JP4344339B2 - Information evaluation device, content search device, information evaluation method, content search method, program thereof, and recording medium - Google Patents

Description

  The present invention relates to a technique for evaluating the reputation of information provided in a community on a network.

In recent years, many communities that share information among users who share the same interest have been formed on the Internet. Examples of such communities include 1) a problem solving community, 2) a product / service evaluation community, 3) a user creation community, and 4) a topic-based community.
A problem-solving community is a community that provides user-supported mutual support for problems in a topic (for example, troubles in product use). There are user forum type customer support sites such as "Sun Support Forums".

The product / service evaluation community is a community for sharing user reviews on products and services. For example, “cosme.net” from iStyle Inc. and “kakaku.com” provided by Kakaku.com Inc. is there.
The user-creative community is a community for the purpose of creating a new service driven by the user, and examples thereof include “Tanokomu” of Engine Co., Ltd. and “Fancy Life” of Elephant Design Co., Ltd.
The topic-based community is a community where discussions and information sharing are performed between users on a certain topic, and the topics handled vary widely depending on the community. Examples include “Slashdot.jp” and “Benesseui Men's Park” of Benessewi Corporation.

  The present invention is not applied only to these specific communities, but can be applied to various information sharing communities including them. These information-sharing communities have the characteristic of gathering articles and content widely from community participants, and since the community includes various participants, information and content provided by these participants The quality of each is incredible. While there are people who provide useful and reliable information that many people are interested in, participants who input a large amount of information that no one is interested in, or information that is intentionally or unconsciously incorrect There are also participants who participate. In addition, there are also participants who intentionally or unconsciously input an evaluation that is not true with respect to the input of evaluation for information input to the community. Therefore, in order to extract useful information from information provided to the community, an information evaluation device or an information evaluation method that takes into account the information providing ability and information evaluation ability of the community participants is important.

An example of a system for scoring the importance of information posted to the community in consideration of the ability of such participants is described in Patent Document 1. In this document, the information reputation score that quantifies the usefulness of information posted to the community, the information provision score that quantifies the information provision ability of the participant (described as the information utterance score in the document), the participant's We have proposed a calculation algorithm that introduces three information evaluation scores that quantify information evaluation ability and calculates based on the interdependence of these scores.
However, in the method described in the document, the information provision score and the information evaluation score corresponding to the information provision ability and the information evaluation ability are expressed in the [Formula (2)] and the paragraph [ [Formula (3)] described in [0063], the information evaluation score and the information provision score are simply changed in proportion to the change in the information reputation score. For this reason, when a large amount of data is actually read based on this algorithm, the information provision score and information evaluation score of a person who obtained a high score in the initial stage will increase and diverge over time. There is a problem that a useful value cannot be obtained as it is. This is due to the fact that the method described in Patent Document 1 does not strictly define the semantics for the score, and the calculated score is not consistent as a whole.

  That is, in order to stabilize the three scores of the information reputation score, the information provision score, and the information evaluation score, and to obtain a score that matches the actual situation of the activities of the community participants, it is necessary to define the semantics of the score strictly. is necessary. For example, in Non-Patent Document 1, an index called PageRank representing the importance of a Web page is calculated based on a semantic called a random surfer model. In the random surfer model, the probability of arrival at each page when a hyperlink existing on a Web page is randomly clicked is defined as PageRank. A specific score calculation method based on this model is roughly as follows.

  When each web page is regarded as a node in graph theory and a link is an edge, if the number of web pages is N, it can be represented by an adjacency matrix L of N × N. This adjacency matrix is transposed, and a matrix divided by the number of links so that the sum of the column vectors is 1 is created. This is a transition probability matrix. In the case of a transition probability matrix, it is known that the final arrival probability vector is calculated by obtaining the eigenvector belonging to the maximum eigenvalue of the transition probability matrix according to the basic theory of linear algebra. Therefore, the arrival probability of each page by the random surfer model can be realized by calculating the eigenvector of the transition probability matrix. A number of methods for calculating the eigenvector have been proposed, and will not be described in detail. However, as the simplest method, the eigenvector belonging to the maximum eigenvalue can be obtained by repeating the multiplication of the probability transition matrix. There are methods to calculate.

Similarly, Non-Patent Document 2 proposes a web page scoring algorithm called HITS. In this algorithm, the Authority vector a representing the information providing capability of each page, by introducing an indication that hub vector h representative of the information evaluation capacity of each page, when the adjacency matrix L of the Web page and its transposed matrix L ^T, a is the eigenvector matrix L ^T L, h is proposed to calculate the eigenvectors of the matrix LL ^T.
JP 2003-316915 A S. Brin and L. Page, “The anatomy of a large-scale hyper-textual web search engine.” Proceedings of 7th WWW Conference, (1998). JM Kleinberg, “Authoritative sources in hyperlinked environment”, Journal of ACM, Vol. 48, pp. 604-632 (1999).

  However, as described above, the matrix modeling and the calculation method using the eigenvectors of the matrix are widely used as scoring methods for web pages. However, the method is applied to the above-described community analysis targeted by the present invention. How to do is not fully studied. In the case of a web page, an adjacency matrix, a transition probability matrix, etc. are generated based on one type of link called a hypertext link, whereas in the community analysis targeted by the present invention, participants are given information. There are two types of links: an information provision link corresponding to the information provision performed on the information and an information evaluation link corresponding to an information evaluation action performed on the information by the participant. In addition, there are many links, such as when the number of community participants is M and the number of information provided or evaluated is N, the information provision link and the information evaluation link each constitute an M × N adjacency matrix. It differs from Web page scoring in that point, and it is difficult to apply the algorithms such as PageRank and HITS described above as they are.

  The present invention has been made in view of the above-described problems, and includes 1) problem solving community, 2) product / service evaluation community, 3) user creation community, 4) topic-based community, and the like. To provide a technology that can calculate a consistent score as a whole by extracting information provision links and information evaluation links from participants to information and analyzing these two types of links Objective. That is, an object of the present invention is to provide a technique for analyzing two types of links and calculating a score of a stable value that is meaningful as an evaluation value.

In order to solve the above problems, in the present invention, at least one of an information reputation score, an information provision score, and an information evaluation score is calculated as follows. The information reputation score means a value obtained by quantifying the reputation of content, which is electronic information provided by a user. The information provision score means a value obtained by quantifying the information provision ability of the user who provided the content. Furthermore, the information evaluation score means a value obtained by quantifying the information evaluation ability of the user who has evaluated the content.
First, the information provision link information in which the user ID of the user who provided the content and the content ID that is the identifier of the content provided by the user are associated is stored in the information provision link storage means. Further, the information evaluation link storage means stores the user ID of the user who evaluated the content, the content ID which is the identifier of the content evaluated by the user, and the numerical value of the evaluation performed on the content by the content evaluator. Information evaluation link information in which evaluation information that is information is associated is stored.

Next, in the score calculation means, the above-described information provision link information and information evaluation link information are used, and a score transition matrix (information provision link information and information evaluation link information and any score (information evaluation score, information reputation score or Eigenvector of a matrix) indicating a defined relationship with any of the information providing scores). Then, the score calculation means stores the element of this eigenvector as any score in the information reputation score storage means, the information provision score storage means, or the information evaluation score storage means.
Here, the score transition matrix generated by the score calculation means is a matrix indicating a defined relationship between the information providing link information, the information evaluation link information, and any score (definition of semantics for the score). In the present invention, the eigenvalue of this score transition matrix is obtained, and the above-mentioned score is obtained. Therefore, the score obtained by the present invention is a strictly maintained definition of the introduced semantics. As a result, the obtained score becomes a mathematically meaningful and stable value under the introduced semantics.

Preferably, in the present invention, the score calculation means uses the information providing link information or the information evaluation link information and the eigenvector, and any score according to a relational expression indicating a defined relationship between these and any score. Is calculated. This makes it possible to calculate two or more stable scores that are mathematically meaningful.
In the present invention, preferably, the score calculation means calculates the score as follows. First, in the first adjacency matrix generation means, information provision link information stored in the information provision link storage means is searched, and for each combination of user ID and content ID, the number of information provision link information having the combination is calculated. Count and generate an adjacency matrix (information provision link adjacency matrix) with the number as an element of the matrix component corresponding to the combination. Further, in the second adjacency matrix generation means, the information evaluation link information stored in the information evaluation link storage means is searched, and for each combination of user ID and content ID, evaluation information of information evaluation link information having the combination And an adjacency matrix (information evaluation link adjacency matrix) having the added value as an element of a matrix component corresponding to the combination is generated and output. Then, in the eigenvector calculation means, the eigenvector of the score transition matrix (matrix indicating the relationship between the information provision link adjacency matrix, the information evaluation link adjacency matrix, and any score) from the information provision link adjacency matrix and the information evaluation link adjacency matrix And the element of this eigenvector is output as any score. Thereafter, the calculated score is stored in the information reputation score storage means, the information provision score storage means, or the information evaluation score storage means in the writing means.

Furthermore, preferably in the present invention, in the score calculation means, the information providing link adjacency matrix or the information evaluation link adjacency matrix and the eigenvector are substituted into a relational expression indicating a defined relationship between these and any score, Calculate and output the score.
Still further preferably in the present invention, the eigenvector calculation means calculates the eigenvector as follows. First, in the first normalization means, each element of the information provision link adjacency matrix is set to “content associated with the user ID in the information provision link storage means” for each set of elements corresponding to each user ID. Normalization is performed by a function depending on the number of IDs, and the normalized information provision link adjacency matrix is output. Further, in the second normalization means, each element of the information evaluation link adjacency matrix is set to “content associated with the user ID in the information evaluation link storage means for each set of elements corresponding to each user ID. Normalization is performed using a function depending on the number of IDs, and the normalized information evaluation link adjacency matrix is output. Further, the calculation means calculates an eigenvector of the score transition matrix using the normalized information providing link adjacency matrix and the normalized information evaluation link adjacency matrix, and outputs an element of the eigenvector as any score. Thereby, it is possible to prevent the information provision score and the information evaluation score of the user from being increased by providing a large number of contents or evaluating a large number of contents by the same user.

According to another embodiment of the present invention, the score transition matrix S is either ^{S = αC T C + (1} -α) E T E, S = CE T EC T, or S = ^{EC T} CE T. Note that C is an information provision link adjacency matrix, E is an information evaluation link adjacency matrix, α is a real number between 0 and 1, and β ^T means a transpose matrix of β.
Further, the present invention can also be applied to provision of content information stored in a content server device connected to a network. In this case, first, content information, which is content information, is received by the content information receiving means and stored in the content information storage means. Thereafter, the information providing link generating means uses the content information stored in the content information storing means, and the user ID of the user who provided the content and the content ID which is the identifier of the content provided by the user are: The associated information provision link information is generated and stored in the information provision link storage means. Further, in the information evaluation link generation means, using the content information stored in the content information storage means, the user ID of the user who evaluated the content, the content ID which is the identifier of the content evaluated by the user, and the content Information evaluation link information in which evaluation information which is numerical information of evaluation performed on the content by the evaluator is associated with each other is generated and stored in the information evaluation link storage means. Next, in the eigenvector calculation means, the information provision link information and the information evaluation link information are used, and the score transition matrix (the defined relationship between the information provision link information, the information evaluation link information and the information reputation score or the information provision score) Eigenvector of the matrix) is calculated, and the elements of the eigenvector are output as an information reputation score or an information provision score corresponding to each content information. Then, the writing means stores the information reputation score or the information provision score in the content information storage means in association with the corresponding content information. Thereafter, the search condition receiving unit receives the search condition transmitted from the user terminal device, and the search unit receives the content information satisfying the search condition as at least an information reputation score and an information provision score associated therewith. Together with one, it is extracted from the content information storage means and output.

The apparatus having such a configuration can process (for example, rearrange) each piece of content information that matches the search condition according to the calculated score. The apparatus can provide the user with a search result that appropriately suggests the importance of the content by using the processed information.
Further, preferably in the present invention, the score calculation means uses the information provision link information or the information evaluation link information and the eigenvector, and according to a relational expression indicating a defined relationship between these and the information reputation score or the information provision score, An information reputation score or an information provision score corresponding to the content information is calculated, and the score is output. More preferably, the information provision link generation means generates only the information provision link information corresponding to the content for which the evaluation exists. Then, in the information reputation score estimation means, information on the calculation result of the function depending on the information provision score of the user and the number of contents provided by the user is the content provided by the user and having no evaluation. Output as a reputation score. In this case, the score calculation speed can be improved without significantly reducing the validity of the score.

  Preferably, in the present invention, the content information includes content provider information (information corresponding to the user who provided the content), and the search means associates the content information satisfying the search condition with the content information. The information reputation score is extracted from the content information storage means together with the information reputation score. Then, the information reputation score totaling unit adds the information reputation score extracted by the search unit for each content provider information included in the content information, and outputs each added value together with the content provider information. Such an added value is a sum of evaluation values for content for each content provider only for content that matches the search condition. That is, by using the added value and the content provider information, it is possible to accurately identify an opinion reader of content that matches the search condition (details will be described later).

  In the present invention, preferably, the content information includes content provider information that is information corresponding to a user who provided the content, content address information indicating the address of the content, and pointer information indicating the address of other content. It is information including. Then, the information provision link generation unit generates the information provision link information by using the content provider information as the user ID and the content address information as the content ID. The information evaluation link generation means generates information evaluation link information using the content provider information as the user ID, the pointer information as the content ID, and a value determined by whether or not the content information includes pointer information as evaluation information. To do.

  In the present invention, it is preferable that the information evaluation link generation unit generates information evaluation link information using pointer information other than trackback. Thereby, it is possible to prevent erroneous information evaluation link information from being generated (details will be described later).

  As described above, according to the present invention, it is possible to calculate a score that is consistent and meaningful as an evaluation value from two types of link information, that is, information provision link information and information evaluation link information.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[First Embodiment]
First, a first embodiment of the present invention will be described.
<Configuration>
FIG. 1 is a diagram showing a configuration of an information evaluation system 1 in the present embodiment.
As illustrated in FIG. 1, the information evaluation system 1 of the present embodiment is connected to the information evaluation device 100 and the network 300 through a user terminal device 200-connected to the user i (i = 1... M). 1 to m.

  The information evaluation apparatus 100 is a known computer device including a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), a bus, an auxiliary storage device, a communication device, and the like. It is comprised by performing. As illustrated in FIG. 1, the information evaluation apparatus 100 of this example includes an information acquisition unit 101, an information provision link storage unit 102, an information evaluation link storage unit 103, a score calculation unit 104, and an information reputation score storage unit. 105, an information provision score storage unit 106, an information evaluation score storage unit 107, a score output unit 108, a temporary memory 111, and a control unit 110. In FIG. 1, lines connecting the respective units indicate the flow of information, but the description of the flow of information input to and output from the control unit 110 and the temporary memory 111 is omitted.

FIG. 2 is a diagram illustrating a detailed configuration of the information acquisition unit 101 in FIG.
As illustrated in FIG. 2, the information acquisition unit 101 in this example includes a communication unit 101a, a message authentication unit 101b, an information verification unit 101c, a content ID generation unit 101d, an information provision link generation unit 101e, Evaluation link generation means 101f and writing means 101g are provided. The arrows in FIG. 2 indicate the flow of information, but some descriptions of the flow of information input to and output from the control unit 110 and the temporary memory 111 are omitted.
FIG. 3 is a diagram illustrating a detailed configuration of the score calculation unit 104 in FIG.

As illustrated in FIG. 3, the score calculation unit 104 in this example includes a first adjacency matrix generation unit 104a, a second adjacency matrix generation unit 104b, an eigenvector calculation unit 104c, a score calculation unit 104d, and a writing unit 104e. And have. The eigenvector calculation means 104c includes a first normalization means 104ca, a second normalization means 104cb, and a calculation means 104cc. Note that the arrows in FIG. 3 indicate the flow of information, but the description of the flow of information input to and output from the control unit 110 is omitted.
FIG. 4 is a diagram illustrating a detailed configuration of the score output unit 108 in FIG.

As illustrated in FIG. 4, the score output unit 108 in this example includes a reception unit 108a, a message determination unit 108b, a score search unit 108c, a message generation unit 108d, and a transmission unit 108e. The arrows in FIG. 4 indicate the flow of information, but a part of the description of the flow of information input / output to / from the control unit 110 is omitted.
In addition, each means mentioned above is not limited to what exists in the same housing | casing, You may distributely arrange in several housings. The communication unit 101a, the reception unit 108a, and the transmission unit 108e described above are configured by communication devices such as a LAN (Local Area Network) card, a modem, a router, and a hub. The information provision link storage unit 102, the information evaluation link storage unit 103, the information reputation score storage unit 105, the information provision score storage unit 106, the information evaluation score storage unit 107, and the temporary memory 111 are, for example, a hard disk device, a flexible disk, An auxiliary storage device such as an MO (Magneto-Optical disc), a RAM, a register in the CPU, a cache memory, and the like are included. The control unit 110, the message authentication unit 101b, the information verification unit 101c, the content ID generation unit 101d, the information provision link generation unit 101e, the information evaluation link generation unit 101f, the writing unit 101g, the first adjacency matrix generation unit 104a, the second For example, the adjacency matrix generation unit 104b, the eigenvector calculation unit 104c, the score calculation unit 104d, the writing unit 104e, the message determination unit 108b, the score search unit 108c, and the message generation unit 108d read a predetermined program into the CPU, for example. This is configured by executing the program.

Further, the user terminal devices 200-1 to 200-m (FIG. 1) are also known computer devices that operate based on a predetermined program.
<Operation>
Next, the operation of the information evaluation apparatus 100 of this embodiment will be described.
[Overview]
The information evaluation apparatus 100 according to this embodiment includes an information reputation score obtained by quantifying the reputation of content, which is electronic information provided by a user, an information provision score obtained by quantifying information provision ability of a user who provided the content, and content Is an apparatus for calculating an information evaluation score obtained by quantifying the information evaluation ability of the user who has evaluated the information. First, an outline of processing of the information evaluation apparatus 100 will be described. Each process of the information evaluation apparatus 100 is executed under the control of the control unit 110.

First, a plurality of user terminal devices 200-1 to 200-m receive a content providing message including two sets of <user ID, content> or <user ID, content ID, evaluation information> via the network 300. A content evaluation message including the triplet is transmitted to the information evaluation apparatus 100.
The information evaluation apparatus 100 receives these at the information acquisition unit 101. Here, when the received message is a content providing message, the information acquisition unit 101 obtains the user ID of the user who provided the content and the content ID that is the identifier of the content provided by the user. The associated information provision link information is stored in the information provision link storage means 102. On the other hand, when the received message is a content evaluation message, the information acquisition unit 101 includes the user ID of the user who evaluated the content, the content ID that is the identifier of the content evaluated by the user, Information evaluation link information associated with evaluation information, which is numerical information of evaluation performed by the evaluator on the content, is stored in the information evaluation link storage means 103.

  After that, the score calculation means 104 uses the information provision link information and the information evaluation link information stored in the information provision link storage means 102 and the information evaluation link storage means 103 to obtain a score transition matrix (information provision link information and information evaluation). The eigenvector of the matrix indicating the defined relationship between the link information and any score (details will be described later) is calculated, and the elements of this eigenvector are output as any score. The score calculation means 104 uses the information provision link information or the information evaluation link information and the eigenvector, and calculates and outputs any score according to a relational expression indicating a defined relationship between these and any score. To do. Then, the score calculation unit 104 stores the calculated scores in the information reputation score storage unit 105, the information provision score storage unit 106, or the information evaluation score storage unit 107.

After that, the score output means 108 searches the information reputation score storage means 105, the information provision score storage means 106, or the information evaluation score storage means 107 in response to the request from the user terminal devices 200-1 to 200-m. Information about the score is transmitted to the user terminal devices 200-1 to 200-m.
[Details of Processing of Information Acquisition Unit 101]
Next, details of the processing of the information acquisition unit 101 will be described.
FIG. 5 is a diagram illustrating a flow for explaining details of processing of the information acquisition unit 101. The details of the processing of the information acquisition unit 101 will be described below with reference to FIG.

First, the communication means 101a (FIG. 2) of the information acquisition means 101 is a content providing message including two sets of <user ID, content> transmitted from a plurality of user terminal devices 200-1 to 200-m, or A content evaluation message including a triplet of <user ID, content ID, and evaluation information> is received via the network 300 (step S1).
FIG. 6A shows an example of the data structure of the content providing message 121 received in this way, and FIG. The content providing message 121 in this example includes header information 121a indicating its attributes, bit configuration, and the like, a user ID 121b of the user who provided the content 121c, and the provided content 121c. Here, the content 121c is electronic information input by the user. Examples of the content 121c include text information such as “00 restaurants are delicious”, music information created by the user, image information, moving images, and the like. The user ID 121b is the ID of the user who provided the content 121c. In the case of a membership-based community, the ID used when logging in to the system when writing the content 121c is used. The content ID 121c is an identifier uniquely assigned to the content.

  Further, the content evaluation message 122 of this example includes header information 122a indicating its attributes, bit configuration, etc., a user ID 122b of the user who evaluated the content, and a content ID 122c which is an identifier of the content evaluated by the user. And evaluation information 122d which is numerical information of evaluation performed on the content by the content evaluator. Here, the user ID 122c is the ID of the user who has evaluated the content. In the case of a membership-based community, the user ID 122c is an ID used when logging into the system when writing the content evaluation. In addition, the evaluation information 122d is information representing evaluation for the content 122d. In this embodiment, the evaluation information 122d is information that is explicitly input by the user using the user terminal devices 200-1 to 200-m. As an example of the evaluation information, there is numerical information that takes a value in a predetermined range such as “+1”, “+2”, “+3”, and the like.

  The communication unit 101a stores the received content provision message and content evaluation message in the temporary memory 111, and then the message authentication unit 101b performs authentication processing of the user terminal devices 200-1 to 200-m (step S2). This authentication process is performed when it is desired to prohibit one user from evaluating the same content a plurality of times, or when it is desired to prohibit providing a content falsely with another user. This authentication process is executed between the message authentication unit 101b and the user terminal devices 200-1 to 200-m via the communication unit 101a and the network 300 according to an authentication protocol using an existing password or the like. Note that the process of step S2 can be omitted depending on the application field.

  The result of the authentication process in step S2 is transmitted to the control means 110. When the authentication process is NG, the control unit 110 ends the session with the user terminal device 200-1 to 200-m. On the other hand, if the authentication process is OK, the control unit 110 causes the information verification unit 101c to determine whether the received message is a content providing message or a content evaluation message (step S3). Specifically, the information verification unit 101c in this example reads the message received by the communication unit 101a in step S1 from the temporary memory 111, and refers to the header information 121a or 122a so that the message is the content provision message 121. Or content evaluation message 122 (FIG. 6). The determination result by the information verification unit 101 c is sent to the control unit 110.

  Here, when the determination result received by the control unit 110 is a determination result indicating that “the message is a content providing message”, the control unit 110 gives an instruction to the content ID generation unit 101d to generate a content ID. . Upon receiving this instruction, the content ID generation unit 101d reads the content provision message received by the communication unit 101a in step S1 from the temporary memory 111. Then, the content ID generating unit 101d generates a unique content ID for the content included in the read content providing message and stores it in the temporary memory 111 (step S4). As a method for generating the content ID, for example, there are a method of using a one-way hash value of the entire content and a method of sequentially assigning serial numbers. When the content ID is generated, the information providing link generating unit 101e reads the content providing message received from the temporary memory 111 at step S1 and the content ID generated at step S4. And the information provision link production | generation means 101e produces | generates the triple (information provision link information) of <user ID, content ID, content> using these, and stores it in the temporary memory 111 (step S5). Thereafter, the information providing link information is read from the temporary memory 111 by the writing means 101g and stored in the information providing link storage means 102 (step S6).

  On the other hand, when the determination result of step S3 received by the control unit 110 is a determination result that “the message is a content evaluation message”, the control unit 110 sends information evaluation link information to the information evaluation link generation unit 101f. Give instructions to generate. Upon receiving this instruction, the information evaluation link generation unit 101f reads the content evaluation message received in step S1 from the temporary memory 111, and includes information on the triplet of <user ID, content ID, evaluation information> contained therein. It produces | generates as evaluation link information and stores it in the temporary memory 111 (step S7). Thereafter, the information providing link information is read from the temporary memory 111 by the writing means 101g and stored in the information evaluation link storage means 103 (step S8).

FIG. 7A is a diagram illustrating a data configuration of the information provision link information 131a to 131e stored in the information provision link storage unit 102 as described above. As shown in FIG. 7A, the information provision link storage unit 102 stores and manages information provision link information 131a to 131e in which a user ID, a content ID, and a content are associated with each other.
FIG. 7B is a diagram illustrating a data configuration of the information evaluation link information 132 a to 132 f stored and managed in the information evaluation link storage unit 103. As shown in FIG. 7B, the evaluation link storage means 103 stores and manages information evaluation link information 132a to 132f in which a user ID, a content ID, and evaluation information are associated with each other. The correspondence relationship between the content ID and the information content is stored and managed in the information provision link storage unit 102 shown in FIG.

  In this embodiment, the information acquisition unit 101 directly receives content and evaluation information from the user terminal devices 200-1 to 200-m, and provides information provision link information 131a to 131e and information evaluation link information 132a to 132f as information provision links, respectively. The data is stored in the storage unit 102 and the information evaluation link storage unit 103. However, the information acquisition unit 101 may extract the information providing link information 131a to 131e and the information evaluation link information 132a to 132f from an existing electronic bulletin board system or Web server device. For example, in a problem-solving community, the present invention may be applied with the action “question” or “answer” as “providing content”. That is, the “question” or “answer” written on the electronic bulletin board may be “content”, and the user who performed the writing may be “user who provided the content”. In addition, the user who referred to or answered the “question” is defined as the “user who evaluated the content”, and the number of times the reference to the “question” or the number of answers to the question is evaluated. That is, the present invention may be applied as “evaluation for content”. Further, the present invention may be applied to an act of selecting a most useful answer from among “answers” by the questioner as an evaluation of the goodness of the “question”, that is, “evaluation of content”. Similarly, even in a product / service evaluation type community, the present invention may be applied with the posting of a review article as “provision” of “content” and the support of other participants for this as “evaluation of content”. Other specific examples will not be given, but the same applies to topic-based communities and user-creative communities. In order to realize this, it is necessary to create a robot that collects information from the Web server device that constitutes such an existing community, but a general robot creation method that collects information from the We server device The method described in the literature: Kevin Hemenway, Tara Calishain, “Spidering hacks”, O'Reilly (2003), etc. can be used. When such an embodiment is taken, the URL (Uniform Resource Identifier) of the resource in which the content exists may be used as the content ID of the information providing link information 131a to 131e shown in FIG. . In this case, no content column is required for the information providing link information 131a to 131e.

[Details of processing of score calculation means 104]
Next, details of the processing of the score calculation means 104 (FIGS. 1 and 3) will be described.
First, the principle for the score calculation means 104 to calculate each score will be described.
As described above, the information provision link information 131a to 131e and the information evaluation link information 132a to 132f stored in the information provision link storage means 102 and the information evaluation link storage means 103 are logically as shown in FIG. It can be thought of as showing two types of links from content users to content. In FIG. 8, the circle on the left indicates the user and the square on the right indicates the content. Moreover, the broken line of this figure represents the association (information provision link) of the user and content corresponding to the user ID in the information provision link information 131a to 131e. Further, the solid line in this figure represents the association (information evaluation link) between the user corresponding to the user ID in the information evaluation link information 132a to 132f and the content corresponding to the content ID. Moreover, the numerical value given to the information evaluation link represents the evaluation information in the information evaluation link information 132a to 132f.

のような利用者i (i=1…m)を行とし、コンテンツj (j=1…n)を列とする隣接行列で表現することができる。すなわち、情報提供リンクは行列C=[c_ij] (i=1…m, j=1…n)で表現できる。ただし、c_ijは行列Ｃのｉ行ｊ列目の要素を示し、iからjにリンクがあればc_ij=1、なければc_ij =0となる。また情報評価リンクは行列E=[e_ij] (i=1…m, j=1…n)で表現できる。ただし、e_ijは行列Ｅのｉ行ｊ列目の要素を示す。またe_ij は、報評価リンク情報１３２ａ〜１３２ｆにおいて、利用者iに対応する利用者ＩＤと、コンテンツjに対応するコンテンツＩＤとに対応付けられている評価情報の値である。そして、e_ijの値が大きい程、コンテンツｊが利用者ｉから強い支持を受けていることになる。尚、以下では特に断りがない限り、行列Ｃ及び行列Ｅは、一般化された行列C=[c_ij] (i=1…m, j=1…n)及びE=[e_ij] (i=1…m, j=1…n)を意味する。 These links

User i (i = 1... M) as a row and content j (j = 1... N) as a column. That is, the information providing link can be expressed by a matrix C = [c _ij ] (i = 1... M, j = 1... N). Here, c _ij indicates the element in the i-th row and j-th column of the matrix C, and c _ij = 1 if there is a link from i to j, and c _ij = 0 if there is no link. The information evaluation link can be expressed by a matrix E = [e _ij ] (i = 1... M, j = 1... N). Here, e _ij indicates the element in the i-th row and j-th column of the matrix E. E _ij is a value of evaluation information associated with the user ID corresponding to the user i and the content ID corresponding to the content j in the report evaluation link information 132a to 132f. The larger the value of e _ij, the stronger the content j receives from the user i. In the following, unless otherwise specified, the matrix C and the matrix E are generalized matrices C = [c _ij ] (i = 1... M, j = 1... N) and E = [e _ij ] (i = 1 ... m, j = 1 ... n).

In this embodiment, two types of links from matrix C and matrix E respectively, to calculate the three scores information providing scores a _i, information evaluation score h _i and information reputation score r _j. The meaning and notation of these scores are described below.
The information provision score a _i is an index indicating how much the user i (i = 1... M) has provided content supported by the community in the past. The higher this score, the more likely this user is to be able to contribute to the community in terms of providing information. Here, the information provision score a _i of all users is expressed as a = [a ₁ ... A _m ] ^T, and this is hereinafter referred to as an information provision vector. Here, M ^T is a transposed matrix of the matrix M.

The information evaluation score h _i is an index showing how much user i (i = 1 ... m) has input evaluation information according to the direction of the community in the past for the content provided to the community. . The higher this score, the more likely this user is to contribute to the community in terms of information input. Here, the information evaluation score h _i of all users is expressed as h = [h ₁ ... H _m ] ^T, and this is hereinafter referred to as an information evaluation vector.
The information reputation score r _j is an index indicating how much the content j (j = 1... N) is supported by the user (ie, in line with the direction of the community). The higher this score, the more the information is considered to be supported by the community. Here, the information reputation score r _j of all contents is expressed as r = [r ₁ ... R _n ] ^T, and this is hereinafter referred to as an information reputation vector.

The user's information provision score, the information evaluation score, and the content information reputation score affect each other according to the following rules.
(1) Content provided by a user with a high information provision score is content having a high information reputation score.
(2) Content evaluated by a user having a high information evaluation score is content having a high information reputation score.
(3) A user who provides content with a high information reputation score is a user with a high information provision score.

(4) A user who evaluates content having a high information reputation score is a user having a high information evaluation score.
In this form, corresponding to these four rules,
r ＝ C ^T a… (1)
r ＝ E ^T h… (2)
a ＝ Cr… (3)
h = Er (4)
The four mathematical relationships are defined. However, in order to make both the formula (1) and the formula (2) compatible, the formula (1) and the formula (2) are integrated by a convex combination,
r = αC ^T a + (1-α) E ^T h (5)
Is used. Here, α is a real number whose domain is [0, 1], and is adjusted according to the characteristics of the community. That is, the closer the α is to 1, the more importance is placed on the information providing link, and the closer it is to 0, the more important is the information reputation link that the information reputation score is calculated.

Substituting Equation (3) and Equation (4) into Equation (5) above,
r = αC ^T Cr + (1-α) E ^T Er
= (ΑC ^T C + (1-α) E ^T E) r… (6)
Get. (αC ^T C + (1-α) E ^T E) is an n × n square matrix, which is defined as a score transition matrix S (“definition of information providing link information, information evaluation link information, and any score) Equivalent to the score transition matrix indicating
r = Sr (7)
It becomes. Generally, r that satisfies the above equation (7) does not exist, but according to the basic theory of linear algebra, when the element of S is a non-negative real number,
λr = Sr (8)
There exists a real vector r that satisfies. However, λ is a constant and is called an eigenvalue of S, and r is called an eigenvector of S. An example of a specific method for calculating these eigenvalues and eigenvectors will be described later. If r (information reputation vector) can be calculated in this way, the expression (3) and the expression (4) (" A) (information provision vector) and h (information evaluation vector) can be calculated respectively.

In addition, since the element e _{ij of the} matrix E is not normalized, there is a problem that if a specific person inputs evaluations for a lot of information, a very high information evaluation score is obtained. In order to prevent such an action, the matrix E may be normalized so that the sum of the values of evaluation information provided by one user becomes 1, like PageRank. In other words, the matrix E may be normalized so that the sum of the elements of the row vectors of the matrix E becomes 1, and the score may be calculated using the normalized matrix. Specifically, the following matrix E ′ may be generated and used.
E '= [e' _ij ] (i = 1… m, j = 1 ... n)
e ' _ij = e _ij / Σ _{j = 1..n} (e _ij )
Similarly, normalization may be performed so that the sum of the elements of the row vectors of the matrix C becomes 1, and each score may be calculated using the matrix C after normalization. In addition, you may normalize so that the sum total of the element of the row vector of the matrix C or E may become a value other than one. Also, these normalizations are not essential, and the score may be calculated without performing such normalization depending on the application.

Based on the above concept, the score calculation means 104 calculates r (information reputation vector), a (information provision vector), and h (information evaluation vector) from the matrices C and E, and provides stable information. Score a _i , information evaluation score h _i and information reputation score r _j are calculated.
FIG. 9 is a diagram illustrating a flow for explaining the processing of the score calculation unit 104. Hereinafter, the processing of the score calculation means 104 will be described with reference to this figure. Hereinafter, the scores are obtained by normalizing the matrices C and E so that the sum of the elements of the row vectors of the matrices C and E becomes 1.

を生成することになる。 First, the first adjacency matrix generation unit 104a of the score calculation unit 104 generates a matrix C, and the second adjacency matrix generation unit 104b generates a matrix E (step S11).
In the case of this example, first, the first adjacency matrix generation unit 104a searches the information provision link information stored in the information provision link storage unit 102 and has the combination for each combination of the user ID and the content ID. Count the number of information link information. Then, the first adjacency matrix generation unit 104a generates a matrix C, which is an adjacency matrix (information providing link adjacency matrix) with the number as an element of the matrix component corresponding to the combination, and outputs the matrix C to the temporary memory 111. To store. For example, in the case of the example of FIG. 7A, (user ID, content ID) is (A0001, C0001), (A0001, C0002), (A0002, C0003), (A0002, C0004), (A0003, C0005). There is one piece of information providing link information. Further, matrix components (rows, columns) corresponding to combinations of these user IDs and content IDs are respectively (1, 1), (1, 2), (2, 3), (2, 4), ( 3, 5), the first adjacency matrix generation means 104a

Will be generated.

を生成する。
尚、行列Ｃ，Ｅ生成の際、第１隣接行列生成手段１０４ａ或いは第２隣接行列生成手段１０４ｂは、各利用者ＩＤ及びコンテンツＩＤの情報も一時メモリ１１１に格納しておく。 Next, the second adjacency matrix generation unit 104b searches the information evaluation link information stored in the information evaluation link storage unit 103, and information evaluation link information having the combination for each combination of the user ID and the content ID. Add the value of the evaluation information. Then, the information evaluation link storage means 103 generates a matrix E that is an adjacency matrix (information evaluation link adjacency matrix) using the added value as an element of a matrix component corresponding to the combination, and outputs the matrix E to the temporary memory 111. To store. For example, in the example of FIG. 7B, (user ID, content ID, evaluation information) are (A0001, C0003, +2), (A0001, C0004, +3), (A0002, C0001, +2), (A0002). , C0002, +1), (A0002, C0005, +1), and (A0003, C0004, +3), one piece of information evaluation link information exists. Further, matrix components (rows, columns) corresponding to combinations of these user IDs and content IDs are respectively (1,3), (1,4), (2,1), (2,2), ( 2, 5), (3, 4), the second adjacency matrix generation means 104b

Is generated.
When generating the matrices C and E, the first adjacency matrix generation unit 104a or the second adjacency matrix generation unit 104b also stores information on each user ID and content ID in the temporary memory 111.

のように正規化される。尚、これは一例であり、行列ＣとＥの正規化は必須でなく、あえてこのような正規化を省略してもよいし、別の正規化方法を利用してもよい（他の正規化方法の例については後述する）。 After the matrices C and E are generated, the eigenvector calculation means next uses the eigenvector r of the score transition matrix S (matrix indicating the relationship between the information providing link adjacency matrix, the information evaluation link adjacency matrix, and any score). And the element of this eigenvector r is output as the information reputation score r _j . In this example, first, the eigenvector calculation unit 104 c reads the matrices C and E from the temporary memory 111. Then, in the first normalization means 104ca and the second normalization means 104cb, by a large number of provisions by the same user (such as posting a large number of articles) and a large number of evaluations by the same user (such as comments on a large number of articles). In order not to increase the information provision score or information evaluation score of the user, the matrices C and E are normalized, and the normalized matrices C and E are stored in the temporary memory 111 (step S12). In the case of this example, in the first normalizing means 104ca, each element of the matrix C (information providing link adjacency matrix) is changed to “information providing” for each set of elements corresponding to each user ID (that is, for each row). The link storage unit 102 normalizes by the function (normalization function 1) depending on the number of content IDs associated with the user ID, and outputs the normalized matrix C. Further, in the second normalization means 104cb of this example, each element of the matrix E (information evaluation link adjacency matrix) is set to “in the information evaluation link storage means 103 for each set of elements corresponding to each user ID. Normalization is performed using a function (normalization function 2) depending on the number of content IDs associated with the user ID, and the normalized matrix E is output. For example, when the normalization function 1 is c _ij / Σ _{j = 1..n} (c _ij ) and the normalization function 1 is e _ij / Σ _{j = 1..n} (e _ij ), The matrices C and E in equations (9) and (10) are respectively

Normalized as follows. Note that this is an example, and normalization of the matrices C and E is not essential, and such normalization may be omitted or another normalization method may be used (other normalizations). An example of the method will be described later).

Next, in the calculation means 104cc of the eigenvector calculation means 104c, the normalized matrices C and E stored in the temporary memory 111 are read, and the matrix S = (αC ^T C + (1−α) is read from the read matrices C and E. E ^T E) is calculated and stored in the temporary memory 111. This matrix S is called a score transition matrix (step S13). An example of the score transition matrix S generated as α = 0.3 from the matrices C and E shown in Expression (11) is shown below.

この例では、情報評判スコアｒ_３＝０．２５９２６，ｒ_４＝０.９６５８１となる。尚、固有ベクトルは固有値の数だけ存在するが、本形態では、例えば、固有値最大の固有ベクトル(principal eigenvector)を用いる。 Next, the computing means 104cc reads the score transition matrix S from the temporary memory 111, calculates its eigenvector r, and stores the element of this eigenvector r as the information reputation score r _j in the temporary memory 111 (step S14). As a method for calculating the eigenvector, many methods such as an iterative calculation method using a power method, a QR decomposition method, and a direct method can be applied. The eigenvectors for the score transition matrix S of Equation (12) calculated using the iterative calculation method based on the power method are shown below.

In this example, the information reputation score r ₃ = 0.25926, r ₄ = 0.96581. Although there are as many eigenvectors as the number of eigenvalues, in this embodiment, for example, the eigenvector having the maximum eigenvalue (principal eigenvector) is used.

この例では、情報提供スコアａ_２＝０．６１２５３、情報評価スコアｈ_１＝０．６８３１９，ｈ_３＝０．９６５８１となる。
次に、書き込み手段１０４ｅにおいて、上述の情報評判スコアｒ_ｊ、情報提供スコアａ_ｉ及び情報評価スコアｈ_ｉと、各利用者ＩＤ及びコンテンツＩＤとを、一時メモリ１１１から読み込む。そして、書き込み手段１０４ｅは、情報評判スコアｒ_ｊ（１４１ｂ）を対応するコンテンツＩＤ（１４１ａ）に関連付けて（図１０（ａ））情報評判スコア格納手段１０５に格納する。また、書き込み手段１０４ｅは、情報提供スコアａ_ｉ（１４２ｂ）及び情報評価スコアｈ_ｉ（１４２ｃ）を対応する利用者ＩＤ（１４２ａ）に関連付けて（図１０（ｂ））、情報提供スコア格納手段１０６及び情報評価スコア格納手段１０７にそれぞれ格納する（ステップＳ１６）。 Next, in the score calculation means 104d, the matrix C or matrix E and the eigenvector r are substituted into a relational expression indicating the defined relationship between these and any of the scores a and h, and the scores a and h are calculated. Then, the data is output to the temporary memory 111 and stored therein (step S15). Specifically, for example, the score calculation unit 104d reads the matrix C, the matrix E, and the eigenvector r from the temporary memory 111, and applies the equations (3) and (4) to them to obtain the scores a, h is calculated, and these elements are stored in the temporary memory 111 as the information provision score a _i and the information evaluation score h _i . The information provision score vector a and the information evaluation score vector h calculated by the equations (3) and (4) are shown below using the matrix C and the matrix E of the equation (11) and the eigenvector r of the equation (13). .

In this example, the information provision score a ₂ = 0.61253, the information evaluation score h ₁ = 0.68319, and h ₃ = 0.96581.
Next, the information reputation score r _j , the information provision score a _{i, the} information evaluation score h _i , each user ID, and the content ID are read from the temporary memory 111 by the writing unit 104 e. Then, the writing means 104e associates the information reputation score r _j (141b) with the corresponding content ID (141a) (FIG. 10 (a)) and stores it in the information reputation score storage means 105. Further, the writing means 104e associates the information provision score a _i (142b) and the information evaluation score h _i (142c) with the corresponding user ID (142a) (FIG. 10 (b)), and provides the information provision score storage means 106. And stored in the information evaluation score storage means 107 (step S16).

[Details of processing of score output means 108]
Next, details of the processing of the score output means 108 (FIGS. 1 and 4) will be described.
FIG. 11 is a diagram showing a flow for explaining details of the processing of the score output means 108. Details of the processing of the score output means 108 will be described below with reference to this figure.
The score output means 108 is a means for searching and returning the information reputation score, information provision score, and information evaluation score calculated and stored in the above steps in response to a request from the user terminal. When this processing is performed, first, the reception unit 108a of the score output unit 108 transmits the information reputation score acquisition message including the content ID or the information provision score including the user ID transmitted from the user terminal devices 200-1 to 200-m. An information evaluation score acquisition message is received through the network 300 (step S21). The received message is stored in the temporary memory 111.

Next, the message determination unit 108b reads the message received in step S21 from the temporary memory 111, and determines whether the message is an information reputation score acquisition message or an information provision score / information evaluation score acquisition message based on header information thereof. (Step S22). The determination result is sent to the control means 110.
If it is determined in step S21 that the information reputation score acquisition message has been received, the control unit 110 gives a search instruction for the information reputation score to the score search unit 108c. Receiving this, the score search means 108c searches the information reputation score storage means 105 using the content ID of the information reputation score acquisition message stored in the temporary memory 111 as a key, and the information reputation score (corresponding to this content ID ( FIG. 10A is extracted (step S23). Note that the extracted information reputation score is stored in the temporary memory 111.

  On the other hand, when it is determined in step S21 that the information provision score / information evaluation score acquisition message has been received, the control unit 110 gives a search instruction for the information provision score and the information evaluation score. Upon receiving this, the score search means 108c searches the information provision score storage means 106 and the information evaluation score storage means 107 using the user ID of the information provision score / information evaluation score acquisition message stored in the temporary memory 111 as a key. Then, an information provision score and an information evaluation score (FIG. 10B) corresponding to this user ID are extracted (step S24). Note that the extracted information provision score and information evaluation score are stored in the temporary memory 111.

  Next, in the message generation unit 108d, the score acquired in the above steps S23 and S24 and the corresponding content ID and user ID are read from the temporary memory 111, and the information is obtained, for example, for each content ID and user ID. Messages that are sorted or rearranged in descending order of score are generated (step S25). The generated message is stored in the temporary memory 111, further read by the transmission unit 108e, and transmitted to the user terminal devices 200-1 to 200-m through the network 300 (step S26).

By providing each score to the user in this way, the user can acquire useful information for the community from a large amount of information in the community and can obtain a ranking of the useful information. It becomes.
<Modifications>
The present invention is not limited to the embodiment described above. For example, the above-described various processes are not only executed in time series according to the above description, but may also be executed in parallel or individually as required by the processing capability of the apparatus that executes the processes. Good.

Further, for example, the calculation procedure from step S11 to step S16 shown in FIG. 9 can be calculated in exactly the same way by performing iterative calculation according to FIG. In this case, since r, a, and h are calculated simultaneously by iteration, steps S13 to S15 are performed in an integrated manner. This calculation procedure can also be expressed as shown in FIG. 12B in terms of matrix element expression. Here, prov (j) is the user who provided the content j (prov (j) ε {1 ... m}), and eval (j) is the set of users who are evaluating the content j (eval (j) ε {1 ... m}), prov (i) is a set of contents provided by user i (prov (i) ∈ {1 ... n}), eval (i) is a set of contents evaluated by user i (eval (i) ∈ {1 ... n}). Such information can be easily extracted by searching using the user ID or the content ID from the information providing link storage unit 102 and the information evaluation link storage unit 103 shown in FIG. In FIG. 12B, || x || ₁ and || x || ₂ represent the norm of the vector x, || x || ₁ is the sum of the absolute values of the elements of x, and || x || ₂ represents the square root (Euclidean norm) of the sum of the squares of the elements of x. As described above, in the present invention, since eigenvectors are used for the calculation of scores, it is possible to easily calculate these scores by applying an iterative algorithm.

FIG. 13 is a diagram illustrating a flow for explaining processing when r, a, and h are calculated by the algorithm of FIG. In FIG. 13, an arrow indicating a vector is omitted. In the case of the process of FIG. 13, first, similarly to step S11, the first adjacency matrix generation unit 104a and the second adjacency matrix generation unit 104b generate matrices C and E (step S31), and first normalize them. Normalized by means 104ca and second normalizing means 104cb and stored in temporary memory 111 (step S32). Further, in the calculation means 104cc, the vector a ⁽⁰⁾ = (1, ..., 1) ^T , the vector h ⁽⁰⁾ = (1, ..., 1) ^T , and the variable k = 0 are set, Store in the memory 111 (step S33). Next, in the operation means 104cc, the normalized matrices C and E and the vectors a ^(k) and h ^(k) are read from the temporary memory 111, and the vector r ^{(k + 1)} = αC ^T a ^(k) + (1-α) E ^T h ^(k) is calculated and stored in the temporary memory 111 (step S34). Α is determined in advance as described above. Next, the computing means 104cc reads the vector r ^{(k + 1)} from the temporary memory 111, normalizes it as shown in FIG. 12B, and stores it in the temporary memory 111 (step S35). Next, the score calculation unit 104d reads the normalized matrices C and E and the vector h ^{(k + 1)} from the temporary memory 111, and the vector a ^{(k + 1)} = C · r ^{(k + 1)} and h ^{(k + 1)} = E · r ^{(k + 1)} is calculated, and vectors a ^{(k + 1)} and h ^{(k + 1)} are stored in the temporary memory 111 (step S36). Next, the control means 110 reads the vectors r ^(k) and r ^{(k + 1)} from the temporary memory 111, and r _j ^(k) −r _j ^{(k + 1) for} all j∈ {1. It is determined whether or not <δ holds (step S37). R _j ^(k) represents the j-th element of the vector r ^(k) , and r _j ^{(k + 1)} represents the j-th element of the vector r ^{(k + 1)} . Also, δ is set to a sufficiently small constant set in advance. Here, if r _j ^(k) −r _j ^{(k + 1)} <δ is not satisfied, the control means 110 gives an instruction to the calculation means 104cc. Using this as a trigger, the calculation means 104cc reads the variable k from the temporary memory 111, calculates k + 1, and stores the calculation result as a new k in the temporary memory 111 (step S38). And the control means 110 returns a process to step S34. On the other hand, if it is determined in step S37 that r _j ^(k) −r _j ^{(k + 1)} <δ, the control unit 110 gives an instruction to the writing unit 104e. Using this as a trigger, the writing means 104e reads the vectors r _j ^{(k + 1)} , a ^{(k + 1)} , and h ^{(k + 1)} from the temporary memory 111, and provides each of these elements as an information reputation score and information. The scores and information evaluation scores are stored in the information reputation score storage means 105, the information provision score storage means 106, and the information evaluation score storage means 107 as described above (step S39).

In addition, when the above-described embodiment is applied to a community in which information evaluation link information monotonously increases, the longer the period of activity in this community, the smaller the impact of the new activity (content provision and evaluation) on the score. turn into. This is due to the influence of a large amount of information evaluation link information created by the user's past activities and the influence of the normalization method in the second normalization means 104cb (FIG. 3). Such a situation has problems such as reducing the incentive to perform the activities of the participants. Therefore, in such a case, it is effective to normalize the matrix E (information evaluation link adjacency matrix) so that these elements become smaller as time passes. Specifically, for example, in the second normalization means 104cb (FIG. 3), each element of the matrix E is read from the temporary memory 111, and a time decay factor such as the following equation is introduced to normalize the matrix E. Store in the temporary memory 111. Then, the matrix E ^(t) at the time t is used as the matrix E ⁽ information evaluation link adjacency matrix after normalization ⁾ of this embodiment.

ただし、ｔは現在時刻、time(e_ij)はe_ijが入力された時刻、ρは［0，1］を定義域とする減衰定数である。またe_ij ^(t)は現在時刻ｔにおける行列Ｅ^（ｔ）のｉ行ｊ列目の要素である。尚、この場合、第２正規化手段１０４ｃｂ（図３）は、例えば、情報評価装置１００内の図示していない内部時計等から出力される情報を現在時刻ｔとして用いる。さらに、この場合、情報取得手段１０１が情報評価リンク格納手段１０３に情報評価リンク情報を格納する際、その受信時間や格納時間等を各情報評価リンク情報に関連付けて格納しておく。そして、第２正規化手段１０４ｃｂ（図３）は、例えば、この情報評価リンク格納手段１０３に格納された受信時間や格納時間等を、それに関連付けられている情報評価リンク情報に対応するtime(e_ij)として利用する。

Where t is the current time, time (e _ij ) is the time when e _ij is input, and ρ is an attenuation constant with [0, 1] as the domain. E _ij ^(t) is an element in the i-th row and j-th column of the matrix E ^{(t) at} the current time t. In this case, the second normalization means 104cb (FIG. 3) uses, for example, information output from an internal clock (not shown) in the information evaluation apparatus 100 as the current time t. Further, in this case, when the information acquisition unit 101 stores the information evaluation link information in the information evaluation link storage unit 103, the reception time, the storage time, and the like are stored in association with each information evaluation link information. Then, the second normalization means 104cb (FIG. 3) uses, for example, the reception time and storage time stored in the information evaluation link storage means 103 as time (e) corresponding to the information evaluation link information associated therewith. _ij ).

Further, in this embodiment, the above-described equations (1) to (4) are used to perform the transformation as equation (6), and (αC ^T C + (1−α) E ^T E) is converted to the score transition matrix S. Then, the eigenvector r is obtained, and the vectors a and h are obtained from the equations (3) and (4). However, using the above equations (1) to (4),
a = Cr = (CE ^T ) h = (CE ^T E) r = (CE ^T EC ^T ) a… (15)
Or
^{h = Er = (EC T)} a = (EC T C) r = (EC T CE T) h ... (16)
Perform modifications such as, a score transition matrix may be placed with S = CE ^T EC ^T or S = EC ^T CE ^T from these equations. In this case, the calculation means 104cc (FIG. 3) calculates the eigenvector a or h, and the score calculation means 104d calculates using the equations (1) to (4), the vectors h, r or a , R. Note that the details of the processing in this case are the same as those in the above-described embodiment except that the score transition matrix is different and the types of equations and vectors used by the score calculation unit 104d are different. Of course, it goes without saying that a modification of the iterative calculation according to FIG. 12 may be used in this case.

In the above-described embodiment, the three scores of the information reputation score, the information provision score, and the information evaluation score are calculated. However, only a part of these scores may be calculated.
Furthermore, in the present embodiment, the score output means 108 is configured to accept two types of messages: an information reputation score acquisition message and an information provision score / information evaluation score acquisition message. However, depending on the application, the score output unit 108 may not accept the information provision score / information evaluation score acquisition message, and the process of step S24 (FIG. 11) may not be executed.

[Second Embodiment]
Next, a second embodiment of the present invention will be described.
<Principle>
As described above, the information evaluation apparatus of the present invention can be applied to a specific community such as a problem-solving community or a product / service evaluation community. However, the present invention is not limited to this, and the present invention can also be applied to ranking display of search results in various content search systems such as videos, images, music, and news on the Internet. The present embodiment is an example in which the present invention is applied to such a content search system. In this embodiment, a blog article is used as a specific example of content, and a case where the content search system is used as a blog article search system will be described.

In this embodiment, a blog article (content) has the following structural features. Note that the technique of this embodiment is not involved in the contents of blog articles.
(A) An individual is an editing subject and has a top page (blog site top) in an individual unit.
(B) There is a link from the top page to each article (blog entry), and each article has a URI (Uniform Resource Identifier).
(C) A mechanism (trackback) for generating a bi-directional link is provided.

  In order to apply the present invention to such a blog article, it is necessary to extract information provision link information and information evaluation link information from a set of collected blog articles. For this purpose, first, the user IDs of the information provider and the information evaluator must be acquired. Here, in the case of a blog article, the editing subject is an individual as described above, and the URI format of the top page of each blog article is almost determined for each blog hosting service. Therefore, in this embodiment, the URI of the blog site top is used as the user ID. In the blog system, basically, a permanent URI is given to each blog article entry. Therefore, in this embodiment, the URI given to the blog article is used as the content ID. In this way, the information providing link information is generated as a set including two sets of URI (user ID) of the blog site top and URI (content ID) of the blog article entry under the blog site. it can.

Similarly, the information evaluation link information includes two sets of a blog site top URI (user ID) and an external blog article URI (content ID) which is a link destination included in each article under the blog site. Can be generated as a set. This means that “There is a link from a blog article of a certain blogger (user) i to an external blog article j”, “This blogger (user) i is associated with a blog article j (content). This is because it is considered to be interested. In this case, the matrix E (information evaluation link adjacency matrix) corresponding to the information evaluation link information is generated by setting e _ij = 1 (with link) or e _ij = 0 (without link) depending on the presence or absence of the link. Can do. However, in the blog, a reverse link to the blog article of the link source is automatically generated in the linked blog article by the trackback function. That is, a mutual link is automatically formed. As a result, a reverse link (trackback link) from the blog article on the trackback side to the blog article on the trackback side is also set for the blog article that is not of interest. As a result, there is an attack called so-called trackback spam, in which many backlinks are made to articles that are not of interest. Therefore, in this embodiment, the link from the trackback side to the trackback side (that is, a normal link) is used for generating information evaluation link information, but the trackback link is not used for generating information evaluation link information. Of course, links between general blog articles that do not depend on trackbacks are used to generate information evaluation link information.

In the first embodiment, a method of normalizing the matrices C and E so that the total number of contents provided by one user and the sum of evaluation information values is 1 is exemplified. However, in the blog space, some people write thousands of blog articles, and many people stop writing a few blog articles. For all such various bloggers (users), if the matrices C and E are normalized so that the total number of contents and evaluation information provided by one user is 1, the provision of blog articles is small. When a link is made to a user's blog article, the information provision score calculated based on the link information becomes a very high value. This may be sufficient when the information provision ability of the user is really high, but there is a problem that the information provision score is easily influenced by luck. -On the other hand, when normalization is not performed, that is, for example, 1 is simply added to the element c _ij of the matrix C (information providing link adjacency matrix) with the addition of a blog article, and with the addition of a link between blog articles If 1 is simply added to the element e _ij of the matrix E (information evaluation link adjacency matrix), there is a problem that the user can intentionally increase the score by writing many articles.

Therefore, in this embodiment, this problem is solved by performing normalization at an intermediate level using values calculated according to the number of posts of blog articles and the number of links. Specifically, for example, as described below, the number of posts per user is normalized with the weight of the square root (Sqrt) of the number of posts of content and evaluation.
C '= [c' _ij ] (i = 1, ..., m, j = 1, ..., n)
c ' _ij = sqrt (Σ _{j = 1 ... n} (c _ij )) × c _ij / Σ _{j = 1 ... n} (c _ij ) = c _ij / sqrt (Σ _{j = 1 ... n} ( c _ij ))… (17)
E '= [e' _ij ] (i = 1, ..., m, j = 1, ..., n)
e ' _ij = sqrt (Σ _{j = 1 ... n} (e _ij )) × e _ij / Σ _{j = 1 ... n} (e _ij ) = e _ij / sqrt (Σ _{j = 1 ... n} ( e _ij ))… (18)
It has been empirically known that normalization by the above Sqrt function is particularly effective when the present invention is applied to a blog article. However, when the present invention is applied to another community such as an electronic bulletin board system (BBS), good results may be obtained by performing normalization using a natural logarithm (ln) function as follows.

C '= [c' _ij ] (i = 1, ..., m, j = 1, ..., n)
c ' _ij = ln (Σ _{j = 1 ... n} (c _ij )) × c _ij / Σ _{j = 1 ... n} (c _ij )… (19)
E '= [e' _ij ] (i = 1, ..., m, j = 1, ..., n)
e ' _ij = ln (Σ _{j = 1 ... n} (e _ij )) × e _ij / Σ _{j = 1 ... n} (e _ij )… (20)
<Configuration>
Next, based on the concept described above, the configuration of a content search system in which the information evaluation apparatus of the present invention is applied to search for blog articles will be described. In the following, differences from the first embodiment will be mainly described, and description of matters common to the first embodiment will be omitted.

FIG. 14 is a diagram showing a configuration of the content search system 1 in the present embodiment.
As illustrated in FIG. 14, the content search system 1 of this example includes user terminal devices 500-1 to 500-1 used by users i = 1,..., P who provide blog articles and collect content information. Content server apparatuses 600-1 to q for storing p and blog articles, a PING server (Weblog Updates Ping server) apparatus 700 for distributing update information of contents transmitted from the content server apparatuses 600-1 to 600-q, and a content search apparatus 1000, which are connected through a network such as the Internet. In addition, the content search device 1000 is collected with content information collection means 1100 that collects content (herein also referred to as blog articles and blog entries) information published on the Internet from the content server devices 600-1 to q. Content information storage means 1200 for accumulating and storing stored content information;-whenever a certain time interval or a certain number of content information is collected, information providing link information and information evaluation link information are extracted from the content information storage means, A search term is received from the information evaluation device 1300 for calculating each score and the user terminal devices 500-1 to 500-p via the network 300, and the keyword is matched with the keyword from the content data set recorded in the content information storage unit 1200. Search for content information to search The content information search means 1400 for returning the results to the user terminal devices 500-1 to 500-p, a temporary memory 1010, a control means 1020, and an internal clock 1030 are included. As in the first embodiment, the content server apparatuses 600-1 to 600-q, the PING server apparatus 700, and the content search apparatus 1000 are configured by causing a known computer to execute a predetermined program.

FIG. 15 is a diagram illustrating a detailed configuration of the content information collection unit 1100 of FIG.
As illustrated in this figure, the content information collection unit 1100 of this example includes a communication unit 1110, an update information reading unit 1120, a content information reading unit 1130, and a content information analysis unit 1140. The arrows in FIG. 15 indicate the flow of data, but the description of the flow of data input / output to / from the control means 1020 and the internal clock 1030 is omitted.
FIG. 16 is a diagram illustrating a detailed configuration of the information evaluation apparatus 1300 of FIG. In this figure, the same reference numerals as those in the first embodiment are used for portions common to the first embodiment.

As illustrated in this figure, the information evaluation apparatus 1300 of this example includes an information acquisition unit 1301, an information provision link storage unit 102, an information evaluation link storage unit 103, and a score calculation unit 104. Note that the arrows in FIG. 16 indicate the flow of data, but the description of the flow of data input to and output from the control means 1020 is omitted.
FIG. 17 is a diagram illustrating a detailed configuration of the information acquisition unit 1301 of FIG. In this figure, the same reference numerals as those in the first embodiment are used for portions common to the first embodiment.

As illustrated in this figure, the information acquisition unit 1301 of this example includes a reading unit 1301a, an information providing link generation unit 101e, an information evaluation link generation unit 101f, and a writing unit 101g. Note that the arrows in FIG. 17 indicate the flow of data, but the description of the flow of data input to and output from the control means 1020 is omitted.
FIG. 18 is a diagram illustrating a detailed configuration of the score calculation unit 104 in FIG. In this figure, the same reference numerals as those in the first embodiment are used for portions common to the first embodiment.

As illustrated in this figure, the score calculation unit 104 of this example includes a first adjacency matrix generation unit 104a, a second adjacency matrix generation unit 104b, an eigenvector calculation unit 104c, a score calculation unit 104d, and a writing unit 104e. Yes. The eigenvector calculation unit 104c includes a first normalization unit 104ca, a second normalization unit 104cb, and a calculation unit 104cc. Note that the arrows in FIG. 18 indicate the flow of data, but the description of the flow of data input to and output from the control means 1020 is omitted.
FIG. 19 is a diagram illustrating a detailed configuration of the content information search unit 1400 in FIG. In this figure, the same reference numerals as those in the first embodiment are used for portions common to the first embodiment.

As illustrated in this figure, the content information search unit 1400 of this example includes a reception unit 1410, a content information search unit 1420, a content information provision list generation unit 1430, and a transmission unit 1440.
<Operation>
Next, the operation of the content search apparatus 1000 of this embodiment will be described.
[Overview]
The content search system 900 according to the present embodiment is a system that processes and provides information related to content stored in the content server apparatuses 600-1 to 600-p using the above-described score. When providing this information, first, the content information collection unit 1100 of the content search device 1000 collects information on blog articles (contents) stored in the content server devices 600-1 to 600-p, and the content information storage unit 1200 is collected. To store. Then, in the information evaluation apparatus 1300, as in the first embodiment, each score is calculated from the content information stored in the content information storage unit 1200 and stored in the content information storage unit 1200 in association with the content information. . Then, the content information search unit 1400 searches the information stored in the content information storage unit 1200 according to the search conditions sent from the user terminal devices 500-1 to 500-p, and the search result is used as the user terminal device 500-1. Return to ~ p. Each process of the content search apparatus 1000 is executed under the control of the control means 1020.

[Details of score calculation processing]
FIG. 20 is a diagram illustrating a flow for explaining score calculation processing performed in the content search apparatus 1000 according to the present embodiment. Hereinafter, this process will be described with reference to FIG.
First, the control means 1020 refers to the internal clock 1030 and determines whether or not a predetermined time has elapsed since the last update information was read (step S101). Here, when it is determined that a predetermined time has elapsed, the control unit 1020 gives an instruction to the update information reading unit 1120, and the update information reading unit 1120 updates from the PING server apparatus 700 through the communication unit 1110 and the network 300. Information is read and stored in the temporary memory 1010 (S102). Here, as the update information, for example, summary information (RSS (Rich Site Summary)) of blog articles published by the Changes.Xml format published by the PING server apparatus 700, the blog hosting service, or the like can be exemplified.

  Next, the control unit 110 reads the update information stored in the temporary memory 1010, and using this, whether or not the content (blog article) stored in any of the content server devices 600-1 to 600-q has been updated. Is determined (step S103). If it is determined that there is no content update, the control unit 110 returns the process to step S101. On the other hand, when it is determined that the content is updated, the control unit 110 gives an instruction to the content information reading unit 1130, and the content information reading unit 1130 passes through the communication unit 1110 (content information receiving unit) and the network 300. Information regarding the updated content is acquired from the content server devices 600-1 to 600-q and stored in the temporary memory 1010. For the method and program code for collecting articles by crawling the content server devices (blog sites, etc.) 600-1 to q, for example, the documents “Tatsuhiko Miyagawa and Naoya Ito“ BIog Hacks ”, O'Reilly It is described in detail in “Japan”.

FIG. 22 is a conceptual diagram exemplifying blog information 1500 that is information regarding the content stored in the temporary memory 1010 as described above. In FIG. 22, only the blog information related to one blogger (user) is illustrated, but actually more blog information related to the logger is collected.
As illustrated in this figure, the blog information 1500 in this example includes a site top URI 1510 that is a URI of the blog site top, an article 1511 of the blog site top, and contents 1520 and 1530. The contents 1520 and 1530 are a blog URI 1521 which is a URI of a resource where the contents 1520 and 1530 are located, normal link URIs 1522 and 1532 which are link destinations set by the user who posted the contents 1520 and 1530, and an article 1523, respectively. 1533, and trackback URIs 1524, 1525, 1534, 1535, and 1536, which are reverse link destinations based on the trackback function.

  The content information analysis means 1140 reads such content information from the temporary memory 1010, analyzes it, extracts information as illustrated in FIG. 23 as a content information list 1600, and uses this as a table such as a DBMS The information is stored in the information storage unit 1200. Here, DocID is a column that stores a unique ID assigned to content such as a blog article collected by the content information collection unit 1100. The URI is a column for storing the URI (blog URI 1521, 1531, etc.) of the resource whose contents are disclosed. However, if the content is the blog site top, the URI (site top URI 1510 or the like) is stored in the URI column. “keywords” is a column for storing extracted words after performing morphological analysis on collected blog articles (articles 1511, 1523, 1533, etc.). The blog-URI is a column for storing the URI of the blog site top to which the blog article belongs (site top URI 1510, etc.). However, if the content is the blog site top itself, <nil> is stored. The Refer-URI is a column for storing URIs of other blog articles that are referred to as general links in the body of the blog article (normal link URIs 1522, 1532, etc.). However, <nil> is stored for a content that does not refer to another blog article as a general link, and when there are a plurality of links, URIs of a plurality of link destinations are stored. In the example of FIG. 23, URIs referred to as trackbacks (trackback URIs 1524, 1525, 1534, 1535, 1536, etc.) are not stored. Date is a column that stores the date and time when the blog article was acquired by patrol. For example, date information output from the internal clock 1030 is used as the date when the blog article is acquired. The Score is a column for storing the score calculated by the information evaluation apparatus 1300 as will be described later. However, when the content is the blog site top, the information provision score for the blogger (user) is stored, and when the content is the blog article, the information reputation score of the article is stored. Note that the information evaluation score is not used in this embodiment. Further, as described above, this system uses the URI of the blog site top (the URI of the blog-URI column) as the user ID. The URI of the URI and refer-URI column is used as the content ID. The data in the blog-URI column corresponds to “content provider information that is information corresponding to the user who provided the content” in the claims, and the data in the URI column is “the content in question” in the claims. The data in the refer-URI column corresponds to “information including pointer information indicating the address of other content”. Further, the data of the URI column, the keywords column, the blog-URI column, and the refer-URI column of the content information list 1600 correspond to “content information” in the claims.

  Thereafter, the control means 1020 determines whether or not a predetermined number or more of content information before score calculation has been accumulated in the content information storage means 1200 (step S106). If it is determined that the content information before score calculation is not accumulated in the content information storage unit 1200 in a predetermined number or more, the control unit 1020 returns the process to step S101. On the other hand, if it is determined that a predetermined number or more of content information before score calculation is stored in the content information storage unit 1200, the information evaluation device 1300 generates information provision link information and information evaluation link information. Then, the information evaluation device 1300 calculates a score from them, and stores it in the content information storage unit 1200 in association with the content information (step S107). It should be noted that step S107 may be executed when the fixed period has elapsed.

[Details of Step S107]
First, the information acquisition means 1301 extracts the blog-URI and URI column elements (content information) stored in the content information storage means 1200, and uses them to describe them in FIG. 24 (a). Information provision link information 1611 to 1615 in which two combinations of a user ID and a content ID are associated with each other are generated and stored in the information provision link storage unit 102. In addition, the information acquisition unit 1301 extracts the elements (content information) of the blog-URI and refer-URI columns stored in the content information storage unit 1200, and uses them to describe them in FIG. 24 (b). Information evaluation link information 1621 to 1626 associated with such a triplet of user ID, content ID, and evaluation information is generated and stored in the information evaluation link storage means 103.

FIG. 21 is a diagram illustrating a flow for explaining details of the calculation processing of the information provision link information and the information evaluation link information. The details of the calculation processing of the information provision link information and the information evaluation link information will be described below with reference to this figure.
First, the control means 1020 substitutes an initial value for the variable v and stores it in the temporary memory 1010 (step S111). Specifically, for example, the top DocID of the content information list 1600 for calculating the score is substituted for v.
Next, the reading means 1301a reads the content information (see FIG. 23) of the v-th row from the content information storage means 1200 and stores it in the temporary memory 1010. Then, in the information provision link generation means 101e, the data (content provider information) in the v-th row blog-URI column is used as the user ID from the temporary memory 1010, and the data in the v-th URI column (content address). Information) as a content ID. And the information provision link production | generation means 101e produces | generates the information provision link information with which these were linked | related (step S112), and stores it in the temporary memory 1010. If the data in the blog-URI column is <nil>, the information provision link information for the row is not generated. Then, the writing unit 101g reads the information provision link information from the temporary memory 1010 and stores it in the information provision link storage unit 102 (step S113).

  Also, in the information evaluation link generation means 101f, the data (content provider information) in the v-th row-URI column is used as the user ID from the temporary memory 1010, and the data (pointer information) in the v-th refer-URI column. ) As a content ID. Further, the information evaluation link generation unit 101f sets, as evaluation information, a value determined by whether or not the content information list 1600 of the v-th row includes data (pointer information) in the refer-URI column. In this example, when the content information list 1600 in the v-th row includes the refer-URI column data, the evaluation information is set to +1. Then, the information evaluation link generation unit 101f generates information evaluation link information in which the extracted user ID, content ID, and evaluation information are associated with each other, and stores the information evaluation link information in the temporary memory 1010 (step S114). Then, the writing unit 101g reads this information evaluation link information from the temporary memory 1010 and stores it in the information evaluation link storage unit 103 (step S115). Note that the refer-URI column of the content information list 1600 stored in the content information storage unit 1200 of this example does not store a URI based on the trackback. Therefore, the information evaluation link generation unit 101f of this example generates information evaluation link information using a URI (pointer information) other than trackback. When the content information list 1600 is generated, the URI based on the trackback is also included in the content information list 1600, and when the information evaluation link information is generated, the information evaluation link generation unit 101f uses information other than the URI based on this trackback. Information evaluation link information may be generated.

  Next, the control unit 1020 reads the variable v stored in the temporary memory 1010 and determines whether or not it is the maximum value of DocID (maximum value of data in the DocID column in the content information list 1600) ( Step S116). Here, if the variable v is not the maximum value of DocID, the control means 1020 calculates v + 1, stores the calculation result as a new variable v in the temporary memory 1010, and returns the process to step S112 ( Step S117). On the other hand, if the variable v is the maximum value of DocID, the calculation process of the information provision link information and the information evaluation link information ends.

  After that, as in the first embodiment, the information provision link stored in the information provision link storage means 102 and the information evaluation link storage means 103 in the first adjacency matrix generation means 104a and the second adjacency matrix generation means 104b. The matrices E and C are generated using the information and the information evaluation link. Then, as in the first embodiment, the eigenvector calculation means 104c calculates the eigenvector of the score transition matrix S from the matrices E and C as an information reputation score or information provision score, and the score calculation means 104d obtains another score. calculate. Note that the first normalization means 104ca and the second normalization means 104cb in this example normalize the matrices E and C according to, for example, the above-described equations (17), (18) or (19), (20), and calculate the calculation means. 104 cc calculates the eigenvector of the score transition matrix S using the normalized matrices E and C. Finally, the writing unit 104e associates the calculated scores with the content information list 1600 as described above and stores them in the content information storage unit 1200.

[Details of content information search processing]
Next, details of content information search processing performed by the content information search means 1400 will be described.
FIG. 25 is a diagram showing a flow for explaining details of the content information search processing. Hereinafter, details of the content information search processing will be described with reference to FIG.
In the case of this example, first, using a general web server device such as Apache (not shown) and the HTTP protocol, the user terminal devices 500-1 to 500-p are displayed with a screen for promoting the input of a search term. . The users of the user terminal devices 500-1 to 500-p input search words (search conditions) to the user terminal devices 500-1 to 500-p in accordance with this, and send messages including this to the content search device 1000 via the network 300. Send. The receiving means 1410 (FIG. 19) of the content information searching means 1400 receives the message including the search word transmitted from the user terminal devices 500-1 to 500-p (step S121) and stores it in the temporary memory 1010.

  Next, in the content information search means 1420, all the contents satisfying the condition specified by the search word from the content information stored in the content information storage means 1200 by reading the search word from the temporary memory 1010. Search for information. Then, the content information search means 1420 extracts the content information satisfying this condition from the content information storage means 1200 together with at least one of the information reputation score and the information provision score associated therewith and stores it in the temporary memory 1010 (step S122). Specifically, for example, the content information search unit 1420 reads the above-described search word from the temporary memory 1010 and searches the Keywords column of the content information list 1600 (FIG. 23) of the content information storage unit 1200 using the search word as a key. To do. Then, the content information search unit 1420 reads all the records (including the column information of the score) of the content information list 1600 in which a word matching the search word exists, and stores it in the temporary memory 1010.

Next, the content information provision list generation unit 1430 generates a content information provision list by sorting the records extracted from the temporary memory 1010 in step S122 using, for example, the information reputation score value as a key, and a message including the list. Is stored in the temporary memory 1010 (step S123). Specifically, the content provision information list generation means 1430 of this example adds HTML tags necessary for the user terminal devices 500-1 to 500-p to display this list using the web browser. Generate the assigned message.
Thereafter, the transmission unit 1440 reads the message generated in step S123 from the temporary memory 1010, and transmits the message as a search result for the search term to the user terminal devices 500-1 to 500-p that have transmitted the search term (step S1440). S124). The user terminal devices 500-1 to 500-p receive this message and display a content information list included in the message as a search result screen using, for example, a web browser.

<Modifications>
The present invention is not limited to the embodiment described above.
For example, the information provision link generation unit 101e does not generate the information provision link information for all the records stored in the content information storage unit 1200, but only the information provision link information corresponding to the content for which the evaluation exists. It may be generated.
FIG. 26 is a diagram illustrating the configuration of score calculation means 2104 corresponding to this modification. FIG. 27 is a diagram illustrating a flow for explaining the processing of this modification.
In the case of this modification, a score calculation unit 2104 is arranged instead of the score calculation unit 104. In addition, the process of FIG. 27 is executed instead of the process of FIG. 21 described above. That is, first, the control means 1020 assigns an initial value to the variable v and stores it in the temporary memory 1010 (step S131). Next, the reading means 1301a reads the content information (see FIG. 23) of the v-th row from the content information storage means 1200 and stores it in the temporary memory 1010. Then, the information providing link generation unit 101e searches for another row (record) having the value of the URI column of the content information of the v-th row as the value of the refer-URI column (step S132).

  Here, when it is determined that there is another row with the value of the URI column of the content information of the v-th row as the value of the refer-URI column, the information provision link generation unit 101e reads the v-th row from the temporary memory 1010. The blog-URI column data is read as the user ID, and the v-th row URI column data is read as the content ID. And the information provision link production | generation means 101e produces | generates the information provision link information with which these were linked | related (step S133), and stores it in the temporary memory 1010. Then, the writing unit 101g reads the information provision link information from the temporary memory 1010 and stores it in the information provision link storage unit 102 (step S134). On the other hand, if it is determined in step S132 that there is no other row in which the value of the URI column in the content information of the v-th row is the value in the refer-URI column, the information providing link generation unit 101e The information providing link information using the information is not generated.

  Next, the information evaluation link generation unit 101f reads from the temporary memory 1010 the data in the blog-URI column of the vth row as the user ID and the data of the refer-URI column in the vth row as the content ID. Further, the information evaluation link generation unit 101f sets, as evaluation information, a value that is determined depending on whether or not the content information list 1600 of the v-th row includes data of the refer-URI column. Then, the information evaluation link generation unit 101f generates information evaluation link information in which the extracted user ID, content ID, and evaluation information are associated with each other, and stores the information evaluation link information in the temporary memory 1010 (step S135). Then, the writing unit 101g reads this information evaluation link information from the temporary memory 1010 and stores it in the information evaluation link storage unit 103 (step S136).

Next, the control means 1020 reads the variable v stored in the temporary memory 1010, and determines whether or not it is the maximum value of DocID (step S137). Here, if the variable v is not the maximum value of DocID, the control means 1020 calculates v + 1, stores the calculation result as a new variable v in the temporary memory 1010, and returns the process to step S112 ( Step S138). On the other hand, if the variable v is the maximum value of DocID, the control means 1020 ends the calculation process of the information provision link information and the information evaluation link information.
Thereafter, in the information reputation score estimation means 2104f (FIG. 26), the calculation result of the function depending on the information provision score of the user and the number of contents provided by the user is evaluated as the content provided by the user. May be output to the content information storage unit 1200 and stored as an information reputation score.

FIG. 28 is an example of a flow for explaining the processing of the information reputation score estimating means 2104f in this case.
In this case, first, the control means 1020 assigns an initial value to the variable v and stores it in the temporary memory 1010 (step S141). Next, the information reputation score estimation unit 2104f determines whether or not data is stored in the score column of the content information list 1600 of the v-th row stored in the content information storage unit 1200 (see FIG. 23), for example (see FIG. 23). Step S142).

If it is determined that no data is stored, the information reputation score estimation unit 2104f searches the URI column of the content information list 1600 using the value of the blog-URI column of the content information list 1600 of the v-th row as a key. To do. The information reputation score estimator 2104f sets the value of the Score column on the line and the value of this key and URI column matches substituted into _{a v,} in the _{a v} (user of the information providing scores) temporary memory 1010 Store (step S143). Next, all the rows of the blog-URI column of the content information list 1600 are searched using the value of the blog-URI column of the content information list 1600 of the v-th row as a key, and the value of the blog-URI column is the blog of the v-th row. counting the number of rows equal to the value of -URI column, and stores it in _{n v} temporary memory 1010 as (the user number of contents provided) (step S144). Next, the information reputation score estimator 2104f the temporary reads the _{a v} and _{n v} from the memory 1010, it performs the calculation of the function that depends on the _{a v} and _{n v.} Then, the information reputation score estimating means 2104f stores this calculation result (information reputation score of the content provided by the user and having no evaluation) in the Score column of the content information list 1600 in the v-th row ( Step S145). As the function that depends on the _{a v} and _{n v,} for _example, can be exemplified _{r j = a v / sqrt (} n v). Here, sqrt (X) is a function that returns the square root of X. Further, a _v corresponds to the information providing scores of content provider v that provided the content j, n _v is equivalent to the number of providing the content of the content provider v.

When step S145 is completed, or when it is determined in step S142 that data is stored in the score column of the content information list 1600 of the v-th row stored in the content information storage unit 1200, the control unit 1020 Performs the following step S146.
First, the control means 1020 reads the variable v stored in the temporary memory 1010, and determines whether or not it is the maximum value of DocID (step S146). Here, if the variable v is not the maximum value of DocID, the control means 1020 calculates v + 1, stores the calculation result as a new variable v in the temporary memory 1010, and returns the process to step S142 ( Step S147). On the other hand, if the variable v is the maximum value of DocID, the process ends.
By adopting such an information reputation score calculation method, when a link is not made to almost all contents like a blog article (= no information evaluation link information), an object to be calculated by the information evaluation apparatus 1300 The number of contents can be significantly reduced. As a result, high-speed score calculation is possible.

  The content information search means may search for and output a content provider (blogger when applied to blog) related to the search word, instead of searching for an article related to the search word. Thereby, the user can find the blog site of the opinion leader related to the search term. And this user can acquire the comprehensive knowledge relevant to a search word by reading the blog site of an opinion reader. One method of generating a content provider list of information providers related to such a search term is to create a list of content providers in which at least one content including the search term is written, and use it as an information provision score. It is a method of sorting. However, the information provision score is calculated by extracting the information provision link information and the information evaluation link information in advance at a predetermined time interval before a search word is given. Therefore, the relevance between the value of the information provision score and the related word is low. Therefore, when the content provider list of the search result is sorted by the information provision score, if an information provider having a high information provision score happens to write an article containing one related word, it immediately becomes higher in the content provider list. There is a possibility.

Therefore, the information reputation score counting means in this example totals information reputation scores of a plurality of contents satisfying the search term condition for each content provider, and based on the total score, wrote an article about the search term. Sort the list of content providers to generate a content provider list. And the information reputation score totaling means of this example transmits this content provider list to the user terminal devices 500-1 to 500-p as a search result for the search word.
FIG. 29 is a diagram illustrating the configuration of such content information search means 2400. FIG. 30 is a diagram illustrating a flow for explaining the processing of the content information search means 2400.

When the receiving unit 1410 receives a search condition including a search word (step S151), the content information search unit 2400 stores the search condition data in the temporary memory 1010. Next, the content information search unit 1420 reads the search condition data from the temporary memory 1010, and the data in the keywords column satisfies the search condition from the content information list 1600 (FIG. 23) of the content information storage unit 1200. -All content information rows whose URI column is not <nil> are extracted together with the information reputation score (Score column data) associated with them, and stored in the temporary memory 1010 in association with each other (step S152). . Next, the information reputation score totaling unit 2450 adds the information reputation score stored in the temporary memory 1010 for each blog-URI column value (content provider information) of the content information associated therewith. Then, the information reputation score totaling means 2450 outputs the added value and the value of the corresponding blog-URI column to the temporary memory 1010 and stores it (step S153). After that, the content information provision list generation unit 1430 reads the added value and the corresponding blog-URI column value from the temporary memory 1010, and rearranges the blog-URI column value using the added value as a key. A message including the list is generated and stored in the temporary memory 1010. Further, this message is sent from the temporary memory 1010 to the transmission unit 1440, and the transmission unit 144 transmits this message to the user terminal devices 500-1 to 500-p that transmitted the search condition.
Needless to say, other modifications are possible without departing from the spirit of the present invention.
Further, when the above-described configuration is realized by a computer, processing contents of functions that each device should have are described by a program. The processing functions are realized on the computer by executing the program on the computer.

The program describing the processing contents can be recorded on a computer-readable recording medium. The computer-readable recording medium may be any medium such as a magnetic recording device, an optical disk, a magneto-optical recording medium, or a semiconductor memory. Specifically, for example, the magnetic recording device may be a hard disk device or a flexible Discs, magnetic tapes, etc. as optical disks, DVD (Digital Versatile Disc), DVD-RAM (Random Access Memory), CD-ROM (Compact Disc Read Only Memory), CD-R (Recordable) / RW (ReWritable), etc. As the magneto-optical recording medium, MO (Magneto-Optical disc) or the like can be used, and as the semiconductor memory, EEP-ROM (Electronically Erasable and Programmable-Read Only Memory) or the like can be used.
The program is distributed by selling, transferring, or lending a portable recording medium such as a DVD or CD-ROM in which the program is recorded. Furthermore, the program may be distributed by storing the program in a storage device of the server computer and transferring the program from the server computer to another computer via a network.

  A computer that executes such a program first stores, for example, a program recorded on a portable recording medium or a program transferred from a server computer in its storage device. When executing the process, the computer reads a program stored in its own recording medium and executes a process according to the read program. As another execution form of the program, the computer may directly read the program from a portable recording medium and execute processing according to the program, and the program is transferred from the server computer to the computer. Each time, the processing according to the received program may be executed sequentially. Also, the program is not transferred from the server computer to the computer, and the above-described processing is executed by a so-called ASP (Application Service Provider) type service that realizes the processing function only by the execution instruction and result acquisition. It is good. Note that the program in this embodiment includes information that is used for processing by an electronic computer and that conforms to the program (data that is not a direct command to the computer but has a property that defines the processing of the computer).

  In this embodiment, the present apparatus is configured by executing a predetermined program on a computer. However, at least a part of these processing contents may be realized by hardware.

FIG. 1 is a diagram showing the configuration of the information evaluation system in the first embodiment. FIG. 2 is a diagram illustrating a detailed configuration of the information acquisition unit in FIG. FIG. 3 is a diagram illustrating a detailed configuration of the score calculation means in FIG. It is the figure which illustrated the detailed structure of the score output means in FIG. FIG. 5 is a diagram illustrating a flow for explaining details of processing of the information acquisition unit. FIG. 6A shows an example of the data structure of the content providing message received in this way, and FIG. 6B shows an example of the content evaluation message. FIG. 7A is a diagram illustrating a data configuration of information providing link information. FIG. 7B is a diagram illustrating a data configuration of the information evaluation link information. It is the conceptual diagram which showed two types of links with respect to a content from a user. FIG. 9 is a diagram illustrating a flow for explaining the processing of the score calculation means. FIG. 10A is a data structure diagram managed by the information reputation score storage means. FIG. 10B is a data structure diagram managed by the information provision score storage means and the information evaluation score storage means. FIG. 11 is a diagram showing a flow for explaining details of processing of the score output means. FIG. 12A is a diagram showing the iterative calculation procedure of the score calculation means in matrix expression. FIG. 12B is a diagram showing the iterative calculation procedure of the score calculation means in element expression. FIG. 13 is a diagram illustrating a flow for explaining processing when r, a, and h are calculated by the algorithm of FIG. FIG. 14 is a diagram illustrating a configuration of a content search system according to the second embodiment. FIG. 15 is a diagram illustrating a detailed configuration of the content information collection unit of FIG. FIG. 16 is a diagram illustrating a detailed configuration of the information evaluation apparatus in FIG. FIG. 17 is a diagram illustrating a detailed configuration of the information acquisition unit in FIG. FIG. 18 is a diagram illustrating a detailed configuration of the score calculation means in FIG. FIG. 19 is a diagram illustrating a detailed configuration of the content information search means in FIG. FIG. 20 is a diagram illustrating a flow for explaining the score calculation processing according to the second embodiment. FIG. 21 is a diagram illustrating a flow for explaining details of the calculation processing of the information provision link information and the information evaluation link information in the second embodiment. FIG. 22 is a conceptual diagram illustrating blog information. FIG. 23 is a conceptual diagram illustrating a content information list. FIG. 24A is a diagram illustrating a data configuration of information providing link information. FIG. 24B shows the data structure of the information evaluation link information. FIG. 25 is a diagram illustrating a flow for explaining details of the content information search processing. FIG. 26 is a diagram illustrating a configuration of a modified example of the score calculation means. FIG. 27 is a diagram illustrating a flow for explaining a modification of the processing of the score calculation unit. FIG. 28 is an example of a flow for explaining the processing of the information reputation score estimating means. FIG. 29 is a diagram illustrating a configuration of a modified example of the content information search means. FIG. 30 is a diagram illustrating a flow for explaining a modification of the processing of the content information search means.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Information evaluation system 100 Information evaluation apparatus 200-1 to m, 500-1 to p User terminal apparatus 300 Network 600-1 to q Content server apparatus 1000 Content search apparatus

Claims (7)

In an information evaluation device that calculates an information reputation score that quantifies the reputation of content that is electronic information provided by a user,
A user ID of the content provider, and the information providing link storage means for storing a content ID, the set information providing link information having the an identifier of content provided by the content provider,
A user ID of the content evaluator, a content ID is an identifier of the content evaluated by the content evaluator, and evaluation information is numerical information, reviews the content evaluator performed with respect to the content, a set Information evaluation link storage means for storing information evaluation link information having
For each set of a Subscriber ID and the content ID, a numerical value indicating whether or not the information providing link information having a set of the said user ID and the content ID is stored in the information providing link storage means, the use First adjacency matrix generation means for generating and outputting an information provision link adjacency matrix C as a matrix element corresponding to a set of a person ID and the content ID ;
For each set of a Subscriber ID and the content ID, the value of the evaluation information information evaluation link information having a set of the said user ID and the content ID has a user ID and the content ID of the content evaluator Second adjacency matrix generation means for generating and outputting an information evaluation link adjacency matrix E as matrix elements corresponding to the set of
When α is a real number between 0 and 1 and β ^T is a transposed matrix of β,
S = αC ^TC + (1-α) E ^T E
Eigenvector calculation means for calculating a score transition matrix S by calculating the eigenvector r of the score transition matrix S , and outputting an element of the eigenvector r as the information reputation score;
Writing means for storing the information reputation score in an information reputation score storage means ;
An information evaluation apparatus characterized by that. The information evaluation device according to claim 1 ,
The eigenvector calculating means is
Each element of the information provision link adjacency matrix C is normalized by a first function determined for each set of elements corresponding to the same user ID of the content provider, and the normalized information provision link adjacency matrix is output. First normalizing means to
Each element of the information evaluation link adjacency matrix E is normalized by a second function determined for each set of elements corresponding to the same user ID of the content evaluator, and the normalized information evaluation link adjacency matrix is output. A second normalization means to:
The score transition matrix S is calculated using the normalized information provision link adjacency matrix C and the normalized information evaluation link adjacency matrix E, and the eigenvector r of the score transition matrix S is calculated. calculating means for outputting the elements of the eigenvector r as the information reputation score, Ri means der having,
The first function is a function depending on the number of the content IDs paired with the user ID in the information provision link storage unit,
The second function is a function depending on the number of the content IDs paired with the user ID in the information evaluation link storage unit.
An information evaluation apparatus characterized by that. A content search device that provides information about content stored in a content server device,
Includes the URI of the site top of the blog site, the URI of the blog article of the blog site linked to the site top of the blog site, the URI of the blog article, and the other blog articles referenced by the blog article Content information receiving means for receiving content;
The URI of the site top of the blog site, the URI of the blog article of the blog site linked to the site top of the blog site, the keyword extracted from the blog article, and other blogs referenced by the blog article Content information storage means for storing a content information list that is a list of content information associated with the URI of an article ;
From the content information list stored in the content information storage unit, the blog site top URI sites extracted by the user ID of the content provider, of the blog site that links to a site top of the blog site the URI of posts extracted by the content ID of the content provided by the content provider generates and outputs the information providing link information having a set of the user ID and the content ID of the content provider An information providing link generating means;
From the content information list stored in the content information storage unit, the blog site top URI sites extracted by the user ID of the content evaluator, of the blog site that links to a site top of the blog site the URI of the other blog post blog post is referring extracted by the content ID of the content that has been evaluated by the content evaluator, and the user ID of the content evaluator, it has been evaluated by the content evaluator and the content ID of the content, and the evaluation information is a numerical value information which is determined by whether or not the blog site other blog post URI that blog post is a reference to link to the site top of the blog site exists, set information evaluation link students to generate and output the information evaluation link information with a And means,
Information providing link storage means for storing the information providing link information;
Information evaluation link storage means for storing the information evaluation link information;
For each set of a Subscriber ID and the content ID, a numerical value indicating whether or not the information providing link information having a set of the said user ID and the content ID is stored in the information providing link storage means, the contents First adjacency matrix generation means for generating and outputting an information provision link adjacency matrix C as a matrix element corresponding to a set of a provider user ID and the content ID;
For each set of a Subscriber ID and the content ID, the value of the evaluation information information evaluation link information having a set of the said user ID and the content ID has a user ID and the content ID of the content evaluator Second adjacency matrix generation means for generating and outputting an information evaluation link adjacency matrix E as matrix elements corresponding to the set of
When α is a real number between 0 and 1 and β ^T is a transposed matrix of β,
S = αC ^TC + (1-α) E ^T E
Eigenvector calculation means for calculating the score transition matrix S by calculating the eigenvector r of the score transition matrix S, and outputting an element of the eigenvector r as an information reputation score corresponding to each content information;
The information reputation score, it in association with the corresponding the content information, and writing means for storing in the content information storing means,
Search condition receiving means for receiving a search term transmitted from the user terminal device;
Search means for searching the content information list using the search terms, and extracting the content information including the keywords that match the search terms from the content information storage means together with the information reputation score associated therewith; Comprising
A content search apparatus characterized by that. In the information evaluation method of the information evaluation apparatus that calculates the information reputation score that quantifies the reputation of the content that is electronic information provided by the user,
The information providing link storage means of the information evaluation device, storing the user ID of the content provider, a content ID is an identifier of content provided by the content provider, the set information providing link information having the When,
The information evaluation link storage means of the data evaluator, a user ID of the content evaluator, a content ID is an identifier of the content evaluated by the content evaluator, the content evaluator performed against the content Storing information evaluation link information having a set of evaluation information which is numerical information of evaluation; and
In the first adjacency matrix generating means of the data evaluator, for each set of a Subscriber ID and the content ID, the information providing link information having a set of the said user ID and content ID the information providing link storage means Generating and outputting an information providing link adjacency matrix C having a numerical value indicating whether or not stored as a matrix element corresponding to a set of the content provider's user ID and the content ID; and
In the second adjacency matrix generating means of the data evaluator, for each set of a Subscriber ID and the content ID, the value of the evaluation information information evaluation link information having a set of the said user ID and the content ID has, Generating and outputting an information evaluation link adjacency matrix E as a matrix element corresponding to a set of the user ID of the content evaluator and the content ID;
In the eigenvector calculation means of the information evaluation apparatus, when α is a real number not less than 0 and not more than 1 and β ^T is a transposed matrix of β,
S = αC ^TC + (1-α) E ^T E
Calculating the score transition matrix S by calculation, calculating an eigenvector r of the score transition matrix S, and outputting an element of the eigenvector r as the information reputation score;
Information reputation score storage unit of the information evaluation device, and a step of storing the information reputation score,
An information evaluation method characterized by that. A content search method of a content search device that provides information about content stored in a content server device,
The content information receiving means of the content search device refers to the URI of the site top of the blog site, the URI of the blog article of the blog site linked to the site top of the blog site, the blog article, and the blog article. Receiving content including URIs of other blog articles that are present ;
In the content information storage means of the content search device, the URI of the site top of the blog site, the URI of the blog article of the blog site linked to the site top of the blog site, the keyword extracted from the blog article, Storing a content information list which is a list of content information associated with URIs of other blog articles referred to by the blog article ;
The information providing link generation unit of the content retrieval device, said from the content information list stored in the content information storing means, and extracted with user ID of the content provider site top URI of the blog site, the the URI of the blog site of the blog post that links to the site the top of the blog site extracted by the content ID of the content provided by the content provider, the combination of the user ID and the content ID of the content provider Generating and outputting information providing link information having :
Storing the information provision link information in an information provision link storage means of the content search device ;
In the above information evaluation link generating means of the content retrieval device, from the content information list stored in the content information storing means, and extracted with user ID of the content evaluator sites top URI of the blog site, the the URI of the other blog post blog post of the blog sites that link to the site the top of the blog site is referring extracted by the content ID of the content that has been evaluated by the content evaluator, use of the content evaluator A user ID, a content ID of the content evaluated by the content evaluator, and a URI of another blog article referenced by the blog article of the blog site linked to the site top of the blog site. and the evaluation information is a numerical value information which is determined by a set of Yes And a step for generating and outputting a that information evaluation link information,
Storing the information evaluation link information in an information evaluation link storage means of the content search device ;
In the first adjacency matrix generation unit of the content retrieval device, for each set of a Subscriber ID and the content ID, the information providing link information having a set of the said user ID and content ID the information providing link storage means Generating and outputting an information providing link adjacency matrix C having a numerical value indicating whether or not stored as a matrix element corresponding to a set of the content provider's user ID and the content ID; and
In the second adjacency matrix generation unit of the content retrieval device, for each set of a Subscriber ID and the content ID, the value of the evaluation information information evaluation link information having a set of the said user ID and the content ID has, Generating and outputting an information evaluation link adjacency matrix E as a matrix element corresponding to a set of the user ID of the content evaluator and the content ID;
In eigenvector computing means of the content retrieval device, in the case where the α is 0 or 1 less real, and beta ^T was transposed matrix of beta,
S = αC ^TC + (1-α) E ^T E
Calculating the score transition matrix S by the calculation of, calculating an eigenvector r of the score transition matrix S, and outputting an element of the eigenvector r as an information reputation score corresponding to each content information ;
In the writing means of the content retrieval device; the information reputation score, it in association with the corresponding the content information to be stored in the content information storing means,
In the search condition receiving means of the content search device, receiving a search term transmitted from the user terminal device;
In search means of the content retrieval device, the content of the content information including the keyword searching the content information list by using the search word, matches the search term, the information reputation score together with their associated It has a step of extracting and outputting the information storage means, and
A content search method characterized by the above. A program for causing a computer to function as the information evaluation device according to claim 1 or 2 or the content search device according to claim 3 .   A computer-readable recording medium storing the program according to claim 6.

Images