JP2009157764A - Community system, activity recording method for community system, and program for activity recording of community system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To handle indices of a plurality of activities, in a unified manner, such as writing of a blog, reading of a news, and reading of a massage in an online community such as an SNS. <P>SOLUTION: The actions of a single or a plurality of users online are accumulated together with text information used in accordance with the actions, to generate data used for extracting and matching taste. Thus, analyzing of action/taste of a user which has been performed based on one-dimensional information so far can be performed from two observation points, or action and taste, or from the observation point of integration of them, allowing analysis of action/taste with more information quantity. The data format by which the action of user along with text information are accumulated is called as action matrix for convenience. The action matrix is such a matrix as consists of action of a user and a keyword extracted from the text used with the action. In each element, a value based on the appearance frequency of keywords accumulated upto a certain time point is inputted. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  In the present invention, a plurality of users use their own client computers to log in to a server system and communicate with each other on the server via a network, such as a social networking service (SNS). , On online community systems.

  As Web 2.0 technology permeates, general users write blogs (diaries) in online communities like CGM (customer generated media) or social networking services (SNS), news We are actively engaged in various activities such as reading product information and searching for product information.

  Therefore, it has been studied to extract the interests and interests of individual users from such activity logs and to utilize them for corporate activities such as marketing.

  For example, Japanese Patent Application Laid-Open No. 2007-241753 holds, in a server, a model ontology that is a model of a personal ontology in which personal interest information is conceptually hierarchized. The direct descendant class and instance up to the class or instance are extracted as a personal ontology, and the unexpected information extraction means is included in one personal ontology between personal ontologies whose degree of approximation is a predetermined value or more, but the other personal ontology. A class or instance that is not included in is extracted as unexpected information, and the unexpected information presenting means discloses a system that presents the unexpected information extracted by the unexpected information extracting means to a user who does not have the unexpected information.

  The specification of Japanese Patent Application No. 2007-110559 related to the present applicant is to prevent the appearance of advertisements such as web pages, blogs, etc. Disclose a technology that systematically assigns one of the interest labels of “perpetrator”, “victim”, and “beneficiary” to applicable items. Then, the advertisement selection mechanism is controlled so that the appearance of advertisements associated with such interest labels is suppressed. The ad selection mechanism used for this purpose can be calculated by calculating the affinity or relevance between the article and the advertisement as a numerical value, and therefore, suppression of the advertisement is achieved by reducing the numerical value of the affinity. Is done. Preferably, the degree of suppression by the interest label is gradually attenuated with time. This disclosed technique can be said to be one attempt to systematically model the psychological state of a user accessing a web page.

  In particular, the technology disclosed in Japanese Patent Application Laid-Open No. 2007-241753 extracts interest from text information written by the user, and the technology described in Japanese Patent Application No. 2007-110559 is based on text information read by the user. Extracting interests. As described above, there are various types of actions or activities that a user can take on the online community, and a technique for uniformly handling such a plurality of actions is desired.

  JP-A-2005-327105 relates to a community analysis apparatus for finding an appropriate community. The community analyzer includes an acquisition unit that acquires information including at least a sender identification symbol for identifying a sender of a transmission mail and a receiver identification symbol for identifying a recipient of the transmission mail from a predetermined mail server; and an acquisition unit The common interest analysis unit that calculates the common interest level that represents the degree of interest common to the sender and receiver from the information acquired by the user, and the community formed by the e-mail user from the common interest level obtained by the common interest analysis unit Having an analysis unit. The analysis unit analyzes the community formed by the e-mail user by calculating the degree of association between the communities using the interest matrix.

Japanese Patent Laid-Open No. 2005-327105 suggests a useful technique in that the degree of association between communities is calculated using a matrix. However, it is related to a single operation of analysis from a sent mail, and still does not teach a technique for handling a plurality of actions in a unified manner.
JP 2007-241753 A Japanese Patent Application No. 2007-110559 JP 2005-327105 A

  An object of the present invention is to provide a technology that makes it possible to uniformly handle a plurality of behavior indicators such as blog writing, reading news, reading messages, etc. in an online community such as SNS. .

  Another object of the present invention is to calculate a plurality of behavioral indicators calculated between individuals, groups or communities, and between individuals and groups or communities in an online community. It is to provide a technique that makes it possible to give useful results.

  The present invention accumulates the actions of a single user or a plurality of users on the online together with text information used in accordance with the actions, and uses it for preference extraction and matching. This makes it possible to analyze the user's behavior and preferences that have been projected to one-dimensional information such as behavior or text, from the two viewpoints of behavior and preference, or from the viewpoint of integrating them. Enables a lot of behavior and preference analysis. The actions mentioned here are not limited to this, but write / read blogs, write / read messages, read news, write / read on bulletin boards, purchase products, read product descriptions, other users See their profile, etc.

In the present invention, in order to accumulate user actions together with text information, the data format held is referred to as an action matrix for convenience. The behavior matrix data is preferably held on a hard disk drive of an online community server or the like. The behavior matrix is a matrix composed of the user's behavior and keywords extracted from the text used together with the behavior, and each element has a value based on the appearance frequency of the keyword accumulated up to a certain time. enter. As illustrated above, there are a plurality of behaviors that are subject to keyword extraction, and behavior A = {a ₁ , a ₂ , a ₃ . The behavior matrix is a storage of keywords W = {w ₁ , w ₂ , w ₃ , ...} extracted from text d associated with the behavior a performed by the user in a form corresponding to the behavior a. . When the keyword w _i appears once in the text d accompanied by a certain action a _j , 1 (or a value obtained by multiplying this by a weight) is entered in the element ij of the action matrix.

The behavior matrix preferably also reflects past accumulation. According to one aspect of the present invention, an action matrix at a certain time T _i is weighted with an action matrix (referred to as a temporary action matrix) between ΔT = T _{i + 1} -T _{i at} the next time T _{i + 1} . And updated in a mixed form.

  One behavior matrix is defined for each user. Further, when a plurality of users are set as one explicit community, the behavior is accumulated and defined as a community behavior matrix.

  By using the action matrix created in this way, it is possible to perform analysis with text action information. Examples include specifying two users and comparing their behavior matrices to find duplicate events from both user behavior, text, and both perspectives.

  The present invention is realized by two systems. The first is a system that generates a user's temporary action matrix from text associated with the user's action and integrates it with the action matrix accumulated up to that point. This is a language processing unit that extracts keywords and preference information from text, and a behavior matrix generation unit that generates a temporary behavior matrix from these texts and behaviors and integrates the behavior matrix stored in the behavior matrix storage unit described below. Consists of. The second is a system that accumulates these behavior matrices for each user or static community and uses them for analysis and search. This system provides an external application program with an interface (API) for various analyzes on the behavior matrix.

  As operations necessary for analyzing the behavior matrix, a sum / difference / product and a row / column degeneration specific to the behavior matrix are defined. The sum of two behavior matrices is to create a pseudo user (or community) behavior matrix that integrates the two users (or two communities) from the users (or communities) that have the respective behavior matrices. Comparable to In addition, by the product of the behavior matrix, the behavior / keywords of the two behavior matrices and the rate at which the behavior and the keyword are correlated can be obtained. Further, row reduction and column reduction mean weighted keyword vectors or weighted action vectors of the action matrix, respectively.

  According to the present invention, in an online community such as SNS, a plurality of actions and an action matrix across a plurality of keywords are calculated, and the result is stored in a storage device of a computer. By performing various computer analysis on the behavior matrix, such as sum, difference, product, and degeneracy, it is possible for the unified users and communities hidden in multiple behaviors that could not be known in the past. It can highlight trends, and the results can be used effectively for marketing, security, and other purposes, thus increasing the value of the online community itself.

  Embodiments of the present invention will be described below with reference to the drawings. Unless otherwise noted, the same reference numerals refer to the same objects throughout the drawings. Also, it is to be understood that what is described below is one embodiment of the present invention, and that the present invention is not intended to be limited to the contents described in this example.

  1, a plurality of client computers 106a, 106b,... 106z are connected to the online community server 102 via the Internet 104. In the system of FIG. 1, a user of a client computer logs in to the online community server 102 through a line of the Internet 104 through a Web browser. Specifically, a predetermined URL is typed into a Web browser and a predetermined page is displayed. The log-in may be performed using a predetermined client application program instead of the Web browser.

  When logging in, the user of the client computer uses a given user ID and a password associated therewith. Once a client computer user logs in, he or she writes a diary in the online community, browses the diary of another person who is allowed access, writes a comment, sees the news, and is a good friend Activities such as creating groups with each other, chatting, and searching for hobby communities.

  Next, referring to FIG. 2, a hardware block diagram of the client computer indicated by reference numerals 106a, 106b... 106z in FIG. In FIG. 2, the client computer has a main memory 206, a CPU 204, and an IDE controller 208, which are connected to the bus 202. In addition, a display controller 214, a communication interface 218, a USB interface 220, an audio interface 222, and a keyboard / mouse controller 228 are connected to the bus 202. A hard disk drive (HDD) 210 and a DVD drive 212 are connected to the IDE controller 208. The DVD drive 212 is used for introducing a program from a CD-ROM or DVD as necessary. A display device 216 having an LCD screen is preferably connected to the display controller 214. The display device 216 displays an online community screen through a Web browser.

    Devices such as a dedicated controller and an acceleration sensor / device can be connected to the USB interface 220 as necessary. These devices can be used to improve operability within the online community.

  A speaker 224 and a microphone 226 are connected to the audio interface 222. By providing the client computer with a speech synthesis function, the chat content with the other party can be voiced and output from the speaker 224 in the online community. Further, by providing a voice recognition function, the contents spoken by the user to the microphone 226 in the online community are converted into text by the voice recognition function and sent to the other party as chat contents. can do.

  A keyboard 230 and a mouse 232 are connected to the keyboard / mouse controller 228. The keyboard 230 is typically used in an online community to write a chat message or describe community content to be searched. The mouse 232 is used in the online community to click on a link to read news, select and execute an action from a menu, or select a diary to read.

  The CPU 204 may be, for example, an arbitrary one based on a 32-bit architecture or a 64-bit architecture, and Intel Pentium (trademark of Intel Corporation) 4, AMD Athlon (trademark), or the like can be used.

  The hard disk drive 210 stores at least an operating system and a web browser (not shown) that runs on the operating system, and the operating system is loaded into the main memory 206 when the system starts up. . As the operating system, Windows XP (a trademark of Microsoft Corporation), Windows Vista (a trademark of Microsoft Corporation), Linux (a trademark of Linus Torvalds), or the like can be used.

  The communication interface 218 uses the TCP / IP communication function provided by the operating system to communicate with the online community server 102 using the Ethernet (trademark) protocol or the like.

  FIG. 3 is a schematic block diagram of a hardware configuration on the online community provider side. As shown in FIG. 3, client computers 106 a, 106 b... 106 z are connected to the communication interface 302 of the online community server 102 via the Internet 104. The communication interface 302 is further connected to a bus 304, and a CPU 306, a main memory (RAM) 308, and a hard disk drive (HDD) 310 are connected to the bus 304.

  Although not shown, a keyboard, a mouse, and a display may be further connected to the online community server 102, and the entire online community server 102 may be managed and maintained by these.

  The hard disk drive 310 of the online community server 102 stores a correspondence table of user IDs and passwords for login management of operating systems, client computers, client computers 106a, 106b,... 106z. . The hard disk drive 310 further stores software such as Apache for causing the online community server 102 to function as a Web server, and is loaded into the main memory 308 when the online community server 102 is started up. To do. As a result, the client computers 106a, 106b,... 106z can access the online community server 102 using the TCP / IP protocol.

  In addition, the hard disk drive 310 of the online community server 102 further includes information on messages, diaries or blogs, bulletin boards, etc. of each user of the online community service, and information on the online community service. An HTML file and a multimedia format such as a graphic image, a moving image file, and a music file are stored.

  The user can write in the diary or the blog and the bulletin board, and other users can read the blog and the bulletin board and make comments according to the authorized authority.

  As will be described in detail later, the hard disk drive 310 stores a module for calculating an action matrix according to the present invention, a module for performing calculations to extract various information from the calculated action matrix, and the like. ing.

  The configuration of blogs, bulletin boards, etc. and user access control for them can be realized with tools of well-known programming languages such as Perl, Ruby, PHP, Servlet, JSP. Alternatively, C, C ++, C #, Java (trademark of Sun Microsystems), etc. can be used.

  Furthermore, it is possible to embed JavaScript (trademark) in the HTML file as appropriate and configure the system to cooperate with Perl, Ruby, PHP, etc.

  Content such as blogs, bulletin boards, news, etc. can be stored in a content management database (CMDB) and managed centrally.

  As the online community server 102, models such as IBM (trademark of International Business Machines Corporation) System X, System i, System p, which can be purchased from International Business Machines Corporation. You can use any server. In this case, usable operating systems include AIX (trademark of International Business Machines Corporation), UNIX (trademark of The Open Group), Linux (trademark), Windows (trademark) 2003 Server, and the like.

Next, the basic concept of the behavior matrix according to the present invention will be described with reference to FIG.
In FIG. 4, the user 402, in this online community service, shows “Write Message” 404a, “Write to Bulletin Board” 404b, “Write Blog” 404c... It is assumed that actions (activities) such as 404i, “read bulletin board” 404j, “read blog” 404k, “read news” 404l,. The types of actions that the user can take in the online community service are determined in advance by the functions provided by the provider of the online community server 102.

  Therefore, for each of the above-described activities, the user 402 reads / writes a text 406a read / written as a message, a text 406b read / written on a bulletin board, a text 406c read / written as a blog, and a text 406d read as news for a certain period of time. Is temporarily stored as text information 406 in the hard disk drive 310 (FIG. 3). At this time, it is called collectively as text 406a read and written as a message, but it should be noted that the read text and the written text are stored separately and identifiable. The same applies to bulletin boards and blogs. Therefore, the frequency of keywords and specific expressions appearing in the hard disk drive 310 is stored in the hard disk drive 310 by using a syntax analysis technique known from JP-A-2001-84250, JP-A-2002-251402, and JP-A-2004-246440. The behavior matrix 408 for the user 402 is obtained by calculating with the analysis calculation module.

  To supplement the explanation, the behavior matrix 408 of the present invention uses rows as keywords and columns as behaviors. Therefore, in general, the matrix is quite long with several thousand rows and several tens of columns. The keywords appearing on the line cover all the keywords extracted by parsing from all the text stored as the text information 406.

  Then, the value of the component of a certain action column in a certain keyword row indicates the appearance frequency of the keyword included in the text related to the action.

  The values of the action matrix created in this way are stored in the hard disk drive 310 as individual files for each user. At this time, the format to be stored can be identified by programming tools such as C, C ++, C #, Java (trademark), Perl, Ruby, and PHP as row and column values such as CSV, HTML, and XML. It can be in any format.

  FIG. 5 is a diagram showing an action matrix DB 502 that stores action matrix data created in the format described in FIG. Actually, the behavior matrix DB 502 is preferably in a CMDB format and is stored in the hard disk drive 310. As described above, behavior matrix data 504a, 504b, 504c,... For each user is stored in the behavior matrix DB 502.

  The behavior matrix DB 502 further stores user community behavior matrix data 506a, 506b, 506c,. The user community behavior matrix data is calculated by performing the operation defined in the embodiment of the present invention on the behavior matrix data of the users belonging to the user community. This calculation algorithm will be described in detail later.

  In this way, when the action matrix data 504a, 504b, 504c,... For each single user and the action matrix data 506a, 506b, 506c,. Using the operation of the API 508 defined in the embodiment of the present invention, extraction of an interest of a user indicated by reference numeral 510, extraction of a user having an interest similar to that of a user indicated by reference numeral 512, reference numeral 514 Extraction of a community having an interest similar to a certain user, and extraction of a community having an interest similar to a certain community indicated by reference numeral 516 can be performed. It should be understood that the processing possible in this embodiment is not limited to that illustrated in FIG. 5, but is merely an example. Processing for executing these or processing other than those described here will be described in detail later.

  FIG. 6 is a diagram for explaining the action matrix generation processing from another viewpoint. In FIG. 6, text data 602 is a text file of content written or accessed by the user, and is the same as the text information 406 shown in FIG. The processing module of the language processing unit 604 is stored in the hard disk drive 310 and is called and executed by the CPU 306 in the main memory 308 as necessary.

  Basically, all language-dependent processing is performed by the language processing unit 604. Specifically, morphological analysis, syntax analysis, keyword extraction, proper noun extraction, reputation expression extraction, and the like are performed. The language processing unit 604 refers to a dictionary or an external knowledge database as necessary. For this reason, the hard disk drive 310 stores a dictionary (not shown) such as a synonym dictionary. The external knowledge database may be stored in the hard disk drive 310 in the same manner as the dictionary, or may be referred to via a network as a Web service.

  Since this process is well known in Japanese Patent Laid-Open No. 2001-84250, Japanese Patent Laid-Open No. 2002-251402, Japanese Patent Laid-Open No. 2004-246440, and the like, detailed description thereof is omitted.

  The behavior matrix generation unit 606 writes the keywords extracted by the language processing unit 604, writes a message, writes it on a bulletin board, writes a blog, reads a message, reads a bulletin board, reads a blog, reads news, etc. ) Organize and weight each time to generate an action matrix.

  The behavior matrix generation unit 606 accumulates the behavior matrix data thus generated in the behavior matrix DB 502. Data stored in the behavior matrix DB 502 is as shown in FIG.

  The API 508 calculates various applications 612a, 612b, 612c,... Using the behavior matrix data stored in the behavior matrix DB 502.

  FIG. 7 is a schematic diagram for explaining the processing of the behavior matrix generation unit 606 in more detail. In FIG. 7, the text information block 406 is the same as that shown in FIG. The text information block 406 includes text 702a, 702b of a blog article written by a specific user, blog articles 704a, 704b,... Read by the specific user, and messages 706a, 706b,. A message 708a read by the specific user is stored.

  The behavior matrix generation unit 606 extracts a keyword column 702x and a column 702y of the number of times these keywords are used from the texts 702a and 702b of the written blog article.

  Further, the behavior matrix generation unit 606 extracts a keyword column 704x and a column 704y of the number of times the keywords are used from the texts 704a, 704b,.

  Further, the behavior matrix generation unit 606 extracts a keyword column 706x and a column 706y of the number of times these keywords are used from the texts 706a, 706b,.

  Further, the behavior matrix generation unit 606 extracts a keyword column 708x and a column 708y of the number of times these keywords are used from the text 708a of the read message.

  The behavior matrix generation unit 606 obtains the behavior matrix 710 from the keyword columns 702x, 704x, 706x, and 708x for each behavior and the columns 702y, 704y, 706y, and 708y of the number of times these keywords are used. Generate.

  That is, in principle, the column of the number of times the keyword is used for the corresponding action is arranged as a column vector corresponding to the action. A matrix is formed by arranging such column vectors. However, in this embodiment, preferably the number of times a keyword is used is normalized by dividing by the number of documents viewed or written in relation. As a result, the “Read Blog” column and the “Write Message” column of the behavior matrix 710 contain decimal numbers.

  In addition, the behavior matrix shown in FIG. 7 is schematically shown. In practice, the number of lines may reach several thousand lines depending on the keywords used in the text. Note that the quantity number is larger than shown because it depends on the number of actions to be defined.

Next, processing for creating a behavior matrix will be described using the flowchart of FIG. As a premise of this explanation, we define action vector a ≡ {a ₁ , a ₂ , ... a _m }. Where a ₁ , a ₂ , ... a _m are the behaviors assumed in this community system, for example, writing / reading a blog, writing / reading a message, reading news, writing on a bulletin board This includes actions such as performing / reading, purchasing a product, reading a product description, and viewing other users' profiles.

In the flowchart of FIG. 8, between the steps 802 and 816, the processing described below is performed for each such action a _i (i = 1 ... m). That is, in step 804, the document set D _i generated by the action a _i at a certain time is extracted. For example, in FIG. 7, the document set referred to here is the written blog articles 702 a and 702 b in the text information 406. This is the document set for the action of “writing a blog”. The document set includes one or more documents. In the text information 406 of FIG. 7, the documents included in the document set corresponding to “write blog” are the document 702a and the document 702b. For each document d _ij in such a document set _{D i (j = 1 .. n} ), perform the following processing.

That is, in step 808, the document d _ij, morphological analysis, syntax analysis is applied, and keywords in the document d _ij, their occurrence frequency is extracted.

In step 810, the keywords thus extracted and their appearance frequencies are stored in W _i and X _i , respectively. W _i and X _i are actually vectors. In step 810, normalization may be performed by dividing the numerical value in X _{i by} the number of documents in the document set D _i .

Repeated completion of the document _{d ij (j = 1 .. n} ) in step 812, in step 814, W _i and X _i corresponding to one of the actions a _i is completed.

Thus, in step 816, W _i and X _i are obtained for all actions a _i . In step 818, the process of arranging the label of keywords included in all W _i vertically is performed first, then, in step 820, for each action a _i, the row position corresponding to the keyword in that W _i, A process of entering the corresponding frequency in X _i is performed. Thus, a temporary action matrix is generated. Here, the term “temporary” is used because it is used for calculating the time transition of the behavior matrix described later.

  FIG. 9 is a diagram for explaining a specific method for accessing a document in steps 804 and 806 of FIG. The community server system shown in FIG. 3 records an access log 902 in the hard disk drive 310. The access log 902 includes a date 902a, a time 902b, a process 902c, a user ID 902d, and a document ID 902e.

  For example, when a user having a user ID of 014623 writes a blog message on the screen 904 and clicks the save button 906, this is recorded in the access log 902 as a record 902f indicating a process called PostBlog. Then, the content of the written blog is determined from the document ID of the record 902f, and the text is stored in the hard disk drive 310 as a file 908. Similarly, a text file 910 associated with the message is also stored on the hard disk drive 310.

  The text files 908 and 910 created in this way are used to create a temporary action matrix as the text information 406 shown in FIGS.

Next, the update by the time transition of an action matrix in a present Example is demonstrated.
In this embodiment, the time transition is calculated as follows.
_{A t + Δt = W k A} t W a + T Δt
Here, A _t is the action matrix at time t, W _k is a weighting factor for each keyword, W _a is a forgetting coefficient for each _action, T Δt is a temporary action at time t + Δt Matrix. The method for generating the temporary behavior matrix is as described with reference to FIGS.

Specific examples of W _k and W _a are shown in FIG. As shown, W _k is an N × N square matrix, where N is the number of keyword labels. In the simplest case, only the diagonal component may be a non-zero positive number and the other components may be zero. For example, the component corresponding to the keyword of the proper noun is set to 1.0 or a value close thereto, and the component corresponding to the common noun or the ordinary keyword is selected to have a value considerably smaller than 1.0. By doing so, the numerical value corresponding to the keyword of the proper noun is not attenuated so much, and the numerical value corresponding to the keyword of the common noun is rapidly attenuated.

W _a is an M × M square matrix, where M is the number of action types. In the simplest case, only the diagonal component may be a non-zero positive number and the other components may be zero. For example, the component corresponding to a relatively strong impression such as writing on a blog is set to 1.0 or a value close thereto, and the component corresponding to a relatively weak impression such as reading news is a value considerably smaller than 1.0. Select.

Calculation of W _{k A t} W _a can but a multiplication of normal matrix
Addition of _{W k A t W a + T} Δt is, not necessarily can use the addition of a normal matrix. This is because, in the generation of T _{t +} Δt, by which you find the new keywords that ever met, in the A _t and T _Δt, because not always the line label of keywords match.

Therefore, in general, as _{A t + T = A t +} 1, the sum of the extended matrices, defined as follows.
Matrix _{A t ≡ {a n, m} } and. a _{n, m} is the n, m component of the matrix.
Further, the keyword / label is set as w = {w ₁ , w ₂ ,..., W _n }.
Similarly, T ≡ {t _{n ', m} }
Its keyword label w = {w ' ₁ , w' ₂ , ..., w '_n' }
A _{t + 1} ≡ {a n`` _{, m} }
Its keyword label w = {w '' ₁ , w '' ₂ , ..., w '' _{n ''} }

The definition is made in this way, and the cases are divided as follows.
· A In the _t, in the case of a keyword is not in the T w _i (word was previously used did not this time is used), w '' _i and a _'ij of the decision:
w '' _i = w _i (1 <= i <= n)
a ' _ij = a _ij (1 <= j <= m)
· A keyword _t to also be in T w _i in the case of (the word was previously used did not this time is used), w '' _i and a _'ij of the decision:
w '' _i = w _i (1 <= i <= n)
a ' _ij = a _ij + t _kj
(w _i = w ' _k , 1 <= j <= m, i <= k <= n')
· A not in _t, '(but did not use the past, this time used the word) _k in the case of a, w' keyword w in the T _'i and a' _ij of the decision:
w '' _i = w ' _k (n + 1 <= i <= n'')
a ' _ij = t _kj (1 <= j <= m)

With this extended sum operation,
Based on the formula of _{A t + Δt = W k A} t W a + T Δt, in accordance with the transition of the action matrix A _t the time, to update. Such a calculation is automatically calculated by a computer program prepared in advance. Those skilled in the art will appreciate that such computer programs can be created in any programming language such as C ++, C #, Java, and the like. All operations defined below are similarly created in a programming language, kept executable and called in a timely manner.

  The update frequency of the behavior matrix is typically once a day, but can be appropriately selected according to the frequency of community activities, such as twice a day or once a week.

  By using the behavior matrix created and updated with time, it can be applied to various meaningful applications. The application will be described below.

この絶対値自体を単独で使用することはあまりなく、その主要な用途は、他の計算における、値の正規化である。 [Absolute value of behavior matrix]
The absolute value of the behavior matrix A is defined by the following formula. a _ij is an ij component of the behavior matrix A.

The absolute value itself is rarely used alone, and its primary use is for value normalization in other calculations.

このような行動マトリクスの和を、あるコミュニティ全体に属するユーザの行動マトリクス全体に適用することにより、コミュニティの行動マトリクスが得られる。 [Sum of two action matrices]
The sum of the two behavior matrices A _k and A _l is defined by the following equation. The absolute values of the matrix appearing here are those defined above. Before performing the following calculation, the process of aligning keywords and labels is performed. That is, an operation that takes the union of the keyword labels of A _k and A _l is performed, and the rows of _Ak and A _l are expanded based on the resulting keyword labels, that is, the original keyword rows After all 0s are entered, the following formula is executed.

By applying the sum of such behavior matrices to the whole behavior matrix of users belonging to a certain community, a community behavior matrix can be obtained.

  FIG. 11 shows an example of the sum of two behavior matrices A and B. As can be seen from this example, the sum of the behavior matrices A and B is obtained by appropriately normalizing the values.

このような行動マトリクスの差は、ユーザ間、ユーザとコミュニティ間、あるいはコミュニティ間の活動の差を求めるために使用することができる。なお、ここでの右辺の一番外側の絶対値は、単に行列の要素を全て正の値にする、という意味で、上記で定義した、マトリクスの絶対値でないことに留意されたい。 [Difference between two behavior matrices]
The difference between the two behavior matrices A _k and A _l is defined by the following equation. The absolute values of the matrix appearing here are those defined above. Before performing the following calculation, the process of aligning keywords and labels is performed. That is, an operation that takes the union of the keyword labels of A _k and A _l is performed, and the rows of _Ak and A _l are expanded based on the resulting keyword labels, that is, the original keyword rows After all 0s are entered, the following formula is executed.

Such behavior matrix differences can be used to determine activity differences between users, between users and communities, or between communities. It should be noted that the absolute value on the outermost side of the right side here is not the absolute value of the matrix defined above in the sense that all the elements of the matrix are simply positive values.

  FIG. 12 shows an example of the difference between the two behavior matrices A and B. As can be seen from this example, the sum of the behavior matrices A and B is a value that is appropriately normalized and a difference in value is obtained.

[Product of two behavior matrices]
Here, three types of behavior keyword correlation, behavior correlation, and keyword correlation are considered as products of behavior matrices. In any of these calculations, a process for aligning the keywords and labels is performed before the calculation. That is, an operation that takes the union of the keyword labels of A _k and A _l is performed, and the rows of _Ak and A _l are expanded based on the resulting keyword labels, that is, the original keyword rows After all 0s are entered, the following formula is executed.

その具体的な計算は、下記のとおりである。

この計算は、２つの行動マトリクスが示す興味と行動の共通項を出力する。これによって、興味と行動の両方の側面で関連する項目を抽出することができる。
ここで、a^k _ij、a^l _ijは、それぞれ、マトリクスＡ_k、Ａ_lのij成分である。 The behavior keyword correlation is defined by the following mathematical formula.

The specific calculation is as follows.

This calculation outputs a common term of interest and behavior indicated by the two behavior matrices. This makes it possible to extract items related to both interest and behavior.
Here, a ^k _ij and a ^l _ij are ij components of the matrices A _k and A _l , respectively.

  FIG. 13 shows an example of calculation of behavior keyword correlation matrices of two behavior matrices A and B.

この計算は、２つの行動マトリクスが示す行動の共通項を出力する。これによって、どのような行動について共通のキーワードを用いているかを知ることができる。 The behavior correlation is defined by the following mathematical formula. A ^T is a transposed matrix. The operation is a normal matrix multiplication. If the behavior matrix is an m × n matrix, the result is an n × n matrix.

This calculation outputs a common term of actions indicated by the two action matrices. As a result, it is possible to know what kind of behavior the common keyword is used.

  FIG. 14 shows an example of calculating the keyword correlation matrix of the two behavior matrices A and B.

Keyword correlation is defined by the following mathematical formula. B ^T is a transposed matrix. The operation is a normal matrix multiplication. If the behavior matrix is an m × n matrix, the result is an m × m matrix.

この計算は、２つの行動マトリクスが示すキーワードの共通項を出力する。これによって、どのようなキーワードについて共通の行動をとっているかを知ることができる。
図１５に、２つの行動マトリクスＡ、Ｂの行動相関行列の計算の例を示す。

This calculation outputs a common term of keywords indicated by the two behavior matrices. As a result, it is possible to know what keywords are used in common.
FIG. 15 shows an example of calculating the behavior correlation matrix of the two behavior matrices A and B.

[Degeneration of behavior matrix]
Degeneration of the behavior matrix includes degeneration in the row direction and degeneration in the column direction. Degeneration in the row direction can also be called an interest keyword list, and is given by the following formula.
V _word = AW _act ^T
Here, A is an action matrix, W _act is an action weight vector, and W _act ^T is its transposed vector. The number of dimensions of W _act is equal to the number of types of actions, and its component is a number between 0 and 1 in principle. The value is determined so as to give a larger value (weight) to the action to be emphasized. For example, writing a blog gives more weight than reading a blog. In this way, a weighted keyword sequence associated with an individual or community can be obtained.

  FIG. 16 shows an example in which the behavior matrix A is degenerated in the row direction, that is, an interest keyword list is obtained.

Degeneration in the column direction can also be called an action pattern list, and is given by the following equation.
V _act = W _word ^T A
Here, A is an action matrix, W _word is a keyword weight vector, and W _word ^T is its transposed vector. The number of dimensions of W _word is equal to the number of extracted keywords, and its component is a number between 0 and 1 in principle. The value is determined so that a larger value (weight) is given to a keyword to be particularly focused.

  FIG. 17 shows an example in which the behavior matrix A is degenerated in the column direction, that is, the behavior pattern list is obtained.

Several typical applications will be described below to assist in understanding the invention.
[Application Example 1]
A user wants to find a community close to his / her interest in the community system. In order to achieve such an object, the community system performs an operation for obtaining a keyword correlation between the user's action matrix and the existing community action matrix stored in the hard disk drive 310. . The community system presents only a keyword having a component greater than a certain value in the obtained keyword correlation matrix to the user with an appropriate GUI. As a result, the user can search for a community characterized by keywords that he / she has not been interested in so far but wants to work on in the future.

[Application Example 2]
One community system suffered from spam writing. However, it is difficult to identify who is a spammer in the system as a whole. Therefore, the system administrator tried to create an action pattern list by using the function of the present invention to degenerate each user's action matrix in the column direction. As a result, it was possible to assume that a user who has an action pattern list that has an unusually large number of comments written on a blog is a potential spammer.

[Application Example 3]
In the community system, one of the main revenue sources is advertising revenue. However, it is difficult to determine what kind of advertisement is effective for the community by the conventional method. However, according to this embodiment, it is possible to generate an individual behavior matrix, and further generate a community behavior matrix based on the matrix. Then, by degenerate in the direction of the community, that is, by generating the keyword list of interest, a weighted keyword list can be obtained. Can be displayed on the screen.

  It should be understood that the operations such as matrix sum, difference, and product shown in the above embodiments are examples for carrying out the present invention and are not limited to these specific expressions. I want. For example, the calculation of the sum and difference may be any algorithm as long as it includes the calculation of the actual matrix sum and difference and the calculation of normalizing the values resulting from the calculation within a reasonable range.

It is a figure which shows that a client computer is connected to an online community server via the internet. It is a figure which shows the hardware constitutions of a client computer. It is a figure which shows the hardware constitutions of an online community server. It is a figure which shows action matrix production | generation conceptually. It is a figure which shows the function of API using the action matrix DB and its data. It is a functional block diagram of action matrix generation processing. It is a figure which shows an action matrix production | generation process concretely. It is a flowchart which shows the process of action matrix production | generation. It is a figure which shows access to a document using an access log. It is a figure which shows the example of the process of time transition of an action matrix. It is a figure which shows the example of the sum of action matrix. It is a figure which shows the example of the difference of action matrix. It is a figure which shows the example of calculation of an action keyword correlation matrix. It is a figure which shows the actual example of calculation of a keyword correlation matrix. It is a figure which shows the example of calculation of an action correlation matrix. It is a figure which shows the example of calculation of an interest keyword list. It is a figure which shows the example of calculation of an action pattern list | wrist.

Claims (20)

A community server system in which a plurality of users are connected via individual client computers and communicate through activities such as reading and writing documents.
A storage device capable of reading and writing data by the system;
Means for storing each user's activity and a document related to the activity as a log in the storage device together with the individual user's ID and activity content;
Analyzing a document read and written for each user's activity based on the log, obtaining a keyword in the document for each individual user's activity and its appearance frequency, and writing to the storage device,
Community server system.   The analysis result of the document read / written for each activity of the user includes the keyword in the first direction, the action in the second direction perpendicular to the first direction, and the intersection pointed by the label at the keyword. The system of claim 1, stored in the storage device as matrix-structured data in which the frequency of occurrence is stored.   The system of claim 1, wherein the activities include writing a blog, reading a blog, writing a message, and reading a message within a community server.   The system of claim 1, further comprising means for normalizing by dividing the frequency of occurrence of the keyword by the number of documents associated with the activity.   The system of claim 2, further comprising means for obtaining matrix structure data associated with users in the community by calculating a sum of the matrix structure data of users in the community.   6. The system of claim 5, further comprising means for calculating a difference between the community matrix structure data and a particular user matrix structure data.   3. The system according to claim 2, further comprising: means for multiplying the matrix structure data by a parameter to be attenuated; and means for adding newly calculated matrix structure data to the matrix structure data multiplied by the attenuation parameter. . An activity recording method for a community server system in which a plurality of users are connected via individual client computers and communicate with each other through activities such as reading and writing documents.
Storing each user's activity and a document related to the activity as a log in the storage device of the community server system together with the individual user's ID and activity contents;
Analyzing a document read and written for each activity of each user based on the log, obtaining a keyword in the document for each activity of each user and its appearance frequency, and writing to the storage device,
Activity recording method for community server system.   The analysis result of the document read / written for each activity of the user includes the keyword in the first direction, the action in the second direction perpendicular to the first direction, and the intersection pointed by the label at the keyword. 9. The method of claim 8, wherein the frequency of occurrence is stored in the storage device as matrix structured data.   9. The method of claim 8, wherein the activities include writing a blog, reading a blog, writing a message, and reading a message within a community server.   The method of claim 8, further comprising normalizing by dividing the frequency of occurrence of the keyword by the number of documents associated with the activity.   10. The method of claim 9, further comprising obtaining matrix structure data associated with users in the community by calculating a sum of the matrix structure data of users in the community.   13. The method of claim 12, further comprising the step of calculating a difference between the community matrix structure data and a particular user matrix structure data.   10. The method of claim 9, further comprising: multiplying the matrix structure data by a parameter to be attenuated; and adding the newly calculated matrix structure data to the matrix structure data multiplied by the parameter to be attenuated. . A program for recording activities in a community server system in which a plurality of users are connected via individual client computers and communicate by activities such as reading and writing documents,
The community server system,
Storing each user's activity and a document related to the activity as a log in the storage device of the community server system together with the individual user's ID and activity contents;
Analyzing a document read and written for each individual user activity based on the log, obtaining a keyword in the document for each individual user activity and its appearance frequency, and executing the step of writing to the storage device,
program.   The analysis result of the document read / written for each activity of the user is that the keyword is labeled in the first direction, the action is labeled in the second direction orthogonal to the first direction, and the intersection of the keyword indicates the keyword The program according to claim 15, wherein the program is stored in the storage device as data having a matrix structure in which appearance frequencies are stored.   16. The program of claim 15, wherein the activities include writing a blog, reading a blog, writing a message, and reading a message within a community server.   The program according to claim 15, further comprising the step of normalizing by dividing the appearance frequency of the keyword by the number of documents related to the activity.   The program according to claim 16, further comprising: obtaining matrix structure data related to users in the community by calculating a sum of data of the matrix structure of users in the community.   The program according to claim 16, further comprising: multiplying the matrix structure data by a parameter to be attenuated; and adding the newly calculated matrix structure data to the matrix structure data multiplied by the attenuation parameter. .

Images