JP2010066891A - Document classification method and system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To appropriately classify a plurality of documents extracted from a blog site or the like based on which document results in the contribution of those documents. <P>SOLUTION: A processor 10 is provided with: a document group extraction part 11 for extracting a classification target document from a blog server 22; a describer specification part 12 for specifying a describer by referring to a member server 23 about each classification target document; a document analyzing part 13 for specifying whether or not a nickname exists in the classification target document by document-analyzing the classification target document, and referring to a nickname database 24; and a determination part 14 for performing classification determination processing to the classification target document according to a predetermined type based on whether the describer is a diary contributor or the others, and whether or not the nickname exists in the classification target document. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a method and system for appropriately classifying documents posted from a large number of writers on a site on a communication network.

  In recent years, the number of users of Internet blog sites, SNS (Social Networking Service), etc. has increased explosively. It can be said that many users such as blogs desire to gain the sympathy of others through the publication of diary documents, or to meet others who can sympathize by browsing the diary documents. Blogs provide people with empathy, security, and problem-solving opportunities through the posting of diary documents including experiences, impressions, and opinions related to daily problems and current affairs, and the opportunity to post comment documents on the diary documents. It can be said that it now plays the role of social contribution.

By the way, it is necessary to perform some kind of search in order to find others who can sympathize with each other on a blog site or the like. Examples of such conventional search methods include random search, category search, and attribute / keyword search. Random search is a technique of manually browsing a Web page by performing a search that specifies a posting date and a poster. Category search is a technique for searching by utilizing theme classifications such as “child care” and “volunteer”. The attribute / keyword search is a technique for searching using the poster's age, gender, and keyword. As a keyword search technique for a blog site, for example, Patent Literature 1 can be exemplified.
JP 2007-11651 A

  By the way, blog sites and SNSs tend to become huge due to the large number of bloggers, and there are actually sites with millions of members. Even if such a huge site is searched by relying on a conventional search method, it is difficult for the user to efficiently and accurately find a person whom he / she empathizes with. That is, in random search and category search, it is difficult to find a sympathy destination because the search is too rough. On the other hand, in keyword search, an accurate search result cannot be obtained unless an appropriate keyword is selected and complicated search settings are made. Or, even if complicated search settings are made, the number of hits may become enormous.

  As described above, as a result of not being able to find a person who can sympathize with each other, there is a concern that the posting of comments to the diary document of the blog site is stagnant and the activity of the site is reduced. This is not desirable for publishers.

  Therefore, in order to identify sympathetic people, by extracting and analyzing documents already created by those who have a history of document communication on the blog site, that is, documents actually created by those who actually sympathize A method of deriving a class of judgment formulas for determining the co-sensitivity numerically, applying a document related to the creation of persons who have not yet met this judgment formula, and automatically determining the co-sensitivity between the persons is considered. .

  When taking such a technique, it is important to first classify the extracted document accurately. For example, as a communication result document, a document group including a diary document posted by a specific person on a blog site and a plurality of comment documents created related to this diary document and posted to the blog site in time series is extracted. In this case, the document included in the document group is a comment document posted by another writer in direct response to the diary document, or a comment document posted by the specific person in response to the comment document. In order to improve the accuracy of the determination formula, it is important to accurately classify whether it is a comment document posted by another writer in response to the comment document.

  The present invention has been made in view of the above problems, and can accurately classify a plurality of documents extracted from a blog site or the like to which documents are posted in response to which document. And to provide a system.

  A document classification method according to one aspect of the present invention is a method for classifying documents posted from a large number of writers on a site on a communication network, and is a method created by a first writer and posted to the site. A document created in response to the first document when there is a document group including one document and a plurality of documents created in relation to the first document and posted to the site in time series A second document created by a second writer different from the first writer and posted on the site, and a document created in response to the second document, the first description A third document created by a user and posted to the site, and a document created in response to the second document, created by a third writer different from the first writer and the second writer And a fourth document posted on the site. A first step of identifying, for each classification target document in the document group, the description of the person who posted the classification target document, and the description person other than the first description person In the classification target document, it is determined whether or not a personal identification name predetermined in association with each writer exists, and if it does not exist, it is determined as the second document. A second step of determining that the document is a fourth document in the document group when a document written by the descriptive person related to the individual specific name exists in time series before the document to be classified; Is the first descriptive person, it is determined whether or not a personal identification name previously defined in association with each descriptor exists in the classification target document. And the classification pair is included in the document group. And a third step of determining the third document as the third document when the posted document of the descriptive person related to the personal identification name exists in time series before the document (Claim 1). .

  According to this configuration, whether or not each classification target document in the document group created in relation to the first document and sequentially posted to the site in time series includes an individual specific name in the classification target document. In addition, based on the time-series positional relationship between documents, it can be accurately classified into any of the second to fourth documents. That is, it is possible to accurately grasp which writer is a document posted in response to which document on the site.

  In the third step, when the personal identification name does not exist, it can be determined that the classification target document is the third document corresponding to the immediately preceding second document in time series (claim). 2).

  According to this configuration, whether the third document corresponds to the immediately preceding second document in chronological order or not depending on whether the personal identification name exists or not is posted. Further subdivision can be made depending on whether it is in response to the second document.

  The personal identification name is preferably a nickname assigned to each writer (claim 3). According to this configuration, since a nickname that is widely used for specifying the other party in the document on the site is used, there is an advantage that it is easy to identify the document.

  In the above configuration, it is preferable that the site on the communication network is a specific website on the Internet. In particular, the website is a blog site, the first document is a diary document created by the first writer, and the second document is created by a writer other than the first writer. The first comment document for the diary document, the third document is a second comment document for the first comment document created by the first writer, and the fourth document is the first comment document. It is desirable that it is a third comment document for the first comment document created by a descriptor other than the descriptor.

  According to this configuration, the first comment document for the diary document posted on the blog site and the second and third comment documents for the first comment document are the classification target documents. In this case, it is possible to solve the problem that the second comment document created by the diary contributor himself is handled as the first and third comment documents, and the third comment document is handled as the first comment document.

  A document classification system according to another aspect of the present invention is a system for classifying documents posted from a large number of writers on a site on a communication network, the document database storing documents posted on the site, A descriptor database that stores names of writers who are scheduled to post to the site, a name database that stores personally specified names previously associated with each of the writers, and documents posted to the site Classification processing means for performing classification processing, wherein the classification processing means is created from the document database in association with the first document created by the first writer and posted to the site. A document group extracting unit for extracting a document group including a plurality of documents posted to the site in time series, and for each categorization target document of the document group, the descriptor data A database is referred to, a writer specifying unit for specifying a writer who has posted the classification target document, a document analysis of the classification target document and a reference to the name database, and a personal identification name is assigned to the classification target document. Document analysis unit for specifying whether or not the document exists, whether the writer is the first writer or a person other than the first writer, and whether or not a personal identification name exists in the classification target document And a determination unit that performs a classification determination process on the classification target document according to a predetermined type (Claim 6).

  According to this configuration, each classification target document in the document group created in relation to the first document and sequentially posted to the site in time series is referred to the writer database and the name database, and the classification target document And the presence / absence of the personal identification name can be specified, and based on the result, the determination unit can determine the classification of the document according to the type.

  In this case, the determination unit is a document created in response to the first document, the second document created by a second writer different from the first writer, and posted to the site; A document created in response to the second document, a third document created by the first writer and posted to the site, and a document created in response to the second document, Categorizing at least a fourth document created by a third writer different from the first writer and the second writer and posted on the site, and the writer of the classification target document If the personal identification name is not present in the document to be classified and the personal identification name does not exist, it is determined as the second document. If the personal identification name exists, the classification is included in the document group. The personal identification name in chronological order before the target document If there is a posted document of the writer who relates to the document, it is determined as the fourth document, the writer of the classification target document is the first writer, and the personal identification name exists in the classification target document If there is a document posted by the descriptor related to the personal identification name in time series before the classification target document in the document group, the third document It is desirable to determine (claim 7).

  According to this configuration, the determination unit determines whether the classification target document is any one of the second to fourth documents based on the presence / absence of the personal identification name in the classification target document and the time-series positional relationship between the documents. Can be accurately classified. That is, it is possible to accurately determine which writer is a document posted in response to which document of the site.

  According to the present invention, for each classification target document in the document group created in relation to the first document and sequentially posted to the site, which writer describes in response to which document of the site It is possible to accurately grasp whether the document is a completed document. Therefore, for example, when extracting documents of those who sympathize on the site, the extraction can be performed accurately. Eventually, when the extracted document is subjected to document analysis to derive a determination formula for determining the co-sensitivity numerically, the accuracy of the determination formula can be improved.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram schematically showing a hardware configuration of a network system S (document classification system) to which a document classification method according to the present invention is applied. This network system S is owned by the processing device 10, the blog system 20 for developing the blog site 21 on the Internet, and the members A, B, C, D, E. The terminal devices 30 including the member terminal devices 31, 32, 33, 34, 35... Are connected via the Internet IN so that data communication is possible.

  The blog site 21 (site on the communication network) is a specific web site developed on the Internet, and is a site on which a diary document, a comment document for the diary document, and the like are posted.

  The members A to E are persons who disclose their attribute information to the site operator of the blog site 21 and are registered as members of the blog site 21. The members A to E post a diary document to the blog site 21 via their member terminal devices 31 to 25, post a comment document (first comment document) to the diary document, It is possible to post a comment document (second comment document) to the document or to browse these documents. The terminal device 30 is typically a personal computer connected to the Internet, a mobile phone, a portable information terminal, or the like.

  The blog system 20 includes a blog server 22 (document database), a member server 23 (describer database), and a nickname database 24 (name database).

  The blog server 22 is a server for operating the blog site 21, and stores the document data of a document posted on the blog site 21 in association with various data such as the posting date of the document and the poster. It is. The member server 23 includes attribute information (name, member number, address, IP address of the terminal device, age, sex, etc.) of a member registered in the blog site 21, that is, a writer who is scheduled to post to the blog site 21. This is a server in which interests, etc.) are accumulated.

  The nickname database 24 is a database that stores nicknames (personally specified names) defined in association with each member. This nickname is a nickname of the user registered when he / she joins the blog site 21 and is used to identify himself / herself in a document posted to the blog site 21. Is the name that is planned.

  The processing device 10 (classification processing means) posted a diary document posted to the blog site 21, a comment document posted in response to the diary document, and a comment document further posted in response to the comment document. It has a function to perform classification according to members. For example, a diary document (first document) created by a member A (first writer) and posted to the blog site 21 and members B, C, D, E... It is assumed that there is a document group including a plurality of comment documents that are created by the above and posted to the blog site 21 sequentially in time series.

In this case, the processing apparatus 10 performs a process of classifying the plurality of comment documents into at least the following type 1, type 2, and type 3 as classification target documents.
[Type 1] A document created in response to a diary document, which is created by a member other than member A who is a diary writer, such as member B (second writer) and posted to the blog site 21 Comment document (second document);
[Type 2] A second comment document (third document) created in response to the first comment document and created by member A who is a diary writer and posted on the blog site 21;
[Type 3] A document created in response to the first comment document, created by a member other than member A and member B, for example, member C (third writer) and posted to the blog site 21 Comment document (fourth document).

  FIG. 2 is a functional block diagram showing a functional configuration of the processing apparatus 10. The processing device 10 is a large computer device equipped with a CPU (Central Processing Unit) that performs the above-described classification processing, for example, and is connected to the blog server 22, the member server 23, and the nickname database 24 so that data communication is possible. Has been. The CPU operates to include the functional unit shown in FIG. 2 by executing software programmed to perform the classification process. The processing apparatus 10 functionally includes a document group extraction unit 11, a writer specifying unit 12, a document analysis unit 13, a determination unit 14, and an analysis processing unit 15.

  The document group extraction unit 11 extracts a diary document and a group of comment documents related to the diary document (a thread for the diary document) from the document data stored in the blog server 22. At this time, the data regarding the posting date and time of these documents is attached, and the documents are extracted so that the time-series relationship between the documents (before and after the posting time) can be grasped. For example, when trying to obtain members B, C, D, E,... With high sensitivity to the member A on the blog site 21, the document group extracting unit 11 reads the diary document posted by the member A. A document group of a thread as a starting point is extracted from the blog server 22. This extracted document group becomes a classification target document.

  The writer specifying unit 12 refers to the member server 23 for each classification target document, and specifies the member who posted the classification target document. For example, the writer specifying unit 12 performs a specifying process based on an address and a password used when posting a document to the blog site 21.

  The document analysis unit 13 performs document analysis on the classification target document and specifies whether or not a nickname exists in the classification target document. Specifically, the document analysis unit 13 performs normalization processing, document structure analysis processing, and the like for each classification target document. The normalization process is a process for deleting characters, symbols, and the like unnecessary for analysis from the classification target document and unifying full-width and half-width characters so that the document structure analysis can be normally performed. In the document structure analysis processing, for each classification target document after normalization processing, for example, morphological analysis is performed to divide the document into words, and parsing processing is performed to determine dependency between words (nouns and verbs). And the like. For such document structure analysis processing, the document analysis unit 13 utilizes a thesaurus (synonym dictionary) that absorbs synonyms and notation fluctuations.

  In order to specify whether or not a nickname exists, the document analysis unit 13 refers to the nickname database 24. FIG. 3 is a table format diagram schematically showing the data structure of the nickname database 24. Here, the nicknames “a”, “b”, “c”, “d”, “e”,... Are stored in association with the member names (member numbers) A, B, C, D, E,. An example is shown. These nicknames, for example, were registered as handle names that each member A, B, C, D, E ... would use instead of their real name when posting a document to the site when registering as a member on the blog site 21 It is a name.

  The document analysis unit 13 compares the word group obtained by analyzing the document to be classified with the nickname stored in the nickname database 24, and specifies whether or not the nickname exists in the classification target document. Here, even a name existing in the nickname database 24 may be used in the document with a meaning other than the nickname. For example, if the nickname “a” of member A is “spring”, it is “spring” used as a noun indicating the four seasons, or “spring” as a nickname used to identify member A In some cases, it may be impossible to distinguish between the two.

  Therefore, in the case of use as a nickname, it is presumed that a title such as "~ chan", "~ san", "~ sama", "~ san", "~ sama" etc. is attached, the former and the latter The distinction is made. In the above case, after confirming that the word “spring” exists in the nickname database 24, the document analysis unit 13 follows “spring”, “˜chan”, “˜san”, etc. It is determined whether or not the honorific word exists. If it exists, it is determined as “spring” as the nickname, and it is determined that the document to be classified is a document including the nickname of the member A. On the other hand, if no honorific word follows, “Spring” is treated as a simple noun, and it is determined that the classification target document is not a document including a nickname.

  The determination unit 14 determines whether the categorization target document is based on whether the writer of the categorization target document is a diary contributor or another person and whether the categorization target document has a nickname. Then, a classification determination process is performed to determine which of the above-described type 1, type 2, and type 3 is applicable. The process in this determination part 14 is explained in full detail with reference to FIGS.

  FIG. 4 is a schematic diagram showing the diary 4A and the comment group 4B posted in association with the diary 4A in time series. In other words, it is a diagram schematically showing one thread actually generated starting from posting of the diary document 41 of member A. The arrow group on the left side in the figure indicates the correspondence between the diary and the comment and between the comment and the comment. The comment group 4B includes eight comment documents 42 to 49, and the uppermost comment document 42 in the figure is posted at the time closest to the posting time of the diary document 41. It is assumed that the comment documents 43 to 49 are arranged toward.

  The comment group 4B includes not only the members B, C, D, and E but also the comment document of the member A who is a diary contributor. In addition to the comment document that directly responds to the diary document 41, there is also a comment document that responds to the comment document. More specifically, the comment documents that directly correspond to the diary document 41 are the comment document 42 of the member B, the comment document 44 of the member C, and the comment document 45 of the member D. Member A has posted comment documents 43, 46, 47, and 48.

  That is, the member A immediately posts the comment document 43 in response to the comment document 42 of the member B, and then posts the comment document 47 after a while. The comment document 47 includes the word “Mr. b” in which the nickname “b” that identifies the member B is accompanied by the honorific word “Mr.”. Member A posts a comment document 48 in response to member C's comment document 44. The comment document 48 includes a word “c-san” in which the nickname “c” identifying the member C is accompanied by the honorific word “sama”. Further, the member A immediately posts the comment document 46 in response to the comment document 45 of the member D.

  In response to the member D's comment document 45, the member E also posts a comment document 49 in response. The comment document 49 includes the word “d-chan” in which the nickname “d” for identifying the member D is accompanied by the honorific word “chan”. Here, the member E does not communicate with the member A who is a diary contributor in the thread. That is, member E does not sympathize with member A, but sympathizes with member D and posts comment document 49.

  When the comment documents 42 to 49 are classified in the above factual relationship, two classification methods shown in FIG. 5 (Comparative Example 1) and FIG. 6 (Comparative Example 2) are provided as comparative examples for the present invention. Can be mentioned. Comparative example 1 is a classification method in which all comment documents of a thread are treated as comment documents corresponding to a diary document. Comparative Example 2 is a classification method in which a comment document excluding a comment document posted by the diary contributor (member A in this example) is treated as a comment document corresponding to the diary document.

  As shown in FIG. 5, in the classification method of Comparative Example 1, all comment documents 42 to 49 in the comment group 4 </ b> B are handled as comment documents corresponding to the diary document 41. With this method, the mechanical processing speed is fast, but it is not possible to extract the “diary-comment” relationship that should not be extracted or the “comment-comment” relationship that must be extracted. May occur.

  Specifically, the comment document 43, 46, 47, 48 for the other member A who is the diary contributor is treated as a comment document for the diary document 41, and a clearly contradictory “diary-comment” relationship is extracted. Will end up. Further, an error is caused in which a comment document 49 that is not a comment for the diary document 41 is treated as a comment document for the diary document 41. On the other hand, a comment document 43 corresponding to the comment document 42, a comment document 48 corresponding to the comment document 44, a comment document 47 corresponding to the comment document 45, and a comment document 49 corresponding to the comment document 45 must be originally extracted. There arises a problem that the “comment-comment” relationship that cannot be extracted.

  Next, as shown in FIG. 6, in the classification method of the comparative example 2, the comment document 42 excluding the comment documents 43, 46, 47, and 48 posted by the member A himself / herself of the diary poster in the comment group 4B. , 44, 45, 49 are treated as comment documents corresponding to the diary document 41.

  According to the classification method of the comparative example 2, the contradiction that the comment documents 43, 46, 47, 48 for the other members of the member A are handled as the comment documents for the diary document 41 can be solved. However, it is impossible to eliminate an error that treats a comment document 49 that is not a comment for the diary document 41 as a comment document for the diary document 41. Further, as in Comparative Example 1, there remains a problem that the “comment-comment” relationship that should be extracted cannot be extracted.

  FIG. 7 is a table format in which the accuracy of the classification methods of Comparative Example 1 and Comparative Example 2 is evaluated. Items marked with “x” in the figure are items that are defective in the comparative method. As described above, in the classification method of Comparative Example 1, the comment document of the diary poster himself is extracted as a comment document in the “diary-comment” relationship, and the “comment-comment” relationship is the “diary-comment” relationship. Misleading faults occur. However, it does not occur up to a problem of extracting an erroneous “comment-comment” relationship, for example, treating the comment document 48 as a comment corresponding to the comment document 42. Moreover, although there are errors, the comment documents 42, 44, and 45 having the correct “diary-comment” relationship can be extracted without omission. However, for example, there is a problem that a correct “comment-comment” relationship of the comment document 49 corresponding to the comment document 45 cannot be extracted.

  The classification method of Comparative Example 2 is the same as Comparative Example 1 except that the error of extracting the comment document of the diary contributor himself as a comment document related to “diary-comment” is eliminated. According to such a classification method, for example, in order to detect other members having high co-sensitivity to member A, a document communicated between member A and another member is extracted, and the document is analyzed. In the case where the co-sensitivity is determined numerically, the inter-member document, which is a prerequisite for accurate co-sensitivity determination, is not accurately extracted. Accordingly, the accuracy of the co-sensitivity determination naturally decreases.

  The determination unit 14 describes the classification target document specified by the writer specifying unit 12 and the classification determined by the document analysis unit 13 so that the items marked with “×” shown in FIG. With reference to the presence / absence of a nickname in the target document and the posting time, the classification target document is classified according to the following logics (a) to (c).

  (A) If the writer of the classification target document is other than the member A who is the diary contributor, and there is no nickname in the classification target document, the comment document of the other member posted in response to the diary document Handled (see above [Type 1]). In the example of FIG. 4, comment documents 42, 44, and 45 correspond. Here, in a comment document that responds directly to a diary document, the other party is obvious, and it is not usually necessary to identify the other party in the text, so the assumption that the nickname will not be described in the comment document is adopted. ing.

  (B) In the case where the writer of the classification target document is other than the member A who is the diary contributor and the nickname exists in the classification target document, the comment group 4B is more time-sequential than the classification target document. If there is a member's comment document relating to the nickname first, it is treated as a comment document exchanged with other members ([Type 3] above). In the example of FIG. 4, a comment document 49 corresponds. Here, when posting a comment document in response to another member's comment document in the thread instead of a diary document, it is usually assumed that a nickname identifying the other party will be described in the comment document. Adopted. In the methods of Comparative Examples 1 and 2, the comment document 49 could not be correctly classified. However, according to the method of the present embodiment, the comment document 49 includes a description “d-chan”. This can be classified accurately.

  (C) When the writer of the classification target document is a member A who is a diary contributor and the nickname does not exist in the classification target document, and there is a nickname in the comment group 4B from the classification target document. If there is a member's comment document related to the nickname in chronological order, it is treated as a comment document for a comment document related to another member's post posted by the diary contributor (see [Type 2 ]). If there is no nickname in the classification target document, it is determined that the classification target document is a comment document posted in response to the previous comment document in time series, so that the nickname is included. It is possible to determine which comment document corresponds to a comment document that does not exist. In the example of FIG. 4, comment documents 43, 46, 47 and 48 correspond.

  Here, in the comment document that responds to the comment document posted immediately before, the other party is obvious and usually there is no need to specify the other party in the sentence, so the nickname will not be described in the comment document The estimation is adopted. In addition, when a comment document is posted in response to a comment document that has been posted before, but not immediately before, a presumption is made that a nickname that identifies the other party will be described in the sentence. According to this method, the relationship between the comment document 43 corresponding to the comment document 42 and the comment document 46 corresponding to the comment document 45 can be accurately extracted. The comment document 47 corresponding to the comment document 42 and the comment document 48 corresponding to the comment document 44 based on the description “Mr. b” in the comment document 47 and the description “c” in the comment document 48. The relationship can be extracted accurately.

  The analysis processing unit 15 performs document analysis on posted documents (diary document 41, comment documents 42 to 48) of members (for example, members B, C, and D with respect to member A) in the blog site 21, that is, Documents exchanged between those who already sympathize with each other are analyzed, and a multivariate analysis is performed to derive a judgment formula for evaluating the co-sensitivity. As a method of multivariate analysis, for example, multiple regression analysis, discriminant analysis, quantification class I, and quantification class II can be employed. At this time, parameters obtained by the document analysis are appropriately set as “object variable” and “explanatory variable”.

  Furthermore, the analysis processing unit 15 performs a process of evaluating the co-sensitivity of members who have not yet communicated using the determination formula. In this case, a document created by a member who is subject to co-sensitivity determination is extracted from the blog server 22 and document analysis is performed to derive parameters for the “object variable” and “explanatory variable”. Apply to the judgment formula to evaluate the co-sensitivity between members numerically. When a combination of members whose co-sensitivity exceeds a predetermined threshold is found, that effect is displayed on a display means (not shown). Accordingly, the blog site 21 can be activated by accurately extracting members who are likely to sympathize with the blog site 21 and providing information to these members.

The operation of the document classification process of the processing apparatus 10 described above will be described based on the flowchart shown in FIG. First, the document group extraction unit 11 extracts a diary document and comment documents C _{1 to} C _n related to the diary document from the document data stored in the blog server 22 (step S1). Then, among the extracted comment documents, the first document C ₁ = C _m and the classification target document M = C _m are set (step S2).

Next, the member who contributed the classification target document M is specified by the writer specifying unit 12 (step S3). Thereby, it is determined whether the contributor of the _first document C1 is the diary contributor X or the comment contributor Y other than the diary contributor.

  If the contributor is the diary contributor X in step S3, the document analysis unit 13 analyzes the document to be classified and determines whether or not a nickname exists in the document to be classified M (step S4). . When the nickname does not exist (NO in step S4), the determination unit 14 responds to the comment document in which the document M has posted the diary document (member A in the example of FIG. 4) in time series. It is determined that the comment document has been posted (step S5). In the example of FIG. 4, the comment documents 43 and 46 correspond.

  When the nickname exists in the classification target document M (YES in step S4), the determination unit 14 further determines whether the member's comment document related to the nickname exists in time series before the classification target document M. It is confirmed whether or not (step S6). If such a member's comment document exists (YES in step S6), the comment document including the nickname is determined as a comment document posted by the diary contributor (step S7). In the example of FIG. 4, the comment documents 47 and 48 correspond.

  On the other hand, if there is no comment document posted by the member corresponding to the nickname first (NO in step S6), that is, the member registered in the nickname database 24 but corresponding to the nickname. If the comment document is not included in the comment group 4B, it cannot be specified which document is sympathetic to the document, and is excluded from the classification target (step S12). For example, a comment document that does not appear in FIG.

  On the other hand, if the contributor is a comment contributor Y other than the diary contributor in step S3, the document analysis unit 13 similarly analyzes the document to be classified, and whether the nickname exists in the classification target document M. Whether or not is specified (step S8). When the nickname does not exist (NO in step S8), the determination unit 14 determines that the document M is a comment document of another member posted in response to the diary document (step S9). In the example of FIG. 4, comment documents 42, 44, and 45 correspond.

  When the nickname exists in the classification target document M (YES in step S8), the determination unit 14 further determines whether the member's comment document related to the nickname exists in time series before the classification target document M. It is confirmed whether or not (step S10). If such a member's comment document exists (YES in step S10), the comment document including the nickname is determined as a comment document posted by the comment poster (step S11). In the example of FIG. 4, a comment document 49 corresponds.

  On the other hand, when there is no comment document posted by the member corresponding to the nickname first in time series (NO in step S10), that is, the member registered in the nickname database 24 but corresponding to the nickname. If the comment document is not included in the comment group 4B, it cannot be specified which document is sympathetic to the document, and is excluded from the classification target (step S12).

After completion of the above determination process, whether classifying target document M is still remaining, whether that is M = C _n is checked (step S13). If the classification target document M remains (NO in step S13), _Cm is incremented by 1 (step S14), and the classification target document M = _Cm is set (step S15). Thereafter, the process returns to step S3, and the determination process for the next classification target document is performed. If the classification target document M does not remain (YES in step S13), the process ends.

  According to the network system S described above, a comment document created in relation to a diary document and sequentially posted to the site in time series, whether or not the nickname is included in the document, and the time between documents Based on the sequential positional relationship, the diary-comment relationship and the comment-comment relationship can be accurately classified. That is, it is possible to accurately grasp which writer is a document posted in response to which document on the site. Therefore, for example, when extracting documents of those who sympathize on the site, the extraction can be performed accurately. Eventually, when the extracted document is subjected to document analysis to derive a determination formula for determining the co-sensitivity numerically, the accuracy of the determination formula can be improved.

1 is a configuration diagram schematically showing a hardware configuration of a network system to which a document classification method according to the present invention is applied. It is a functional block diagram which shows the function structure of a processing apparatus. It is a figure of the table format which shows the data structure of a nickname database typically. It is the schematic diagram which showed the diary document and the comment group contributed in connection with this in time series. It is a schematic diagram which shows the comparative example 1 with respect to this invention. It is a schematic diagram which shows the comparative example 2 with respect to this invention. It is the figure of the table format which evaluated the precision of the classification method of the comparative example 1 and the comparative example 2. It is a flowchart which shows the operation | movement of the document classification process by a processing apparatus.

Explanation of symbols

S network system (document classification system)
10. Processing device (classification processing means)
DESCRIPTION OF SYMBOLS 11 Document group extraction part 12 Writer specific part 13 Document analysis part 14 Judgment part 15 Analysis processing part 20 Blog system 21 Blog site (site on communication network)
22 Blog server (document database)
23 Member server (describer database)
24 Nickname database (name database)
30 Terminal device

Claims (7)

A method for classifying documents posted from a large number of writers on a site on a communication network, the first document created by a first writer and posted to the site, and created in relation to the first document When there is a document group including a plurality of documents posted to the site in time series,
A document created in response to the first document, created by a second writer different from the first writer, and posted to the site;
A document created in response to the second document, the third document created by the first writer and posted to the site;
A document created in response to the second document, the fourth document created by a third writer different from the first writer and the second writer, and posted to the site;
In order to classify at least the classification target documents of the document group,
A first step of identifying another description who posted the classification target document;
If the writer is other than the first writer, it is determined whether or not there is a personally specified name previously defined in association with each writer in the classification target document. In the case where it is determined that the document is the second document, and there is a document posted by the descriptor associated with the personally specified name in time series before the classification target document in the document group, the second document A second step of determining four documents;
When the descriptive person is the first descriptive person, it is determined whether or not an individual specific name defined in advance in association with each descriptive person exists in the classification target document. A third step of determining that the document is a third document if there is a document posted by the writer who relates to the personal identification name in time series before the document to be classified in the document group. When,
A document classification method characterized by comprising: In the third step,
2. The document according to claim 1, wherein if the personal identification name does not exist, it is determined that the classification target document is the third document corresponding to the immediately preceding second document in time series. Classification method.   The document classification method according to claim 1, wherein the personal identification name is a nickname assigned to each writer.   The document classification method according to claim 1, wherein the site on the communication network is a specific website on the Internet. The website is a blog site,
The first document is a diary document created by the first writer;
The second document is a first comment document for a diary document created by a writer other than the first writer;
The third document is a second comment document for the first comment document created by the first writer;
The fourth document is a third comment document for the first comment document created by a descriptor other than the first descriptor.
The document classification method according to claim 4, wherein: A system for classifying documents posted by a large number of writers on a site on a communication network,
A document database for storing documents posted on the site; a descriptor database for storing descriptor names scheduled to be posted to the site; and a personally specified name previously determined in association with each of the writers A name database for storing the information, and a classification processing means for performing classification processing of documents posted on the site,
The classification processing means includes
A first document created by a first writer and posted to the site from the document database, and a plurality of documents created in relation to the first document and posted to the site in time series A document group extraction unit for extracting a document group;
For each classification target document of the document group, referring to the description database, a descriptor specifying unit for specifying a descriptor who posted the classification target document;
A document analysis unit that analyzes the document to be classified and refers to the name database to identify whether or not a personal identification name exists in the document to be classified;
Based on whether the descriptive person is the first descriptive person or other than the first descriptive person, and whether or not a personal identification name exists in the classification target document, according to a predetermined type, A determination unit that performs a classification determination process on the classification target document;
A document classification system characterized by including: The determination unit
A document created in response to the first document, created by a second writer different from the first writer, and posted to the site;
A document created in response to the second document, the third document created by the first writer and posted to the site;
A fourth document created in response to the second document, created by a third writer different from the first writer and the second writer, and posted to the site;
Are classified at least,
When the writer of the classification target document is other than the first writer and the personal identification name does not exist in the classification target document, it is determined as the second document, and the personal identification name exists In the case where there is a document posted by the writer who relates to the personal identification name in time series before the classification target document in the document group, it is determined as the fourth document,
The description target of the classification target document is the first description person, and the personal identification name does not exist in the classification target document. If there is a writer's post document related to the personal identification name in time series, the third document is determined.
The document classification system according to claim 6.

Images