer shown in FIG. That is, for example, the terminal devices 10 and 12, the recommendation servers 30, 30 a and 30 b, and the DB servers 32 and 36 shown in FIG. 1 or 11 can be mounted using the general-purpose computer shown in FIG.

  In FIG. 15, a CPU (Central Processing Unit) 902 controls the overall operation of the general-purpose computer. A ROM (Read Only Memory) 904 can store programs and data describing a part or all of a series of processes. A RAM (Random Access Memory) 906 temporarily stores programs and data used by the CPU 902 for arithmetic processing.

  The CPU 902, ROM 904, and RAM 906 are connected to each other via a bus 908. Further, an input / output interface 910 is connected to the bus 908.

  The input / output interface 910 is an interface for connecting the CPU 902, ROM 904, and RAM 906 to the input device 912, output device 914, storage device 916, communication unit 918, and drive 920.

  The input device 912 includes, for example, a mouse, a touch pen, a keyboard, and the like, and receives an instruction and information input from a user. The output device 914 includes, for example, a display device such as a CRT (Cathode Ray Tube) or a liquid crystal display, or an audio output device such as a speaker, and is used to output information to the user.

  The storage device 916 includes, for example, a hard disk drive or a flash memory, and stores programs and data used in the recommendation system 1 or 2. A communication unit 918 performs communication processing between devices via a network such as a LAN or the Internet.

  The drive 920 is a peripheral device provided in the general-purpose computer as necessary. For example, a recording medium 922 can be attached to the drive 920. The recording medium 922 can record a program in which a part or all of a series of processes according to the first or second embodiment is described, for example. Then, the program recorded on the recording medium 922 is read by the CPU 902 via the drive 920.

  As mentioned above, although preferred embodiment of this invention was described referring an accompanying drawing, it cannot be overemphasized that this invention is not limited to the example which concerns. It will be apparent to those skilled in the art that various changes and modifications can be made within the scope of the claims, and these are naturally within the technical scope of the present invention. Understood.

  For example, the processes according to the first and second embodiments do not necessarily have to be executed in the order described in the sequence diagram. In the first and second embodiments, each processing step may include processing executed in parallel or individually independently.

It is a schematic diagram which shows the physical structure of the recommendation system which concerns on 1st Embodiment. It is explanatory drawing which shows an example of the profile data updated according to a user's action history. It is explanatory drawing which shows an example of the profile data updated according to a user's action history and content metadata. It is a block diagram which shows the logical structure of the recommendation system which concerns on 1st Embodiment. It is explanatory drawing which shows an example of the recommendation process based on a cooperation filter. It is explanatory drawing which shows an example of the recommendation process based on a content filter. It is explanatory drawing which shows an example of the data structure of the matrix format which concerns on 1st Embodiment. It is explanatory drawing which shows an example of the data access system which concerns on 1st Embodiment. It is a sequence diagram which illustrates the flow of the profile data generation process which concerns on 1st Embodiment. It is a sequence diagram which illustrates the flow of the production | generation process of the recommendation list which concerns on 1st Embodiment. It is a schematic diagram which shows the physical structure of the recommendation system which concerns on 2nd Embodiment. It is a block diagram which shows the logical structure of the DB server which concerns on 2nd Embodiment. It is explanatory drawing which shows the structure of the data allocation table which concerns on 2nd Embodiment. It is a sequence diagram which illustrates the flow of the data reading process which concerns on 2nd Embodiment. And FIG. 11 is a block diagram illustrating a configuration example of a general-purpose computer.

Explanation of symbols

30 Information processing device (recommended server)
DESCRIPTION OF SYMBOLS 130 Application algorithm part 140 Basic algorithm part 150 Data operation part 160 Data holding part 170 Database input / output part

Claims (15)

It has profile data representing user attributes in at least one of the row direction and the column direction and metadata data items representing the characteristics of the content, and an identifier and content for identifying the user in at least one of the row direction and the column direction. A data holding unit for holding data in a matrix format data structure having data items of identifiers to be
A data operation unit that acquires an arbitrary submatrix from the data holding unit using a specifier that specifies a row or a column;
A basic algorithm unit that performs a typical calculation process on the profile data and the metadata included in the partial matrix acquired by the data operation unit;
An application algorithm unit that obtains a recommendation list in which contents recommended to the user are listed using the result of the routine calculation by the basic algorithm unit;
An information processing apparatus comprising:   2. The designator according to claim 1, wherein the specifier is a hierarchically expressed character string that designates one of the nodes in a hierarchical structure including nodes corresponding to data and nodes corresponding to a category of data. Information processing device.   The information processing apparatus according to claim 1, wherein the basic algorithm unit calculates a degree of correlation of the profile data between a specified user and another user as one of the typical calculations.   The information processing according to claim 1, wherein the basic algorithm unit calculates a correlation degree between the profile data of a specified user and the metadata of an arbitrary content as one of the routine calculations. apparatus.   The basic algorithm unit, as one of the routine calculations, the degree of correlation of the profile data between the specified user and another user, or the profile data of the specified user and the arbitrary content The information processing apparatus according to claim 1, wherein the recommendation list is generated based on a degree of correlation with metadata.   The information processing apparatus according to claim 1, wherein the basic algorithm unit calculates a correlation degree of the metadata between the designated content and another content as one of the typical calculations.   The information processing apparatus according to claim 1, wherein the basic algorithm unit calculates a degree of correlation between the metadata of the specified content and the profile data of an arbitrary user.   The information processing apparatus according to claim 1, wherein the data holding unit has a plurality of data values in one field in the matrix format data structure.   When the application algorithm unit further receives the user's behavior history data from the terminal device, the application algorithm unit includes the user who performed the behavior according to the metadata of the content that is the target of the behavior included in the behavior history data. The information processing apparatus according to claim 1, wherein the profile data is updated.   The information processing apparatus according to claim 1, further comprising: a database input / output unit that requests the database to persist the data held in the data holding unit based on an instruction from the data holding unit. apparatus.   The data holding unit acquires data in a range allocated to the information processing apparatus from a data item of the profile data and a data item of the metadata from the database via the database input / output unit. The information processing apparatus described in 1. It has profile data representing user attributes in at least one of the row direction and the column direction and metadata data items representing the characteristics of the content, and an identifier and content for identifying the user in at least one of the row direction and the column direction. Holding the profile data and the metadata in a matrix-type data structure having identifier data items to be identified;
Obtaining an arbitrary submatrix from the matrix-format data structure using a specifier specifying a row or column;
Performing a routine calculation process on the profile data and the metadata included in the submatrix acquired from the matrix data structure;
Obtaining a recommendation list in which contents recommended to the user are listed using the result of the routine calculation;
Including recommendation method. A computer that controls the information processing device:
It has profile data representing user attributes in at least one of the row direction and the column direction and metadata data items representing the characteristics of the content, and an identifier and content for identifying the user in at least one of the row direction and the column direction. A data holding unit for holding data in a matrix format data structure having data items of identifiers to be
A data operation unit that acquires an arbitrary submatrix from the data holding unit using a specifier that specifies a row or a column;
A basic algorithm unit that performs a typical calculation process on the profile data and the metadata included in the partial matrix acquired by the data operation unit;
An application algorithm unit that obtains a recommendation list in which contents recommended to the user are listed using the result of the routine calculation by the basic algorithm unit;
Program to function as A computer that controls the information processing device:
It has profile data representing user attributes in at least one of the row direction and the column direction and metadata data items representing the characteristics of the content, and an identifier and content for identifying the user in at least one of the row direction and the column direction. A data holding unit for holding data in a matrix format data structure having data items of identifiers to be
A data operation unit that acquires an arbitrary submatrix from the data holding unit using a specifier that specifies a row or a column;
A basic algorithm unit that performs a typical calculation process on the profile data and the metadata included in the partial matrix acquired by the data operation unit;
An application algorithm unit that obtains a recommendation list in which contents recommended to the user are listed using the result of the routine calculation by the basic algorithm unit;
A computer-readable recording medium on which a program is recorded so as to function as a computer. A user application unit that requests the information processing device to recommend content;
A terminal device comprising:
It has profile data representing user attributes in at least one of the row direction and the column direction and metadata data items representing the characteristics of the content, and an identifier and content for identifying the user in at least one of the row direction and the column direction. A data holding unit for holding data in a matrix format data structure having data items of identifiers
A data operation unit for obtaining an arbitrary submatrix from the data holding unit using a specifier for designating a row or a column;
A basic algorithm unit that performs a typical calculation process on the profile data and the metadata included in the partial matrix acquired by the data operation unit;
And an application algorithm unit that acquires a recommendation list in which contents recommended to a user are listed using the result of the routine calculation performed by the basic algorithm unit, and transmits the acquired recommendation list to the terminal device;
An information processing apparatus comprising:
Information processing system including

Images