JP2005128673A - Information recommendation method, server device, program, and recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain content related to a user having a common interest.
SOLUTION: Based on content reproduction information, a user type related to content of a user who requests information search is determined, and the degree of popularity of content included in the content reproduction information of the user who requests information search is quantified, and the reproduction frequency is determined. , Generating a user preference vector to which a weight calculated based on the degree of publicity of content information is generated according to the user type of the information search requesting user, and the generated user preference vector is similar to the user requesting the information search A user other than the user who requests the search is selected, content information included in the content reproduction information of the selected user is acquired, and information useful for the user who requests the information search is selected from the acquired content information. The selected information is presented to the user who requests the information search.
[Selection] Figure 1

Description

  The present invention collects information on user problems such as interests, interests and intentions and information acquired by users in information search via a network, and provides information to other users who have similar problem awareness. The present invention relates to information recommendation technology that supports collection activities.

  Conventionally, collaborative filtering is a method for supporting information gathering activities by collecting user's awareness of interests, interests and intentions and information acquired by users and providing them to other users with similar problem awareness. There is an information recommendation method using. This collaborative filtering is based on an environment where there are a plurality of people, and User profile 1e is a file that describes the criteria for selecting information related to a user, and Recommendation is a file that stores information obtained by the user by collecting information. User modeler, which is a processing unit for extracting user characteristics, Match maker, which is a processing unit for calculating similarity of user relations, and Score predicator, which is a processing unit for calculating ranking of information according to individual users It consists of five elements. When the user inputs User profile and Recommendation, User moerer quantifies the user's characteristics based on the User profile, Match maker searches for others with similar user characteristics, and Score predictor The information is filtered by outputting information evaluated according to each user from the recommendation of the person.

  Various methods have been proposed for the process of calculating the similarity between users performed in the process of Match maker which is one of the elements of the collaborative filtering system.

  A method for calculating similarity between users in GroupLens (for example, see Non-Patent Document 1), which is the most general algorithm among currently proposed methods, will be described below.

GroupLens treats the user's interest as an array of numerical values obtained by scoring the user evaluation regarding the content, and defines the similarity r _KL between the evaluation score array K of the user K and the evaluation score array L of the user L by the following expression. .

  Several studies have been made to solve the problem in collaborative filtering using the GroupLens algorithm, many of which are based on the GroupLens algorithm.

  On the other hand, in the information distribution process in human society, it has been reported that users involved in the process are classified into a plurality of user types according to their behavior characteristics (see, for example, Non-Patent Document 2). As for the determination method of the user type, various researches such as a determination method based on a result of a questionnaire and a determination method based on a Web access log are being conducted. Non-Patent Document 3 reports that when users are classified based on the characteristics of the playback history of favorite content such as music and video, the information acquisition form from others tends to differ for each user type. According to Non-Patent Document 3, there is the following relationship between the feature of content reproduction and the feature of information acquisition form from others.

  (1) Users who are strongly concerned with content and have high reproduction frequency only for specific content (hereinafter referred to as “Mania”) have high expectations for information acquisition from others with their favorite minor hobbies.

  (2) A user who reproduces some contents continuously for a certain period of time (hereinafter referred to as Opinion Leader) is highly expected to acquire information from others who have new information.

(3) A user who reproduces a lot of content a small number of times (hereinafter referred to as “majority”) is highly expected to acquire information from others who have trendy information with many employers in the world.
Paul Resnick, Neophytos Iacovou, Mitsuku Schak, Peter Verstrom, John Riedl, “GroupLens: An Open Architecture for collaborative feeds”. E. M. Rogers, “Science of Communication-Basic Theory of Multimedia Society”, 1992 Endo, Iwata, Otake, Yamada, “Evaluation of Community Formation Method Using Content Attribute Information”, 47th GN Study Group of Information Processing Society, 2002

  In the conventional method, the similarity calculation formula between users includes the variance of user evaluation values related to content in the denominator. Therefore, users with a small variance of evaluation, i.e., there is no concern for specific content, and there is a difference in evaluation for each content. There is a tendency that the degree of similarity with users who do not have is increased. For example, when the three users shown in FIG. 15 exist, if the similarity between User A who has a tendency to concentrate on specific content and the other two users is calculated by the conventional method, User A highly evaluates. The degree of similarity with User B, which has a small evaluation variation with respect to the content, is calculated higher than the degree of similarity with User C that has the same interest in the specific content (content 3). Therefore, if the content (content 3) with the same interest between UserA and UserC is the minor content preferred by UserA, it is difficult to obtain information from others who have minor interests, which is the information acquisition form of Mania. is there.

  In addition, regarding the similarity between User B, which has a tendency of a minority in evaluation of content, and the other two users, the similarity with User C is calculated to be lower than the similarity with User A. However, the content (content 8) in which the evaluations of UserC and UserB are gently matched is the major content, and the content (content 7) in which the evaluations of UserA and UserB are gently matched is the minor content. In this case, it is difficult to acquire information from others who have information on fashion, which is the information acquisition form of the Majority.

  The present invention has been made in view of such circumstances, and an object thereof is to provide an information recommendation method, a server device, a program, and a recording medium on which the program is recorded, which can solve the above-described problems.

  Therefore, in order to solve the above-described problem, the information recommendation method according to claim 1 is configured such that when a user makes an information search request by a client device, a server device connected to the client device performs information search processing, and information is transmitted to the client device. In the information recommendation method to be provided, the concentration of interest is obtained from the content reproduced by the user or the attribute of the content, the variation of the attribute of the content, and the burst property of the content reproduction, and a threshold is set to the value of the concentration A user having a preference vector that determines a user type, generates a preference vector that expresses the user's preference by a method according to the type of the user, and has a similarity greater than a certain value with the user's preference vector that requests information provision Generate a set and request information from the information corresponding to the users in this set And determining the information to be provided to over The.

  The information recommendation method according to claim 2 is an information recommendation method in which when a user makes an information search request by a client device, a server device connected to the client device performs information search processing and provides information to the client device. In the client device, when the user reproduces the content, the process of receiving the content reproduction information from the client device, and the content attribute information corresponding to the content ID included in the received content reproduction information A content information database, a process of associating the content reproduction information with the acquired content attribute information and recording it in the content reproduction information database, and an information search request designating user information from the client device. Reception process and information search required The process of determining the user type of the original user based on the content reproduction information recorded in the content reproduction information database, and the degree of popularity of the content or content attribute included in the content reproduction information of the information search requesting user Based on the calculation process and the content reproduction information of the user recorded in the content reproduction information database and the well-known degree of the content or content attribute, a vector expressing the user preference according to the type of the user is generated. A process, a process of recording the vector as a user preference vector in a preference vector information database, and a similarity between the preference vector of the information search requesting user and the user preference vector recorded in the preference vector information database The process of calculating the degree and the information Selecting a user having a high similarity between the search requesting user and the preference vector, and extracting information useful for the user who has transmitted the information search request from the content or content attribute information included in the content reproduction information of the selected user; And transmitting the extracted information as an information search result to the client device.

  The information recommendation method according to claim 3 is the information recommendation method according to claim 2, wherein the user type is a content reproduced by a corresponding user based on content reproduction information recorded in the content reproduction information database, or Based on the index value calculated to reflect the dispersion of the content attributes and the burstiness of content playback that expresses whether content playback is concentrated within a specific time range or is continuously played back It is determined.

  Further, the information recommendation method according to claim 4 is the information recommendation method according to claim 2 or 3, wherein the preference vector is an element having a content ID or content attribute of a content reproduced by the user in the past as a vector element. Based on the reproduction frequency of content having a content ID or content attribute and the familiarity of the content or content attribute, a value given a weight calculated according to the user type of the information search requesting user is set as an element value It is characterized by doing.

  In addition, when the user makes an information search request by the client device, the server device according to claim 5 receives the request, performs information search processing, and provides information to the client device. The concentration of interest is determined from the content attributes, content attribute variations, and bursts of content playback, and a threshold is set for the concentration value to determine the user type. According to the user type Generate a preference vector that expresses the user's preference by the method, generate a user set having a preference vector that has a certain degree of similarity with the user's preference vector for which information is to be provided, and handle users included in this user set The information to be provided to the user who requests information provision is determined from the information to be provided.

  According to a sixth aspect of the present invention, when a user makes an information search request by a client device, the server device receives the request, performs information search processing, and provides information to the client device. When the user reproduces content, means for receiving content reproduction information from the client device and content attribute information corresponding to the content ID included in the received content reproduction information are acquired from the content information database. Content information acquisition means for performing, content reproduction information, reproduction information registration means for recording the acquired content attribute information in association with each other, and recording information search request specifying user information from the client device The user of the information search requesting user The user type determination means for determining the type based on the content reproduction information recorded in the content reproduction information database, and the content or the content attribute in the content reproduction information of the information search requesting user as numerical values A vector representing the user's preference according to the type of the user, based on the degree-of-name calculation means for calculating, the content reproduction information of the user recorded in the content reproduction information database and the name of the content or the content attribute. Generating a preference vector generating means for recording this vector as a user preference vector in a preference vector information database, a preference vector of the information search requesting user, and a user preference vector recorded in the preference vector information database Calculate the similarity of A user who has selected a preference similarity calculating unit and a user having a high preference vector similarity with the information search requesting user, and has transmitted an information search request from content or content attribute information included in the content reproduction information of the selected user And a search result extraction unit that extracts information useful for the client and transmits the extracted information as an information search result to the client device.

  The server device according to claim 7 is the server device according to claim 6, wherein the user type is content reproduced by the corresponding user based on content reproduction information recorded in the content reproduction information database or the content thereof Judged by an index value that is calculated by reflecting the variation in attributes and the burstiness of content playback that expresses whether content playback is concentrated within a specific time range or is continuously played back. It is characterized by that.

  The server device according to claim 8 is the server device according to claim 6 or 7, wherein the preference vector includes a content ID or content attribute of a content reproduced by the user in the past as a vector element, and a content ID serving as an element Alternatively, an element value may be a value assigned with a weight calculated according to a user type of the information search requesting user based on the reproduction frequency of the content having the content attribute and the degree of publicity of the content or the content attribute. It is characterized by.

  A program according to claim 9 is characterized in that the information recommendation method or server device according to any one of claims 1 to 8 is described as a computer program and can be executed.

  A recording medium according to a tenth aspect is characterized in that the information recommendation method according to any one of the first to eighth aspects or the server device is recorded with a program that can be executed by a computer.

  In the present invention, the user is classified into several types based on the evaluation of the user's content, and the user's preference vector is calculated by a method according to this type. If the information to be provided is provided, there is a high possibility that it is useful information that matches the interest of the user who requests the information provision.

  According to the present invention, for a user, content that suits hobbies and tastes is provided as compared with the conventional recommendation method by providing recommendation information that takes into account the attributes of the information that the user is interested in and preference bias. Can be found and acquired.

  For content providers, in particular, it is possible to expect the effect of promoting the sale of content by efficiently distributing minor content, etc., which has not been easily noticed by users, according to user preferences.

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

  In the description of the embodiment of the present invention, content is a concept including all types of content such as video information, audio information, image information, and character information.

  FIG. 1 illustrates a system configuration example of an information recommendation system realized by the present invention. The information recommendation system is roughly composed of a plurality of client devices 10 and a server device 20. The client device 10 and the server device 20 are connected to the Internet, a LAN, a personal computer communication network via a public line, or any other wired or wireless. It is possible to connect via a communication network (not shown) such as a used network. Although there are a plurality of client devices 10, for simplicity of explanation, only one client device 10 will be described below.

(Configuration of client device 10)
As shown in FIG. 1, the client device 10 includes a user information input unit 11, a content reproduction control unit 12, a content ID acquisition unit 13, a reproduction information notification unit 14, an information search request transmission unit 15, and a search result. Display means 16.

  The user information input unit 11 provides a user information input unit such as a user ID necessary for using the service provided by the server device 20.

  The content reproduction control means 12 performs control such as reproduction and stop of the content file 17 stored in the local disk of the client device 10, a storage medium such as a CD, a content distribution server, or the like.

  The content ID acquisition unit 13 acquires a content ID, which is a unique identifier of content embedded in the header portion of the content file 17, from the content file 17.

  The reproduction information notification unit 14 uses the user information input using the user information input unit 11, the content ID acquired using the content ID acquisition unit 13, the current time as the content reproduction time, and the content reproduction information. To the server device 20.

  The information search request transmitting unit 15 transmits user information input using the user information input unit 11 to the server device 20 as an information search request message.

  The search result display means 16 receives and displays the search result returned from the server apparatus 20 as a response to the information search request transmitted to the server apparatus 20 by the information search request transmission means 15.

(Configuration of server device 20)
As shown in FIG. 1, the server device 20 includes a content information acquisition unit 21, a reproduction information registration unit 22, a user type determination unit 23, a name degree calculation unit 24, a preference vector generation unit 25, and a preference similarity calculation. Means 26, search result extraction means 27, content information DB 28, content reproduction information DB 29, and preference vector information DB 2a are provided.

  An example of the data structure of the content information DB 28 is shown in FIG. The content information DB 28 stores content attribute information such as content ID, content title, artist name, and genre name.

  Next, an example of the data structure of the content reproduction information DB 29 is shown in FIG. The content reproduction information DB 29 stores a user ID, content ID, reproduction time, content attribute information, and the like.

  Next, an example of the data structure of the preference vector information DB 2a is shown in FIG. The preference vector information DB 2a stores a user ID, an element number, an element value, an element name, and the like.

  Subsequently, each component means will be described. The content information acquisition unit 21 acquires content attribute information (content title, artist name, genre name, etc.) of the content managed by the specified content ID from the content information DB 28.

  The reproduction information registration unit 22 receives content reproduction information (user ID, content ID, content reproduction time, etc.) transmitted from the client device 10, and receives the received user ID, content ID, content reproduction time, and content information. The attribute information of the content managed by the content ID included in the content reproduction information acquired using the acquisition unit 21 is stored in the content reproduction information DB 29.

  The user type determination unit 23 receives the user ID as an information search request from the client device 10 and stores the content reproduction information (user ID, content ID, reproduction time, content attribute information, etc.) of the user corresponding to the received user ID as the content. The user type of the user is determined based on the content playback information acquired from the playback information DB 29.

  The famousness calculation means 24 acquires distribution information of the content to be measured, and calculates the content managed by the specified content ID and the famousness of the specified content attribute based on the acquired distribution information of the content To do.

  The preference vector generation unit 25 acquires the degree of popularity of content or content attribute included in the content reproduction information of the user corresponding to the specified user ID from the name calculation unit 24 and the user type determination unit 23. The content reproduction information obtained from the content reproduction information DB 29, and the content reproduction information obtained from the content reproduction information DB 29, for all users who enjoy this service with a calculation method according to the user type of the designated user. The preference vector is generated based on the name recognition degree, and the generated preference vector is stored in the preference vector information DB 2a.

  The preference similarity calculating means 26 acquires the preference vectors of all users who enjoy the service from the preference vector information DB 2a, and the user preference vectors corresponding to the specified user ID and the other user preference vectors. The degree of similarity is calculated, and a user with a high degree of similarity calculated is selected.

  The search result extraction unit 27 acquires information regarding all contents included in the content reproduction information of the user corresponding to the specified plurality of user IDs from the content reproduction information DB 29, and specifies all the acquired information with respect to the specified information. The score obtained by quantifying the benefit for the user corresponding to the second user ID is calculated, and the calculated high score information is transmitted to the client device 10.

  Next, the overall operation of the information recommendation system according to the embodiment of the present invention will be described. 5 and 6 are overall flowcharts of the information recommendation system according to the embodiment of the present invention. A content ID for uniquely identifying the content is given to the content targeted by the present invention, the content ID given from the content file can be acquired, and content attribute information registered in advance by a content provider or the like Must be obtained based on the content ID. Therefore, before the processing of this system, it is necessary to perform the following preprocessing.

(Preprocessing)
(1) A content provider or the like registers content attribute information in the content information DB 28.

  (2) For the recorded content attribute information, the server device 20 issues a content ID that is a unique identifier of the content.

  (3) The issued content ID is embedded in the content file using the header portion of the content file.

  The above processing is not limited to the processing in the server device 20. For example, when the target content is music, it is a music database provided by Gracenote (http://www.gradenote.com). It is also possible to use a processing result performed by a device other than the server device 20, such as using a song ID and content attribute information managed by "CDDB".

  After the above preprocessing, processing is performed according to the following flow.

(Content reproduction information registration processing flow of the client device 10)
(Step S01) User Information Input Processing A user who uses this service transmits content reproduction information to the server device 20. At that time, user information such as a user ID for uniquely identifying the user is input using the user information input means 11. The trigger for inputting user information may be the content playback start time, playback player activation time, content playback information transmission time, etc., but needs to be performed before the content playback information transmission processing in step S05 described later. .

(Step S02) Content Reproduction Processing The content reproduction means 12 reproduces the content file 17 designated by the user. The content reproduction format exists on a content distribution server operated by a content distributor and a format for reproducing a content file existing on a storage device of the client device 10 or a content file stored on a CD / DVD or the like. There are formats for reproducing content files for stream distribution, etc., but any content reproduction format is applicable in this embodiment.

(Step S03) Content ID Acquisition Processing The content ID acquisition means 13 acquires the content ID embedded in the content file in the preprocessing (3) from the content file 17 reproduced using the content reproduction control means 12. Here, the method of acquiring the content ID from the content file 17 includes a method of reading the content ID embedded in the header portion of the content file described above, a method of reading the content ID embedded in the content file by digital watermarking, In the case of a CD or the like, a method of reading the ID using a dedicated tool for reading the ID embedded in the CD can be considered.

(Step S04) Content ID Embedding Presence Determination The presence or absence of content ID embedding in the content file 17 is determined.

(Step S05) Reproduction Information Transmission Processing When the content ID is embedded in the content file 17, the reproduction information notification means 14 sends the user ID input in step S01 and the content of the content file 17 acquired in step S03. The ID and the current time as the content playback time are transmitted to the server device 20 as content playback information.

(Content reproduction information registration processing flow of server device 20)
(Step S06) Content Information Acquisition Processing The server device 20 receives the content reproduction information (user ID, content ID, reproduction time, etc.) transmitted from the client device 10 in step S05. The content information acquisition unit 21 acquires content attribute information (content title, artist name, genre name, etc.) of the content corresponding to the content ID included in the received content reproduction information from the content information DB 28.

(Step S07) Reproduction Information Registration Process The reproduction information registration means 22 includes the content reproduction information (user ID, content ID, reproduction time, etc.) transmitted from the client device 10 in step S05, and the content attribute acquired in step S06. The information is stored in the content reproduction information DB 29 in association with each other.

(Information search processing flow)
(Step S08) User Information Input Processing A user who uses this service uses the user information input means 11 to input user information such as a user ID for uniquely identifying the user. The information search request transmission unit 15 of the client device 10 transmits the input user information to the server device 20 as an information search request message.

(Step S09) User Type Determination Processing The server device 20 receives user information as an information search request message transmitted from the client device 10. The user type determination unit 23 acquires the content reproduction information of the information search requesting user corresponding to the received user ID from the content reproduction information DB 29, and the content reproduced by the user based on the acquired content reproduction information or its content An index value is calculated that reflects the variation in content attributes and the burstiness of content reproduction that expresses numerically whether content reproduction is concentrated within a specific time range or is being continuously reproduced. Then, based on the calculated index value, the user type of the information search requesting user is determined.

  In the following description, the user type will be described by taking the three-level user type reported in Non-Patent Document 3 as Mania, Opinion Leader, and Majority as an example, but as the user type, Japanese Patent Application No. 2003-159118 “Profiling System” It is also possible to use a user type such as a user profile indicated by “Profiling method, program and recording medium”.

(Step S10) Name recognition calculation processing The name recognition means 24 distributes content or content attribute information for all contents corresponding to the content ID included in the content reproduction information of the information search requesting user acquired in step S09. Information is acquired, and based on the acquired distribution information, the degree of popularity of the content or content attribute is calculated.

  Here, the distribution information of content or content attribute information is sales / ranking information (www.oricon.co.jp) related to music CD sales such as provided by Oricon, for example, WinMX (http: //www.winmx). .Com /) and the number of search results when searching the content attribute information of the target content using a search engine, etc., using an application for exchanging content between users. Or a value obtained by quantifying the distribution frequency of content, such as information obtained by combining a plurality of such information. Of course, it is also possible to calculate distribution information of content or content attribute information based on the content reproduction information stored in the content reproduction information DB 29 acquired by the present system.

(Step S11) Preference Vector Generation Processing The preference vector generation means 25 acquires the content reproduction information of all users who enjoy the service from the content reproduction information DB 29, the acquired content reproduction information, and the information search request determined in step S09 Based on the user type of the original user and the familiarity of the content or content attribute acquired in S10, a preference vector, which is a vector expressing the user's preference, is generated for all users who enjoy this service. The generated preference vector is stored in the preference vector information DB 2a.

  The preference vector includes the content ID or content attribute information of the content that the user has played in the past as an element, the reproduction frequency such as the number of times or the reproduction time of the content having the content ID or content attribute information as the element, and the content or content attribute. The product of the weight calculated according to the user type of the information search requesting user based on the degree of name recognition is the value of each element.

(Step S12) Preference Similarity Calculation Processing The preference similarity calculation means 26 acquires the preference vectors of all users generated in step S11 from the preference vector information DB 2a, and requests information search based on the acquired preference vector information. The similarity of preference vectors between the original user and other users is calculated.

  Conventional methods for calculating the preference vector similarity between two users include a method of taking the cosine of two target vectors, and a method of taking the cosine after normalizing the vectors so that the average of the element values of the vectors becomes zero. It is conceivable to use a preference similarity calculation method for collaborative filtering.

(Step S13) Information Selection Processing The search result extraction unit 27 generates a user set having a high similarity to the preference vector of the information search requesting user calculated in step S12, and the user content reproduction information included in the generated user set Is acquired from the content reproduction information DB 29. In consideration of the content reproduction information of the user included in the user set, the presence / absence of reproduction of the information search requesting user, and the like for the content included in the acquired content reproduction information, the content is determined by the information search requesting user. , Calculate a score that represents how useful it is. The calculated high score content is transmitted to the client device 10 as an information search result.

(Step S14) Search Result Display Processing The search result display means 16 of the client device 10 receives and displays the information search result transmitted from the server device 20 in step S13.

  Next, the user type determination process in step S09 in the overall process flow described above will be described in more detail with reference to FIG.

  The user type determination unit 23 receives the user ID as an information search request message from the client device 10 (S0901).

  Based on the received user ID, the content reproduction information of the information search requesting user is acquired from the content reproduction information DB 29 (S0902).

  The acquired content reproduction information of the information search request source user is totaled by content ID or content attribute information and time. As a result, content reproduction indicating what content (how many times, time, etc.) has been reproduced by the information search requesting user from time 1 to time 2 after the set time range. An information summary table is created.

  FIG. 8 shows an example in which a content reproduction information tabulation table is created from content reproduction information. In the example shown in FIG. 8, the number of times of content reproduction obtained from the content reproduction information is totaled based on the content artist name and time (S0903).

  A content reproduction frequency table is created by normalizing the values in the content reproduction information aggregation table of the created information search request source user. The content reproduction frequency table expresses how much the information search requesting user is interested in a certain content or content attribute among all the interests of the user. FIG. 9 shows an example in which a content reproduction frequency table is created from the content reproduction information tabulation table created in FIG. In this example, by dividing each value of the content reproduction information aggregation table by the total number of times of content reproduction, the content reproduction information aggregation table is normalized to create a content reproduction frequency table (S0904).

  Based on the value of the created content reproduction frequency table, a user index value for determining the user type of the information search requesting user is calculated. In the present invention, the variation of content or content attributes to be played back by the user and the burstability of content playback that expresses numerically whether content playback is concentrated within a specific time range or continuously played back. Based on this, a user index value is calculated.

  FIG. 10 shows an example in which the user index value is calculated from the content reproduction frequency table. In the example shown in FIG. 10, the variation of the reproduction content attribute is shown, and the value of the content reproduction frequency table is shown for each content attribute (shown in FIG. 10). In the example, the distribution of values aggregated for each artist) is used. Further, the burst property of content reproduction is calculated as the average of the calculated variance values by calculating the variance value in the time axis direction of the content reproduction frequency table for each content attribute (in the example shown in FIG. 10, artist). The user index value is a quotient of the variation in the reproduction content attribute and the burst property of the content reproduction. In the user index value calculation method shown in FIG. 10, the user index value of a user who plays a large number of times only for specific content and repeats playback over a long period of time is calculated high, and many types of content are Therefore, the user index value of the user who reproduces is calculated to be low. In other words, a user who has the highest user index value in Maria, the lowest user index value in Majority, and takes an intermediate user index value between the user index value in Mania and the user index value in Majority becomes Opinion Leader (S0905). ).

  The user type is determined by comparing the calculated user index value with thresholds for determining Mania, Opinion Leader, and Majority. The threshold for separating Mania and Opinion Leader is “threshold 1”, and the threshold for separating Opinion Leader and Majority is “threshold 2”. First, the user index value is compared with the threshold value 1 (S0906).

  (1) The user index value is smaller than the threshold value 1.

The user index value and the threshold value 2 are compared. (S0907)
(1-1) When the user index value is smaller than the threshold value 2 The user type of the information search requesting user is determined as “Majority” (S0908).

  (1-2) When the user index value is larger than the threshold value 2 The user type of the information search requesting user is determined to be “Opinion Leader” (S0909).

(2) When the user index value is larger than the threshold value 1 The user type of the information search requesting user is determined to be “Mania” (S0910).

  Regarding the threshold setting for determining the user type, a method using a fixed value determined empirically at the start of the service or a user evaluation for the search result provided by the system is fed back and machine learning or the like is used. For example, a method of dynamically setting can be considered.

  It should be noted that the user index value calculation methods shown in FIGS. 7 to 10 are described for easy understanding of an example of the present invention. In the present invention, the user index value is a reproduction content attribute. As long as it is a numerical value that represents a feature relating to the user's interest, calculated from the variation of the content and the burst property of content reproduction, the method is not limited to the above-described method of calculating the user index value.

  Next, according to FIG. 11, for the content name recognition calculation process in step S10 in the overall processing flow described above, a case where the content reproduction information stored in the content reproduction information DB 29 acquired by this system is content distribution information is taken as an example. This will be described in more detail.

  As content distribution information, content reproduction information from the specified arbitrary time to the current time acquired from the content reproduction information DB 29, content ID or content attribute (artist name in the example shown in FIG. 11), and user ID And sum up. Thus, what content (which artist's content in the example shown in FIG. 11) has which content or which content attribute from which user is designated to the present time (how many times, time, etc.) A table (content reproduction totaling table in FIG. 11) indicating whether the content has been reproduced is created. In the example of FIG. 11, the content familiarity is calculated as the number of people who have played back the content having the artist whose name is to be calculated.

  The content recognition calculation method shown in FIG. 11 is described in order to facilitate understanding of an example of the present invention, and the present invention is not limited to the content recognition calculation method described above. Absent.

  Next, the preference vector generation process in step S11 in the overall process flow described above will be described in more detail with reference to FIG. The preference vector generation unit 25 acquires the user type of the information search requesting user from the user type determination unit 23 (S1101).

  The content reproduction information of the information search request source user is acquired from the content reproduction information DB 29 (S1102), and a set C of content or content attributes including all content or content attributes included in the acquired content reproduction information is generated (S1103). ).

  The famousness of all the content or content attributes included in the content or content attribute set C is acquired from the famousness calculation means 24 (S1104).

  Next, the loop function j is initialized to 1 (S1105), and the following processing is repeated.

  The preference vector generation means 25 acquires content reproduction information for the user who enjoys this service from the content reproduction information DB 29 (S1106).

  Based on the acquired content playback information of the user j, the cumulative playback count Pjk of the content having the content k or content attribute k of the user j is set to all the content k or content attributes k included in the content or content attribute set C. Is calculated (S1107).

  Based on the calculated content cumulative reproduction count, preference vectors of all users are generated by the following method according to the user type of the information search request source (S1108).

(1) When the user type of the information search requesting user is “Mania” The information search requesting user has a high expectation for information from others who have minor hobbies with him. It is necessary to emphasize element values of highly rare content or content attributes. Therefore, in this embodiment, each element value of the user's preference vector is calculated as a quotient between the content cumulative number of reproduction times and the content or content attribute name recognition (S1109).

(2) When the user type of the information search requesting user is “Opinion Leader” In this embodiment, each element value of the user's preference vector is set as the cumulative content reproduction count (S1110).

  Here, since the information search requesting user has high expectation for new information, the information limited to a certain period close to the current time is distributed as the distribution information of the content or content attribute information that is the origin of the content or the content attribute. There is a possibility that a more effective preference vector may be generated by a method such as calculating each element value of the user's preference vector as a product of the cumulative number of times of content reproduction and the famousness of the content or content attribute.

(3) When the user type of the information search requesting user is “Majority” Since the information search requesting user has high expectation for information acquisition from users who have popular information in the world, the preference vector Among the elements, it is necessary to emphasize element values of content or content attributes that are highly known and widely spread in the world. Therefore, in this embodiment, each element value of the user's preference vector is calculated as the product of the content cumulative number of reproduction times and the content or the name of the content attribute (S1111).

  The above operation is repeated for all users (the total number N of users) who enjoy this service. That is, it is determined whether j ≧ N (S1112). If j <N, j is incremented by +1 (S1113), and step S1106 and subsequent steps are repeated. When j ≧ N, the process is terminated.

  The preference vector calculation formula shown in FIG. 12 is described for easy understanding of an example of the present invention, and the present invention is not limited to the preference vector calculation formula described above. Absent.

  FIG. 13 shows the implementation of the example of the evaluation table shown in FIG. 15 in the case where appropriate content name recognition is given to the content to be measured and User A whose user type is “Mania” is the information search requesting user. It is a calculation result of each user's preference similarity using an example.

  FIG. 14 shows the implementation of the example of the evaluation table shown in FIG. 15 in the case where appropriate content name recognition is given to the content to be measured and User B whose user type is Majority is the information search requesting user. It is a calculation result of each user's preference similarity using an example.

  The calculation results of FIGS. 13 and 14 are the results of solving the above-described problems in the prior art.

  Note that the present invention can be realized by using a computer by configuring a part or all of the processing functions of the method shown in FIGS. 5 to 12 or the apparatus shown in FIG. 1 as a program, or FIGS. Can be configured as a program and executed by a computer. Also, a computer-readable recording medium such as a flexible disk, MO, ROM, memory card, or the like is stored in the computer. It can be recorded on a CD, a DVD, a removable disk, etc., stored, provided, and distributed via a communication network such as the Internet.

The figure which showed the outline | summary of the system configuration | structure of an information recommendation system. The figure which showed the data structure example of content information DB. The figure which showed the data structure example of content reproduction information DB. The figure which showed the data structure example of preference vector information DB. The whole flowchart of an information recommendation method. The whole flowchart of an information recommendation method. The detailed flowchart of a user classification calculation process. The figure which showed the example of conversion from content reproduction information to a content reproduction total table. The figure which showed the example of conversion from a content reproduction total table to a content reproduction frequency table. The figure which showed the example of calculation of a user index value. The figure which shows the example of the calculation method of content visibility. The detailed flowchart of a preference vector production | generation process. The figure which showed the example of a calculation result of the preference similarity by a present Example. The figure which showed the example of a calculation result of the preference similarity by a present Example. The figure which showed the calculation result example of the preference similarity by the conventional collaborative filtering.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ... Client apparatus 11 ... User information input means 12 ... Content reproduction control means 13 ... Content ID acquisition means 14 ... Reproduction information notification means 15 ... Information search request transmission means 16 ... Search result display means 17 ... Content file 20 ... Server apparatus 21 ... content information acquisition means 22 ... reproduction information registration means 23 ... user type determination means 24 ... name degree calculation means 25 ... preference vector generation means 26 ... preference similarity calculation means 27 ... search result extraction means 28 ... content information 29 ... content reproduction information 2a ... Preference vector information

Claims (10)

In an information recommendation method in which when a user makes an information search request by a client device, a server device connected to the client device performs an information search process and provides information to the client device.
Determine the concentration of interest from the content played by the user or the content attribute, the content attribute variation, and the burstiness of the content playback, determine the type of user by setting a threshold value for this concentration value,
Generate a preference vector that expresses the user's preference by a method according to the type of the user,
A user set having a preference vector having a similarity equal to or greater than a certain value with a user's preference vector for information provision,
An information recommendation method comprising: determining information to be provided to a user who requests information provision from among information corresponding to users included in the user set. When a user makes an information search request by a client device, a server device connected to the client device performs an information search process and provides information to the client device.
A process of receiving content playback information from the client device when a user plays back content on a client device;
Acquiring content attribute information corresponding to a content ID included in the received content reproduction information from a content information database;
Recording the content reproduction information and the acquired content attribute information in a content reproduction information database in association with each other;
Receiving an information search request specifying user information from the client device;
Determining the user type of the information search request source user based on the content reproduction information recorded in the content reproduction information database;
A step of calculating the degree of popularity of the content or content attribute included in the content reproduction information of the information search requesting user as a numerical value;
Generating a vector expressing user preferences according to the type of the user based on the content reproduction information of the user recorded in the content reproduction information database and the degree of popularity of the content or content attribute;
Recording the vector as a user preference vector in a preference vector information database;
Calculating the similarity between the preference vector of the information search requesting user and the preference vector of the user recorded in the preference vector information database;
A user having a high similarity between the information search request source user and the preference vector is selected, and information useful for the user who transmitted the information search request is extracted from the content or content attribute information included in the content reproduction information of the selected user. Process,
Transmitting the extracted information as an information search result to the client device;
An information recommendation method characterized by comprising: The information recommendation method according to claim 2,
The user type is based on the content playback information recorded in the content playback information database, whether the content played back by the user or the content attribute of the user and whether the content playback is concentrated within a specific time range An information recommendation method, characterized in that the determination is based on an index value calculated by reflecting the burst property of content reproduction, which is a numerical value indicating whether the content is continuously reproduced. The information recommendation method according to claim 2 or 3,
The preference vector has a content ID or content attribute of content that the user has played in the past as an element of the vector, and has a reproduction frequency of content having the content ID or content attribute as an element, and a well-known degree of the content or content attribute. And an information recommendation method, wherein a value given a weight calculated according to a user type of the information search requesting user is used as an element value. When a user makes an information search request by a client device, the server receives the request, performs information search processing, and provides information to the client device.
Determine the concentration of interest from the content played by the user or the content attribute, the content attribute variation, and the burstiness of the content playback, determine the type of user by setting a threshold value for this concentration value,
Generate a preference vector that expresses the user's preference by a method according to the type of the user,
A user set having a preference vector having a similarity equal to or greater than a certain value with a user's preference vector for information provision,
A server device characterized in that information to be provided to a user who requests information provision is determined from information corresponding to the user included in the user set. When a user makes an information search request by a client device, the server device receives the request, performs information search processing, and provides information to the client device.
Means for receiving content playback information from the client device when the user plays back the content in the client device;
Content information acquisition means for acquiring content attribute information corresponding to a content ID included in the received content reproduction information from a content information database;
Reproduction information registration means for associating the content reproduction information with the acquired content attribute information and recording it in a content reproduction information database;
A user type determination unit that receives an information search request specifying user information from the client device, and determines a user type of an information search request source user based on content reproduction information recorded in the content reproduction information database;
A well-known degree calculating means for calculating, as a numerical value, a well-known degree of content or content attribute included in the content reproduction information of the information search requesting user;
Based on the content reproduction information of the user recorded in the content reproduction information database and the well-known degree of the content or content attribute, a vector expressing the user's preference is generated according to the type of the user, and this vector is generated by the user. Preference vector generation means for recording in a preference vector information database as a preference vector,
A preference similarity calculating means for calculating a similarity between the preference vector of the information search requesting user and the preference vector of the user recorded in the preference vector information database;
Select a user having a high preference vector similarity with the information search requesting user, and extract information useful for the user who transmitted the information search request from the content or content attribute information included in the content reproduction information of the selected user Search result extraction means for transmitting the extracted information to the client device as an information search result;
The server apparatus characterized by having. The server device according to claim 6, wherein
The user type is based on the content playback information recorded in the content playback information database, whether the content played back by the user or the content attribute of the user and whether the content playback is concentrated within a specific time range The server apparatus is characterized in that the determination is made based on an index value calculated by reflecting the burst property of content reproduction in which whether the content is continuously reproduced is represented by a numerical value. The server apparatus according to claim 6 or 7,
The preference vector has a content ID or content attribute of content that the user has played in the past as an element of the vector, and has a reproduction frequency of content having the content ID or content attribute as an element, and a well-known degree of the content or content attribute. In addition, the server apparatus is characterized in that a value given a weight calculated according to a user type of the information search requesting user is used as an element value.   A program characterized in that the information recommendation method or server device according to any one of claims 1 to 8 is written in a computer program and is executable.   A recording medium on which the information recommendation method or server device according to any one of claims 1 to 8 is recorded so as to be executable by a computer.

Images