JP2009151373A - Citation relation extraction system, citation relation extraction method, and citation relation extracting program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To extract an implicit citation relation in a content, and to enhance the precision of implicit citation relation extraction. <P>SOLUTION: This citation relation extraction system, this citation relation extraction method and this citation relation extracting program calculate a degree of citation indicating a degree of possibility of citation made between the contents, based on a difference of preparation, updating or reference time between contents, and based on a degree of relation between writers who have prepared, updated or cited the content. The citation relation extraction system, the citation relation extraction method and the citation relation extracting program extract the citation relation between the contents, based on the calculated degree of citation. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a citation relationship extraction system, a citation relationship extraction method, and a citation relationship extraction program for extracting citation relationships between contents.

  In daily operations of organizations such as companies, new content is often created by quoting or referring to content created by supervisors, colleagues, or subordinates. For example, a report may be created by incorporating a part of the project member's material. Clarifying the citation relationship directly leads to clarifying the original work and source, and is important for copyright protection. Also, finding citation relationships for the entire content group plays an important role in organizing the content and facilitating searches.

  For example, a research institution or the like may evaluate a researcher by measuring the impact factor of a paper based on the number of citations, or create a network of documents based on a citation relationship and organize a group of documents. It has been broken.

  However, unlike papers used in research institutions, etc., the citation relationship of in-house content used in organizations such as companies is often not specified. For this reason, there are cases in which the merit of the content organization technology based on the citation relationship cannot be fully enjoyed, and the original idea creator cannot be evaluated correctly.

  In the following, citation relationships that are not specified in the in-house content are referred to as implicit citation relationships. In addition, hereinafter, unless otherwise specified, “quoting relationship” indicates an implicit quoting relationship.

  Citation is a relationship between two contents or between parts in two contents (hereinafter referred to as passages). Citation refers to reusing part or all of a certain content (hereinafter referred to as “citation content”) as it is or in a part of another content (hereinafter referred to as “citation content”). Hereinafter, a part or all of the citation source content is referred to as a citation source passage, and a part or all of the citation destination content is referred to as a citation destination passage.

  Since citation relationships are an important factor for organizing documents and facilitating search, many techniques for finding citation and reuse relationships of documents and contents have been proposed. For example, Patent Document 1 describes a method of extracting a content reuse relationship using surface layer information such as the appearance location and the number of appearances of the same character string by finding the same character string by matching character strings. Further, Patent Document 2 describes a method for automatically generating a link relationship based on passage similarity.

International Publication No. 2004/034282 Pamphlet JP 2000-3372 A

  However, the citation relationship may be erroneously determined only by using the methods described in Patent Document 1 and Patent Document 2. For example, papers written by two unconnected researchers at almost the same time can be said to have no citation relationship between them. However, in this case, if only the methods described in Patent Document 1 and Patent Document 2 are used, it is determined that the character strings and passages are similar and may be erroneously recognized as having a citation relationship. In other words, in the methods described in Patent Document 1 and Patent Document 2, if the creation times of two contents are very far apart, or if the relationship between the creators of the contents is very far away in the organization, The possibility that the content is created independently is not considered.

  In general, it is impossible to quote a confidential document without access right in a system for setting access right. However, in the method of determining the citation relationship based on the similarity relationship described in Patent Literature 1 and Patent Literature 2, even if it is a confidential document without access right, if a similar character string or passage is included by chance, If there is a citation relationship, there is a possibility of being misdetected.

  Accordingly, the present invention has been made to solve the above-described problem, and enables the extraction of the implicit citation relationship in the content and the citation relationship extraction system that enables the accuracy of the implicit citation relationship extraction to be improved, It is an object to provide a citation relationship extraction method and a citation relationship extraction program.

  The citation relationship extraction system according to the present invention enables citation between contents based on the difference in creation, update or reference time between contents and the degree of relationship between authors who created, updated or referred to the content. A citation degree calculating means for calculating a citation degree indicating the degree of sexuality, and a citation relation extracting means for extracting a citation relation between contents based on the citation degree calculated by the citation degree calculating means. .

  In addition, the citation relation extraction method according to the present invention performs citation between contents based on the difference in creation, update or reference time between contents and the degree of relation between authors who created, updated or referred to the contents. A citation degree calculating step for calculating a citation degree indicating the degree of possibility, and a citation relation extracting step for extracting a citation relation between contents based on the calculated citation degree.

  In addition, the citation relationship extraction program according to the present invention enables a computer to extract between contents based on the difference in creation, update or reference time between contents and the degree of relationship between authors who created, updated or referenced the contents. A citation degree calculation process for calculating a citation degree indicating the degree of possibility that citations have been performed, and a citation relation extraction process for extracting a citation relation between contents based on the calculated citation degree. is there.

  According to the present invention, the citation relationship between contents is extracted in consideration of the difference in creation, update or reference time between contents and the degree of relationship between authors who created, updated or referenced content. Relationship detection errors can be excluded. Accordingly, it is possible to extract the implicit citation relationship in the content and improve the accuracy of the implicit citation relationship extraction.

Embodiment 1. FIG.
Hereinafter, a first embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing an example of the configuration of an implicit citation relationship discovery system (citation relationship extraction system) according to the present invention. The present invention relates to an in-house information system, for example. For example, an implicit citation relationship discovery system performs processing for finding an implicit citation relationship between in-house content, particularly regarding organization of in-house content such as an electronic document and facilitating search within an organization such as a company.

  The implicit citation relationship discovery system can be applied not only to electronic documents but also to contents such as still images and videos. However, for the sake of convenience, this embodiment is an example of contents unless otherwise specified. An explanation is given using an electronic document. For example, the citation source content and the citation destination content will be described as a citation source document and a citation destination document, respectively.

  First, the concept of the implicit citation relationship discovery system (citation relationship extraction system) according to the present invention will be described. In the present invention, a space-time constraint condition and a citation direction constraint condition for citation relations are introduced to provide a means for finding an implicit citation relation that can exclude erroneous judgment of citation relations.

  The citation direction constraint includes a time condition that the citation content must be created before the citation content, and a permission condition that the author of the citation content must have access to the citation content. including. The implicit citation relationship discovery system uses this citation direction constraint to eliminate a content pair in which a citation relationship cannot exist and to give an order relationship between a citation destination and a citation source.

  For space-time constraints, in order for citation relationships to occur, authors in organizational space must have a time interval condition that the difference in content creation time must be within a certain range and that there must be a connection between authors. Correlation conditions. These spatio-temporal constraints formulate general trends as follows in content citations.

  (1) Time interval condition: There is a low possibility that a citation relationship exists in content created at the same time. In other words, the closer the creation times of the citation source document and the citation destination document are, the higher the possibility that the citation relationship does not exist. Also, if the difference in creation time increases, the possibility of accessing the content increases, so the possibility of citation increases.

  However, if this difference in creation time becomes extremely large, unless the content is very good, it will be forgotten, and the possibility of being quoted will be low. In other words, as the difference in creation time increases, the citation possibility between contents once tends to increase and then gradually decrease.

  (2) Organization correlation condition: In an organization space, authors who are strongly connected are in close spaces, so there is a high possibility that they are communicating closely. Therefore, there is a high understanding of the other party's way of thinking and the content created by the other party, and the possibility of quoting the content is high.

  The implicit citation relationship discovery system extracts citation relationships between contents according to the above-described concept.

  As shown in FIG. 1, the implicit citation relationship discovery system includes a controller 100, a document database 101, an organization configuration table storage unit 102, an access database 103, a virtual citation database 104, an order relationship estimation unit 201, and a citation. Degree calculation means 202.

  Specifically, the implicit citation relationship discovery system is realized by an information processing apparatus such as a personal computer that operates according to a program. Note that the implicit citation relationship discovery system may be realized by one information processing apparatus or may be realized by using a plurality of information processing apparatuses. For example, the implicit citation relationship discovery system may be implemented using a plurality of information processing apparatuses that implement a document sharing system, a personnel management system, or the like installed in an organization such as a company.

  The document database 101 is a database that stores in-house documents (electronic documents) used in the company, and is specifically realized by a database device such as a magnetic disk device or an optical disk device. Further, the document database 101 may be realized by one or a plurality of database servers that realize a text sharing system in an organization such as a company.

  The document database 101 stores a set of document ID, file path, author ID, creation time, access level, and document type as information about the document. Further, the document database 101 stores a passage ID, a document ID, and a set of passages as information related to passages included in the document. Further, the document database 101 stores a set of document type and a unit time distance conversion parameter as information on a calculation parameter for the citation degree for each content type.

  The document database 101 stores, for example, “document information”, “passage information”, and “calculation parameter information” in advance at the timing of registering a document in the document sharing system. For example, the document database 101 stores an author name (for example, author ID) specified based on an ID and a password input at the time of document registration as one of “information about the document”.

  The passage is a part that semantically constitutes one unit included in the content (for example, an electronic document). For example, if the content is a document, the passage is a paragraph.

  The access level is information indicating the level of authority necessary for accessing the document. For example, the access level is represented by a number from 0 to 10. For example, the implicit citation relationship discovery system sets an appropriate access level when registering an electronic document in the document database 101 in accordance with the operation of the author who created the electronic document (content). Further, the user operates the user terminal or the like to access the document stored in the document database 101 according to the access level permitted by the user.

  The document creation time is the time when the electronic document (content) is registered in the document database 101.

  Among the information that makes up the information of the calculation parameters for the citation level for each type of content, the document type is, for example, type classification according to the purpose and use of documents such as notes, weekly reports, monthly reports, reports, and papers. It is information to show.

  The unit time distance conversion parameter is the time (effective period) from the creation of each type of document to the most likely time of reference. For example, in the case of weekly reports, if there is a high possibility that it will be referred to in about one month immediately after creation, and the possibility that it will be referred to after that becomes low, the value of the conversion parameter for unit time distance is one month. is there. Similarly, for example, in the case of monthly reports, the value of the conversion parameter for unit time distance is one year. In the document database 101, it is assumed that conversion parameters determined in advance according to the document type are registered in this manner.

  Specifically, the organization configuration table storage unit 102 is realized by a storage device such as a magnetic disk device or an optical disk device. The organization configuration table storage unit 102 stores an organization configuration table including author information, an adjacency matrix of the organization configuration graph, and an update time of the organization configuration graph. The organization structure table is created by, for example, a personnel department in an organization such as a company and registered in the organization structure table storage unit 102 in advance. Further, the implicit citation relationship discovery system may acquire the organization configuration table from the personnel management system in the organization and execute the processing.

  FIG. 2 is an explanatory diagram illustrating an example of the organization configuration graph, an adjacency matrix for the organization configuration graph, and author information. Among these, Fig.2 (a) has shown the example of the organization structure graph. As shown in FIG. 2B, the organization structure table storage unit 102 accumulates a set of authors and author IDs as author information included in the organization structure table. Further, as shown in FIG. 2C, the organization configuration table storage unit 102 stores a set of graph IDs and adjacency matrices as an adjacency graph expressed using the author ID.

  The organization configuration table storage unit 102 stores a set of graph IDs and update times as the update time of the organization configuration graph. The implicit citation relationship discovery system can manage the update of the organization graph due to the organization revision or personnel change using the update time information stored in the organization configuration table storage unit 102.

  The access database 103 is a database that stores document access right information, and is specifically realized by a database device such as a magnetic disk device or an optical disk device. The access right information is created in advance by a personnel department or a system management department in an organization such as a company and registered in the access database 103 in advance. Further, the implicit citation relationship discovery system may acquire access information from the personnel management system or access right management system in the organization and execute the processing.

  The access database 103 stores a set of author ID, update time, and access level as access right information. The access level is information indicating the level of access authority of the author corresponding to the author ID. For example, the access level is represented by a number from 0 to 10. Therefore, in this embodiment, only an author who has an access level higher than the access level necessary for accessing the document can access the document. The update time is a change time of the access level of the author due to the organization revision or personnel change. Therefore, in order to specify the access level, it is necessary to make a determination using the author ID and the update time at the same time.

  The virtual citation database 104 is a database that stores the citation relationship extraction results extracted by the implicit citation relationship discovery system, and is specifically realized by a database device such as a magnetic disk device or an optical disk device. The virtual citation database 104 includes a citation relationship document ID, a citation source passage ID, a citation destination document ID, a citation destination passage ID, and a citation degree as an extraction result of the passage citation relationship extracted by the implicit citation relationship discovery system. Is stored.

  Specifically, the controller 100 is realized by a CPU of an information processing apparatus that operates according to a program. The controller 100 has a function of extracting a citation relation between contents based on the citation degree calculated by the citation degree calculating means 202 and the order relation estimated by the order relation estimating means 201.

  In the present embodiment, the controller 100 includes the estimation result of the order relation estimation unit 201 (determination result of the estimation process) and the calculation result of the citation degree calculation unit 202 for all or part of the documents stored in the document database 101. To extract the implicit citation relationship. Further, the controller 100 stores the extraction result of the implicit citation relationship between contents (for example, electronic documents) in the virtual citation database 104.

  Specifically, the order relation estimation unit 201 is realized by a CPU of an information processing apparatus that operates according to a program. The order relationship estimation means 201 has a function of estimating the order relationship between content that can be a citation source and content that can be a citation destination. As illustrated in FIG. 1, the order relationship estimation unit 201 includes an access right determination unit 2011 and a time order determination unit 2012.

  In the present embodiment, the order relationship estimation unit 201 estimates the order relationship between the citation source and the citation destination of a document pair with a high citation level based on the citation direction constraint. That is, the order relation estimation unit 201 determines whether the citation source and the citation destination are based on the access right condition related to whether or not the citation source is accessible and the time condition that the citation source needs to be created before the citation destination. Is estimated.

  Among the means included in the order relation estimation means 201, the access right determination means 2011 has a function of examining the access right condition and estimating the order relation between the citation source and the citation destination. The access right judging means 2011 estimates the order relationship between the content that can be cited and the content that can be cited based on the level of access right set for the content and the level of access right set for the author. . In this case, if the access right determination unit 2011 determines that the access right level set for the author is equal to or higher than the access right level set for the content, the access right determination unit 2011 estimates the content as a content that can be cited.

  Further, the time order determination unit 2012 has a function of inspecting the time condition and estimating the order relation between the citation source and the citation destination. The time order determination unit 2012 estimates the order relationship between the content that can be the citation source and the content that can be the citation destination based on the content creation, update, or reference time (creation time in the present embodiment). In this case, the time order determination unit 2012 estimates content that has a long creation, update, or reference time as content that can be cited, and estimates content that has a new creation, update, or reference time as content that can be cited.

  Specifically, the citation level calculation means 202 is realized by a CPU of an information processing apparatus that operates according to a program. The citation level calculation means 202 has a function of calculating a citation level indicating the degree of possibility of citation of passages in an electronic document. The citation level calculation means 202 indicates the possibility that citations have been made between contents based on the difference in creation, update or reference time between contents and the degree of relationship between authors who created, updated or referred to the contents. The degree of citation indicating the degree is calculated. As shown in FIG. 1, the citation degree calculation means 202 includes a similarity degree calculation means 2021, a time distance calculation means 2022, an author distance calculation means 2023, and an integrated calculation means 2024.

  The similarity calculation unit 2021 has a function of calculating the similarity between contents. In this embodiment, the similarity calculation unit 2021 calculates the similarity between passages included in different documents. For example, the similarity calculation unit 2021 can obtain the similarity of passages in a document using a cosine of a keyword vector calculated based on a vector space model. For example, Document A “Takenobu Tokunaga,“ Information Retrieval and Language Processing ”, University of Tokyo Press, uses the cosine of a keyword vector calculated based on a vector space model to determine the similarity of passages in a document. Society, pp. 31, 41-43 ".

  Note that the similarity calculation unit 2021 calculates the similarity between contents using a similarity calculation method prepared according to the type of content when obtaining the similarity of content other than a document.

  The time distance calculation unit 2022 has a function of calculating a time distance indicating a difference in creation, update, or reference time between contents. In this embodiment, the time distance calculation unit 2022 calculates the absolute value of the difference between the creation times of two documents. Note that the time distance calculation unit 2022 may calculate the absolute value of the difference between the update time and the reference time of two documents, for example.

  In addition, the time distance calculation unit 2022 has a function of converting the absolute value of the obtained difference in creation time into a unit time distance. Note that parameters for conversion to unit time distance are determined for each document type and stored in the document database 101 in advance. Then, the time distance calculation unit 2022 acquires (extracts) the parameter for conversion to the unit distance from the document database 101 and normalizes it using the extracted parameter, thereby obtaining the absolute value of the difference in creation time as the unit time. Convert to distance. That is, the time distance calculation means 2022 uses the normalization parameter for normalizing the difference in creation, update or reference time (creation time in the present embodiment) according to the content type, and normalizes the time distance. Is calculated.

  The author distance calculation means 2023 has a function of calculating an author distance indicating the degree of relationship between authors who created, updated, or referred to content. In the present embodiment, the author distance calculation unit 2023 calculates the length of the shortest path between nodes corresponding to the document author in the organization structure graph as the author distance.

  The total calculation unit 2024 has a function of calculating a quoting degree obtained by integrating the time distance calculated by the time distance calculation unit 2022, the author distance calculated by the author distance calculation unit 2023, and the similarity calculated by the similarity calculation unit 2021. Is provided. In the present embodiment, the integrated calculation unit 2024 calculates the similarity between passages obtained by the similarity calculation unit 2021, the author distance obtained by the author distance calculation unit 2023, and the time distance obtained by the time distance calculation unit 2022. Used to calculate the citation level of passages in the document. That is, the degree of citation is a function of similarity, time distance, and author distance based on space-time constraints.

  In this embodiment, the storage device (not shown) of the information processing apparatus that implements the implicit citation relationship discovery system stores various programs for discovering the implicit citation relationship between contents (for example, electronic documents). is doing. For example, the storage device of the information processing apparatus that implements the implicit citation relationship discovery system allows a computer to determine the difference in creation, update, or reference time between contents and the degree of relationship between authors who created, updated, or referenced content. A citation degree calculation process for calculating a citation degree indicating the degree of possibility that citation was performed between contents, and a citation relation extraction process for extracting a citation relation between contents based on the calculated citation degree. An implicit citation relationship discovery program (citation relationship extraction program) for execution is stored.

  Next, the operation will be described. FIG. 3 is a flowchart illustrating an example of a process in which the implicit citation relationship discovery system extracts an implicit citation relationship included between contents (passages in an electronic document). The implicit citation relationship discovery system starts the implicit citation relationship extraction process shown in FIG. 3 at a predetermined timing.

  For example, the implicit citation relationship discovery system starts the extraction process of the implicit citation relationship with the instruction operation of the system administrator as a trigger. Further, for example, the implicit citation relationship discovery system may execute an implicit citation relationship extraction process at predetermined time intervals using a nighttime batch or the like. For example, the implicit citation relationship discovery system may start the extraction process of the implicit citation relationship based on the registration of a new electronic document in the document database 101. Further, for example, the implicit citation relationship discovery system may start the implicit citation relationship extraction process based on a predetermined amount of electronic documents registered in the document database 101.

  First, the controller 100 acquires (extracts) a set D of processing target documents from the document database 101 (step S101).

  The controller 100 repeatedly executes the following steps S103 to S112 for the document di (0 <i <D.count) included in the extracted document set D (step S102). Note that i indicates the document order. “count” indicates the total number of documents.

  In the loop process for the document di, the controller 100 repeatedly executes the following processes in steps S104 to S111 for the document dj (i + 1 ≦ j ≦ D.count) included in the document set D (step S103). Note that j indicates the document order.

  First, the controller 100 passes (outputs) the document di and the document dj to be processed to the order relation estimation unit 201. The order relation estimation means 201 estimates the citation order relation between the document di and the document dj using the access right judgment means 2011 and the time judgment means 2012 (step S104). Then, the estimation result of the order relation is returned (output) to the controller 100.

  The controller 100 receives (inputs) an order relation estimation result from the order relation estimation unit 201. Then, the controller 100 determines whether or not the input estimation result indicates that there is a citation order relationship between the document di and the document dj (step S105). If the estimation result indicates that there is no citation order relationship between the document di and the document dj, the controller 100 jumps (transfers) to step S112 as it is. Then, the loop processing of steps S103 to S112 is repeated.

  If the estimation result indicates that the document di and the document dj have a citation order relationship, the controller 100 passes the document di and the document dj together with the document di and the document dj to the citation degree calculation unit 202. (Output).

  Next, the citation degree calculating unit 202 calculates the time distance between the document di and the document dj using the time distance calculating unit 2022 (step S106). At the same time, the citation level calculation means 202 uses the author distance calculation means 2023 to calculate the author distance between the author of the document di and the author of the document dj (step S107). Further, at the same time, the citation level calculation unit 202 calculates the similarity level of the passages included in the document di and the document dj using the similarity level calculation unit 2021 (step S108).

  In addition, the order which performs the process of step S106-S108 is not ask | required. For example, the citation level calculation means 202 may execute the processing of steps S107 and S108 after executing the time distance calculation processing of step S106, or may execute step S106, after executing the author distance calculation processing of step S107. You may perform the process of S108. Further, the citation level calculation means 202 may execute the processing of steps S106 and S107 after executing the similarity calculation processing of step S108, and the processing of steps S106, S107, and S108 in parallel by time sharing. May be executed.

  Next, the citation degree calculating means 202 obtains the citation degree c using the time distance calculated in step S106, the author distance calculated in step S107, and the passage similarity calculated in step S108. In this case, the citation level calculation unit 202 uses the integrated calculation unit 2024 to calculate the citation level c of the combination of two passages included in the document di and the document dj (step S109). Then, the citation level calculation means 202 passes (outputs) the calculation result of the citation level c to the controller 100.

  Next, the controller 100 receives (inputs) the value of the citation degree c calculated in step S109 from the citation degree calculation means 202. Then, the controller 100 compares the input citation level c with a predefined threshold value, and determines whether or not the value of the citation level c is greater than a predetermined threshold value (step S110).

  When the controller 100 determines that the citation degree c is greater than a predetermined threshold, the controller 100 determines that the combination of passages included in the document di and the document dj has a citation relationship. Then, the controller 100 registers the determination result determined to have a citation relationship in the virtual citation database 104 (step S111).

  Next, the controller 100 adds 1 to the coefficient j indicating the document order (j = j + 1), and proceeds to step S103. That is, the processes of steps S104 to S111 are performed for the next document dj.

  In step S113, the controller 100 adds 1 to the coefficient i indicating the document order (i = i + 1), and proceeds to step S102. That is, the processing of steps S103 to S112 is performed for the next document di.

  As described above, the citation relationship is extracted for the combination of passages included in all electronic documents by repeatedly executing the processes of steps S103 to S112 for all the electronic documents stored in the document database 101. Then, the determination results of all passages determined to have a citation relationship are stored in the virtual citation database 104.

  The determination result stored in the virtual citation database 104 can be provided and used in response to a request from each department in an organization such as a company. For example, the implicit citation relationship discovery system extracts and transmits a citation relationship determination result from the virtual citation database 104 in response to a request from a management department or personnel department terminal in the organization. Then, by displaying the judgment result of the citation relationship received at the terminal of the management department or the personnel department, it can be used for performance evaluation or personnel evaluation of the organization.

  Hereinafter, each step included in the extraction process of the implicit citation relationship illustrated in FIG. 3 will be described.

  (1) Processing for Acquiring Document Set for Processing (Step S101): In step S101, the controller 100 accesses the document database 101 and acquires (extracts) an ID set of documents to be processed from the document database 101. In general, the controller 100 extracts document IDs for all documents stored in the document database 101. The controller 100 can also specify a condition and extract a document ID for a part of documents stored in the document database 101.

  Since document IDs and documents correspond one-to-one, hereinafter, a set of document IDs is referred to as a document set D unless otherwise specified.

  (2) Order relation estimation process (step S104): Citations have an order relation. That is, a document that can be a citation source and a document that can be a citation destination are determined in advance. In this embodiment, the order relation estimation unit 201 introduces a citation direction restriction that is a restriction on the directionality of a document that can be a citation source and a document that can be a citation destination, and determines the citation order relation. The citation direction constraint includes a time condition and an access right condition. The time condition is a condition that the citation source document needs to be created before the citation destination document is created. The access right condition is a condition that the author of the cited document can access the cited document (access right is given).

  The order relation estimation unit 201 can check the time condition by accessing the document database 101, obtaining (extracting) the creation time of the document from the document database 101, and comparing it. The time condition check process is executed by the time order determination unit 2012 of the order relation estimation unit 201.

  The time order determination unit 2012 acquires (extracts) the document creation time from the document database 101 through the order relation estimation unit 201. Then, the time order determination unit 2012 compares the creation times of the extracted documents, and determines a document that can be a citation destination and a document that can be a citation source. The time order determination unit 2012 may access the document database 101 to directly acquire (extract) the document creation time.

  Further, the order relation estimation unit 201 uses the access right determination unit 2011 to determine whether or not the author A can access the document b. The access right determination unit 2011 extracts an access level necessary for the document b from the document database 101 through the order relation estimation unit 201. Further, the access right determination unit 2011 extracts the access level of the author A from the access database 103 through the order relation estimation unit 201. Then, the access right determination unit 2011 compares the access level of the extracted document b with the access level of the author A, and determines whether or not the access right condition is satisfied.

  In this case, the access right determination unit 2011 determines that the author A can access the document b (the access right condition is satisfied) if the access level of the author A is equal to or higher than the access level of the document b. That is, the access right determination unit 2011 determines that the document b can be a citation source. Further, the access right determination unit 2011 determines that the author A cannot access the document b (the access right condition is not satisfied) unless the access level of the author A is equal to or higher than the access level of the document b. That is, the access right determination unit 2011 determines that the document b cannot be a citation source.

  Note that the access right determination unit 2011 may directly access the document database 101 and the access database 103.

  FIG. 4 is a flowchart illustrating an example of the order relation estimation process in step S104. In step S104, the order relation estimation unit 201 extracts the creation time of the document (di or dj) from the document database 101 for the document combination (di, dj) included in the document set acquired in step S101. Further, the order relation estimation unit 201 acquires (extracts) access level information from the access database 103 using the IDs of the authors of the document di and the document dj.

  That is, the order relation estimation unit 201 determines the citation order using the access level information immediately before the creation time of the document (di or dj). Then, the order relation estimation unit 201 determines the citation order of the document di and the document dj according to the following procedure using the citation direction constraint.

  The order relation estimation unit 201 determines whether or not the author of the document di can access the document dj, and determines whether or not the author of the document dj can access the document di. When it is determined that the author of the document dj can access the document di but the author of the document di cannot access the document dj (step S401), the order relation estimation unit 201 determines that the document dj is the cited document, It is determined that the document di is a citation source document (step S402).

  If it is determined that the author of the document di can access the document dj, but the author of the document dj cannot access the document di (step S403), the order relation estimation unit 201 determines that the document di is the cited document. Yes, it is determined that the document dj is a citation source document (step S404).

  When it is determined that the author of the document di can access the document dj and the author of the document dj can access the document di (Y in step S405), the order relation estimation unit 201 determines that the document di and the document dj The order relation between the document di and the document dj is estimated on the basis of the creation time.

  When it is determined that the document di is created before the document dj (step S406), the order relation estimation unit 201 determines that the document di is the citation source document and the document dj is the citation destination document. (Step S407). On the other hand, when it is determined that the document dj is created before the document di (Y in step S408), the order relation estimation unit 201 determines that the document di cites the document dj (step S409). . If it is determined that the creation times of the document di and the document dj are the same (N in step S408), the order relationship estimation unit 201 determines that there is no citation relationship between the two documents di and dj. (Step S410).

  If it is determined that the author of the document di cannot access the document dj and the author of the document dj cannot access the document di (N in step S405), the order relation estimation unit 201 determines that the document di and the document di It is determined that there is no citation relationship with dj (step S410).

  Then, the process proceeds to step S <b> 105, and the order relation estimation unit 201 returns (outputs) an estimation result to the controller 100.

  (3) Citation degree calculation processing (step S106, step S107, step S108, and step S109) using a spatiotemporal constraint condition and citation relationship registration processing (step S110, S111): for document pair (di, dj) Then, the controller 100 uses the citation level calculation means 202 to calculate the citation level in round robin for each passage.

  Since citation is often performed in units of passages in a document, in this embodiment, the implicit citation relationship discovery system determines the citation relationship by calculating the citation level in units of passages. If the quotation level of two passages is high, the implicit quotation relationship discovery system determines that the two passages have a quotation relationship. In this case, there is also a citation relationship between documents including these two passages. Hereinafter, the description will be made with the document d1 as a citation source document and the document d2 as a citation destination document unless otherwise specified.

  The citation degree calculation means 202 calculates the citation degree using the time distance of the document, the author distance, and the similarity between passages. The citation level calculation means 202 calculates the time distance and the author distance in document units. On the other hand, the citation level calculation means 202 calculates the similarity level in units of passages.

  In this embodiment, the time distance is the difference between the creation times of two documents. The author distance is the shortest distance of the path connecting the authors of the document in the organization space, and is a measure indicating the strength of the connection between the authors.

  Basically, the higher the similarity, the higher the possibility of citation between passages, and the higher the citation level. In addition, the shorter the author distance, the higher the possibility of citation between passages and the higher the citation level. In addition, the degree of citation between passages with small or large time distance is small.

  The citation level calculation means 202 calculates the citation level in step S109 using the time distance calculated in step S106, the author distance calculated in step S107, and the passage similarity calculated in step S108. In this case, the citation level calculation unit 202 uses the integrated calculation unit 2024 to calculate the citation level cit between the passage p1 of the document d1 and the passage p2 of the document d2 according to the following equation (1).

  However, in Formula (1), sim is a passage similarity. In the formula (1), timedis is a time distance, and authdis is an author distance.

  If the citation degree cit between the passage p1 of the document d1 and the passage p2 of the document d2 obtained by the citation degree calculation means 202 is larger than a predetermined threshold, the controller 100 determines whether the passage p1 of the document d1 and the passage p2 of the document d2 It is judged that there is a citation relationship between. Then, the controller 100 registers the citation relation determination result in the virtual citation database 104 in a form in which the ID of the document d1, the ID of the passage p1, the ID of the document d2, the ID of the passage p2, and the citation degree cit are associated with each other. Do. The registration process of the citation relation determination result is performed in step S111.

  Hereinafter, a method of calculating passage similarity, time distance, and author distance will be described.

  (3-1) Time distance calculation using time interval conditions (step S106): In general, there is a low possibility that a citation relationship exists in a document created at the same time. In other words, the closer the creation time between the citation source document and the citation destination document is, the higher the possibility that a citation relationship exists. On the other hand, when the difference in creation time between the documents increases, the possibility that the document is read increases, so that the possibility that the document is cited increases.

  However, if the difference in the creation time of the document building becomes extremely large, unless it is a very good document, it will be forgotten, and the possibility of being cited will be reduced. That is, as shown in FIG. 5, as the difference in creation time between documents increases, the citation possibility between documents tends to decrease from a certain point after increasing once.

  In the present embodiment, the time distance calculation unit 2022 obtains a difference in document creation time converted to a unit time distance (normalized) as a time distance. In step S <b> 106, the time distance calculation unit 2022 acquires (extracts) the creation time of each document, the document type, and the conversion parameter of the unit time distance corresponding to the document type from the document database 101 through the controller 100. Then, the time distance calculation unit 2022 calculates the time distance between the document d1 and the document d2 according to the following equation (2) using the extracted information.

timedis (d1, d2) = (| time (d1) −time (d2) |) / μ (d1) (2)

  However, in Expression (2), time (d1) and time (d2) are the creation times of the document d1 and the document d2 in units of time (hour), respectively. Further, μ is a conversion parameter for the unit time distance corresponding to the type of the citation source document. As described above, the unit time distance conversion parameter is set in advance for each document type.

  For example, when the document is a weekly report, the conversion parameter μ can be set to 24 hours (1 day). For example, when the document is an in-house report RN, the conversion parameter can be set to 720 hours (one month). In this way, by converting the unit time distance, the time distance calculation unit 2022 can obtain the time distance in a form that eliminates the effect of the effective time of the document.

  The time distance calculation unit 2022 may access the document database 101 and directly acquire (extract) the document creation time and the conversion parameter μ.

  (3-2) Author distance calculation using organization correlation condition (step S107): Generally, in an organization space, authors with strong connections are close to each other, so there is a high possibility that they are communicating closely. . Therefore, there is a high understanding of the partner's way of thinking and the document created by the partner, and there is a high possibility that the document created by the partner is cited. For example, co-workers, supervisors, and subordinate authors in the same department are likely to cite documents created by the other party.

  In the present embodiment, the author distance calculation unit 2023 calculates the author distance using the organization graph acquired (extracted) from the organization configuration table storage unit 102. For example, in the organization graph shown in FIG. 2A, nodes correspond to employees and branches represent organizational relationships between employees. The author distance calculation means 2023 calculates the author distance according to the following process using the organization graph as shown in FIG.

  The author distance calculation means 2023 acquires (extracts) the organization graph from the organization configuration table storage means 102 through the citation degree calculation means 202 based on the creation time of the cited document d2. That is, the author distance calculation unit 2023 obtains the author distance by acquiring and using the organization graph immediately before the creation time of the document d2. The author distance calculation means 2023 may acquire (extract) information related to the organization graph directly from the organization structure table storage means 102 without passing through the citation degree calculation means 202.

  In step S107, the author distance calculation means 2023 obtains the author distance authdis between the document d1 and the document d2 according to the following equation (3). That is, the author distance calculation means 2023 has the nodes d1. author and node d2. It is calculated as the length of the shortest path connecting the author. Note that the length of the path is counted as the number of branches of the path. Further, when there are a plurality of authors of the document di and the document dj, the author distance calculation means 2023 calculates the author distance for all the combinations.

authdis (d1, d2) = shortestpath (d1.author, d2.author) (3)

  As the calculated author distance is shorter, the author's connection with the document d1 and the document d2 is stronger and the possibility of being in the same space is higher. Therefore, the controller 100 can determine that there is a high possibility of quoting the partner document.

  For example, in the example of the organization graph shown in FIG. 2A, the distance between the “S supervisor” and the “H manager” is 1, and the distance between the “S supervisor” and the “K senior researcher” is two. Therefore, the controller 100 determines that there is a stronger connection between the “S manager” and the “H manager”, and that the “S manager” is more likely to cite the document of the “H manager” than the “K chief researcher” document. .

  The author distance calculation means 2023 can determine the shortest path connecting two authors as the shortest path problem of the graph. For example, the author distance calculation means 2023 can obtain the shortest path connecting the authors using the Dijkstra algorithm. A method for obtaining the shortest path using Dijkstra's algorithm is described in, for example, Document B “Kei Ishihata,“ Algorithm and Data Structure ”, Iwanami Shoten, pp. 260-270”.

  (3-3) Similarity calculation (step S108): The similarity calculation means 2021 can calculate the similarity using, for example, the following equation (4). Expression (4) is an expression for calculating the cosine of the keyword vector based on the vector space model (see Document A).

  However, in Equation (4), the keyword vector for the passage p1 is (x1, x2,..., Xn), and the keyword vector for the passage p2 is (y1, y12,..., Ym).

  The similarity calculation unit 2021 extracts the texts of the passages p1 and p2 from the document database 101 through the citation level calculation unit 202, respectively. The similarity calculation unit 2021 may acquire (extract) the texts of the passages p1 and p2 directly from the document database 101 without passing through the citation level calculation unit 202. If the similarity between the passages is high, there is a high possibility that there is a citation relationship between the passages. In the case of content other than a document, the similarity calculation unit 2021 prepares a corresponding similarity calculation formula and performs similarity calculation.

  As described above, according to the present embodiment, the citation relationship between contents is extracted in consideration of the difference in creation time between contents and the degree of relationship between authors who created the content. False detection can be excluded. Accordingly, it is possible to extract the implicit citation relationship in the content and improve the accuracy of the implicit citation relationship extraction.

  In other words, according to the present embodiment, there is provided an order relation estimation means using citation direction constraints, and a citation degree calculation means using similarity, time distance and author distance based on spatiotemporal constraint conditions. Since the citation relationship can be extracted based on the citation level, it is possible to extract even an implicit citation relationship that is not explicitly specified in the content. In addition, since extraction can be performed in consideration of time distance and author distance, it is possible to exclude false detection of a citation relationship even though there is no citation relationship by using a similarity-based method. . That is, the accuracy of citation relationship detection can be improved.

  By referring to the virtual citation database and the document database of the implicit citation relationship extraction system constructed by the method shown in the present embodiment, the following usage method becomes possible. For example, an in-house document can be displayed with reference to a document ID or file path in the document database, and a document related to the displayed document can be displayed. In addition, it is possible to automatically generate hyperlinks between passages in a document and refer to each other. If the citation relationship is regarded in the same way as a Web link, the ranking of important documents can be performed using the link relationship in the same manner as the Web search.

  Furthermore, it is possible to evaluate performance by determining novelty (novelty) and authority (non-quoting) in citation-related passage units, and identifying authors who have created highly original documents.

  Note that novelty, authority, and originality can be obtained using, for example, the following equation (5).

O (p) = A (p) · N (p)
N (p) = 1 / (refin (p) +1)
A (p) = refout (p)
... Formula (5)

  Here, in Equation (5), O (p) is the originality of the passage p, N (p) is the novelty of the passage p, and A (p) is the authority of the passage p. Further, refout (p) is the number of passages that quote the passage p, and refin (p) is the number of passages that the passage p cites. For example, by obtaining the originality of the passage in the document for each author and obtaining the average value, the average originality of the document created by the author can be obtained.

Embodiment 2. FIG.
Next, a second embodiment of the present invention will be described with reference to the drawings. FIG. 6 is a block diagram illustrating a configuration example of the implicit citation relationship discovery system in the second embodiment. As shown in FIG. 6, in this embodiment, the implicit citation relationship discovery system is different from the first embodiment in that it includes a document registration monitoring unit 301 in addition to the components shown in the first embodiment. . Further, the present embodiment is different from the first embodiment in that the order relation estimation unit 201 does not include the time order determination unit 2012 among the components shown in the first embodiment.

  That is, in this embodiment, the implicit citation relationship discovery system includes the controller 100, the document database 101, the organization configuration table storage unit 102, the access database 103, the virtual citation database 104, the order relationship estimation unit 201, and the citation. Degree calculation means 202 and document monitoring means 301 are included. In the following, the same components as those in the first embodiment are denoted by the same reference numerals as those in the block diagram shown in FIG.

  Specifically, the document registration monitoring unit 301 is realized by a CPU of an information processing apparatus that operates according to a program. The document monitoring unit 301 has a function of monitoring registration of a new document in the document database 101.

  In the present embodiment, the controller 100, with respect to a new document detected by the document monitoring unit 301, for each document stored in the document database 101, a document registered before the detected new document (registered document). The citation relationship with is determined. In this case, the controller 100 determines the citation relationship between the new document and the registered document by using the order relationship estimation unit 201 and the citation degree calculation unit 202. Then, the controller 100 registers the citation relation determination result in the virtual citation database 104.

  The order relation estimation unit 201 includes an access right determination unit 2011. The order relation estimation means 201 estimates the order relation between the citation source and the citation destination of a document pair with a high citation degree based on the citation direction constraint. That is, the order relation estimation means 201 determines the citation order relation between the new document detected by the document monitoring means 301 and the registered document registered before the new document acquired (extracted) from the document database 101 through the controller 100. presume. In this case, the order relation estimation unit 201 uses the access right determination unit 2011 to estimate based on the access right condition that the access right to the citation source document is necessary. That is, the order relation estimation unit 201 estimates and determines the citation source document and the citation destination document.

  The citation degree calculation means 202 calculates a citation degree indicating the possibility of passage citation.

  The controller 100 uses the order relation estimation means 201 and the citation degree calculation means 202 to extract the citation relation between passages and store it in the virtual citation database 104.

  Next, the operation will be described. FIG. 7 is a flowchart illustrating an example of a process for extracting an implicit citation relationship included between contents (passages in an electronic document) by the implicit citation relationship discovery system according to the second embodiment. In the present embodiment, detailed description of steps for performing the same processing as in the first embodiment is omitted.

  In this embodiment, the document registration monitoring unit 301 repeatedly monitors registration of new documents in the document database 101. For example, the document registration monitoring unit 301 determines whether or not a new document is registered in the document database 101 every predetermined time (step S200). If it is determined that a new document has been registered, the document registration monitoring unit 301 notifies the controller 100 that the new document has been registered (for example, outputs notification information). If no new document is registered, the document registration monitoring unit 301 continues the monitoring process in step S200.

  The controller 100 starts a citation relation extraction process for the new document d detected by the document registration monitoring unit 301 (step S201). The controller 100 acquires (extracts) a set D of registered documents created before the registration time of the new document detected by the document registration monitoring unit 301 from the document database 101 (step S202). Then, the controller 100 performs the following processing for each new document d detected in step S200.

  The controller 100 repeatedly executes the following processes of steps S204 to S211 for the document di (0 <i <D.count) included in the extracted document set D (step S203). Note that i indicates the document order. “count” indicates the total number of documents.

  First, the controller 100 passes (outputs) the document d and the document di to be processed to the order relation estimation unit 201. The order relation estimation unit 201 uses the access right judgment unit 2011 to estimate the order relation of the citation between the document d and the document di (step S204). Then, the estimation result of the order relation is returned (output) to the controller 100.

  The controller 100 receives (inputs) an order relation estimation result from the order relation estimation unit 201. Then, the controller 100 determines whether or not the input estimation result indicates that the document d and the document di have a citation order relationship (step S205). If the estimation result indicates that there is no citation order relationship between the document d and the document di, the controller 100 jumps (transfers) to step S212 as it is. Then, the loop processing of steps S204 to S212 is repeated.

  If the estimation result indicates that there is a citation order relationship between the document d and the document di, the controller 100 passes the citation order relationship between the document d and the document di to the citation degree calculation unit 202 together with the document d and the document di. (Output).

  Next, the citation degree calculation unit 202 calculates the time distance between the document d and the document di using the time distance calculation unit 2022 (step S206). At the same time, the citation level calculation unit 202 calculates the author distance between the author of the document d and the author of the document di using the author distance calculation unit 2023 (step S207). Further, at the same time, the citation level calculation unit 202 uses the similarity level calculation unit 2021 to calculate the similarity level of the passages included in the document d and the document di (step S208).

  Next, the citation degree calculating means 202 obtains the citation degree c using the time distance calculated in step S206, the author distance calculated in step S207, and the passage similarity calculated in step S208. In this case, the citation level calculation unit 202 uses the integrated calculation unit 2024 to calculate the citation level c of the combination of the two passages included in the document d and the document di (step S209). Then, the citation level calculation means 202 passes (outputs) the calculation result of the citation level c to the controller 100.

  Next, the controller 100 receives (inputs) the value of the citation degree c calculated in step S209 from the citation degree calculation means 202. Then, the controller 100 compares the input citation level c with a predefined threshold value, and determines whether or not the value of the citation level c is greater than a predetermined threshold value (step S210).

  If the controller 100 determines that the citation degree c is greater than a predetermined threshold, the controller 100 determines that there is a citation relationship in the combination of passages included in the document d and the document di. Then, the controller 100 registers the determination result determined to have a citation relationship in the virtual citation database 104 (step S211).

  Next, the controller 100 adds 1 to the coefficient i indicating the document order (i = i + 1), and proceeds to step S203. That is, the processing of steps S204 to S211 is performed for the next document di.

  Thereafter, when the document registration monitoring unit 301 detects the next new document d (step S213), the controller 100 performs the same processing as steps S201 to S212 on the next new document d.

  Hereinafter, the processes of steps S200, S202, and S204 that perform processes different from those of the first embodiment will be described. First, in step S200, the document registration monitoring unit 301 monitors the document database 101 to monitor whether a new document is registered. When the registration of the new document is detected, the document registration monitoring unit 301 notifies the controller 100 of the newly registered new document group (for example, outputs notification information). Then, the controller 100 extracts a citation relationship between the new document group and the already registered documents previously registered in the document database 101. In other words, in the present embodiment, the implicit citation relationship discovery system performs citation relationship discovery processing with the registration of a new document as a trigger. Detection of new document registration by the document registration monitoring unit 301 plays a role of this trigger.

  In step S202, the controller 100 accesses the document database 101, and acquires (extracts) from the document database 101 a set of registered documents registered before the new document detected in step S200. In this embodiment, the registration time in the document database 101 indicates the document creation time. Therefore, the controller 100 can acquire (extract) a registered document set registered before the new document by checking the creation time of the document stored in the document database 101.

  FIG. 8 is a flowchart illustrating an example of the order relationship estimation processing in step S204 in the second embodiment. In this embodiment, since it is known that the document di included in the document set D is created before the new document d, the order relation estimation unit 201 does not need to perform the time condition inspection process. Therefore, the order relation estimation unit 201 uses the access right determination unit 2011 to show the following based on author information acquired (extracted) from the document database 101 and access right conditions acquired (extracted) from the access database 103. The citation order of document d and document di is determined according to the procedure. Note that the access right determination unit 2011 may directly access the document database 101 and the access database 103 without passing through the order relation estimation unit 201.

  The order relation estimation unit 201 determines whether or not the author of the document di can access the document d, and determines whether or not the author of the document d can access the document di. When it is determined that the author of the document d can access the document di but the author of the document di cannot access the document d (step S451), the order relation estimation unit 201 determines that the document d is a cited document, It is determined that the document di is a citation source document (step S452).

  If it is determined that the author of the document di can access the document d, but the author of the document d cannot access the document di (step S453), the order relation estimation unit 201 determines that the document di is the cited document. Yes, it is determined that the document d is a citation source document (step S454).

  If it is determined that the author of the document d can access the document di and the author of the document di can access the document d (Y in step S455), the order relation estimation unit 201 determines that the document d is the citation destination. It is determined that the document di is the citation source document (step S456).

  When it is determined that the author of the document d cannot access the document di and the author of the document di cannot access the document d (N in step S455), the order relation estimation unit 201 determines that the document di and the document di It is determined that d does not have a citation relationship (step S457).

  Then, the process proceeds to step S <b> 205, and the order relation estimation unit 201 passes (outputs) the estimation result to the controller 100.

  As described above, according to the present embodiment, the order relation estimation unit 201 obtains and processes only registered documents registered before the new document detected by the document registration monitoring unit 301. Citation relationships can be estimated. Therefore, in the process of estimating the citation order relationship, the time condition inspection process can be omitted, and the processing load can be reduced. Further, according to the present embodiment, it is only necessary to perform processing for registered documents, so the number of documents to be processed can be reduced.

  Next, the minimum configuration of the citation relationship extraction system according to the present invention will be described. FIG. 9 is a block diagram illustrating a minimum configuration example of the citation relationship extraction system. As shown in FIG. 9, the citation relationship extraction system includes a document database 101 and a citation degree calculation unit 202 as minimum components.

  The citation level calculation means 202 indicates the possibility that citations have been made between contents based on the difference in creation, update or reference time between contents and the degree of relationship between authors who created, updated or referred to the contents. A function for calculating the degree of citation indicating the degree is provided. In addition, the controller 100 has a function of extracting a citation relationship between contents based on the citation degree calculated by the citation degree calculating unit 202.

  According to the citation relation extraction system with the minimum configuration shown in FIG. 9, the difference between creation, update or reference time between contents and the degree of relation between authors who created, updated or referred to the contents are considered. Since the citation relationship is extracted, erroneous detection of the citation relationship can be excluded. Accordingly, it is possible to extract the implicit citation relationship in the content and improve the accuracy of the implicit citation relationship extraction.

  In each of the above embodiments, the characteristic configuration of the citation relationship extraction system (implicit citation relationship discovery system) as shown in the following (1) to (8) is shown.

  (1) In the citation relationship extraction system, citations can be made between contents based on the difference in creation, update or reference time between contents and the degree of relationship between authors who created, updated or referenced the contents. The citation relationship between the contents is extracted based on the citation degree calculated by the citation degree calculating means (for example, realized by the citation degree calculating means 202) and the citation degree calculated by the citation degree calculating means. Citation relationship extraction means (for example, realized by the controller 100) is provided.

  (2) The citation relation extraction system includes order relation estimation means (for example, realized by the order relation estimation means 201) for estimating the order relation between content that can be a citation source and content that can be a citation destination. The means may be configured to extract a citation relation between contents based on the citation degree calculated by the citation degree calculation means and the order relation estimated by the order relation estimation means.

  (3) In the citation relationship extraction system, the citation degree calculating means calculates a time distance indicating a difference in creation, update or reference time between contents (for example, realized by the time distance calculating means 2022). And a calculation unit (for example, realized by the total calculation unit 2024) that calculates the citation degree based on the time distance calculated by the time distance calculation unit.

  (4) In the citation relation extraction system, the citation degree calculation means is realized by an author distance calculation means (for example, an author distance calculation means 2023) that calculates an author distance indicating a degree of relation between authors who created, updated, or referred to content. And a calculation means (for example, realized by the total calculation means 2024) for calculating the citation degree based on the author distance calculated by the author distance calculation means.

  (5) In the citation relation extraction system, the citation degree calculating means includes a similarity degree calculating means (for example, realized by the similarity degree calculating means 2021) for calculating a similarity degree between contents, and the calculating means calculates a time distance. The citation degree may be calculated by integrating the time distance calculated by the means, the author distance calculated by the author distance calculation means, and the similarity calculated by the similarity calculation means.

  (6) In the citation relationship extraction system, the order relationship estimation means can be a content that can be a citation source and a citation destination based on the level of access right set for the content and the level of access right set for the author. Including an access right order estimating unit (e.g., realized by the access right determining unit 2011) for estimating the order relation with the content, and the access right order estimating unit is configured such that the level of the access right set for the author is set for the content. If it is determined that the access right level or higher, the content may be estimated as content that can be cited.

  (7) In the citation relation extraction system, the order relation estimation means estimates time order estimation means for estimating the order relation between content that can be a citation source and content that can be a citation destination based on the creation, update, or reference time of the content. For example, the time order estimation means estimates the content that has been created, updated, or referred to as the content that can be cited, and the content that has the new creation, update, or reference time. The content may be estimated as a content that can be cited.

  (8) In the citation relationship extraction system, the time distance calculation means is a normalization parameter (for example, for converting to a unit time distance) for normalizing a difference in content creation, update or reference time according to the content type. The parameter may be used to calculate a normalized time distance.

  The present invention can be applied to uses such as an information processing apparatus that organizes in-house content, an in-house content search apparatus, and an in-house performance evaluation support apparatus. It can also be applied to uses such as a device for discovering the reuse relationship of content, clarifying the original and the source, and supporting copyright protection.

It is a block diagram which shows an example of a structure of the implicit citation relationship discovery system by this invention. It is explanatory drawing which shows the example of an organization structure graph, the adjacency matrix with respect to the organization structure graph, and author information. It is a flowchart which shows an example of the process which an implicit citation relationship discovery system extracts the implicit citation relationship contained between contents (passage in an electronic document). It is a flowchart which shows an example of the estimation process of order relation. It is explanatory drawing which shows the relationship between the possibility of quotation, and time distance. It is a block diagram which shows the structural example of the implicit citation relationship discovery system in 2nd Embodiment. It is a flowchart which shows an example of the process which the implicit citation relationship discovery system in 2nd Embodiment extracts the implicit citation relationship contained between content (passage in an electronic document). It is a flowchart which shows an example of the estimation process of the order relationship in 2nd Embodiment. It is a block diagram which shows the minimum structural example of a quotation relation extraction system.

Explanation of symbols

100 controller 101 document database 102 organization structure table storage means 103 access database 104 virtual citation database 201 order relation estimation means 2011 access right judgment means 2012 time order judgment means 202 citation degree calculation means 2021 similarity calculation means 2022 time distance calculation means 2023 Author distance calculation means 2024 Integrated calculation means 301 Document registration monitoring means

Claims (18)

Based on the difference in creation, update, or reference time between contents and the degree of relationship between authors who created, updated, or referenced the contents, the degree of citation that indicates the likelihood of citations between the contents was calculated A citation level calculating means to
A citation relationship extraction system comprising: citation relationship extraction means for extracting a citation relationship between contents based on the citation level calculated by the citation level calculation means. An order relationship estimating means for estimating an order relationship between content that can be cited and content that can be cited;
The citation relationship extraction system according to claim 1, wherein the citation relationship extraction unit extracts a citation relationship between contents based on the citation degree calculated by the citation degree calculation unit and the order relationship estimated by the order relationship estimation unit. Citation level calculation means
A time distance calculating means for calculating a time distance indicating a difference in creation, update or reference time between contents;
The citation relationship extraction system according to claim 1, further comprising a calculation unit that calculates a citation degree based on the time distance calculated by the time distance calculation unit. Citation level calculation means
Author distance calculation means for calculating the author distance indicating the degree of relationship between the authors who created, updated or referred to the content;
The citation relation extraction system according to any one of claims 1 to 3, further comprising a calculation unit that calculates a citation degree based on the author distance calculated by the author distance calculation unit. The citation degree calculating means includes a similarity calculating means for calculating the similarity between contents,
The calculation means calculates a citation degree obtained by integrating the time distance calculated by the time distance calculation means, the author distance calculated by the author distance calculation means, and the similarity calculated by the similarity calculation means. Item 5. The citation relationship extraction system according to item 4. The order relation estimation means estimates the order relation between the content that can be cited and the content that can be cited based on the level of access right set for the content and the level of access right set for the author. A right order estimation means,
3. The access right order estimating unit estimates the content as a content that can be cited from the content when the access right level set for the author is determined to be equal to or higher than the access right level set for the content. Citation relationship extraction system. The order relation estimation means includes time order estimation means for estimating an order relation between content that can be a citation source and content that can be a citation destination based on the creation, update, or reference time of the content,
The time order estimation means estimates content that has an old creation, update, or reference time as content that can be cited, and estimates content that has a new creation, update, or reference time as content that can be cited. 6. The citation relationship extraction system according to 6.   The time distance calculation means calculates a normalized time distance using a normalization parameter for normalizing a difference in content creation, update or reference time according to the content type. The quotation relation extraction system of any one of them. Based on the difference in creation / update / reference time between contents and the degree of relationship between authors who created / updated / referenced the content, the degree of citation indicating the possibility of citation between the contents was calculated. Quoting level calculating step,
A citation relationship extraction method, comprising: a citation relationship extraction step for extracting a citation relationship between contents based on the calculated citation degree. Including an order relation estimation step for estimating an order relation between content that can be cited and content that can be cited;
The citation relationship extraction method according to claim 9, wherein in the citation relationship extraction step, a citation relationship between contents is extracted based on the calculated citation degree and the estimated order relationship. In the citation level calculation step,
Calculate the time distance indicating the difference in creation, update or reference time between content,
The citation relationship extraction method according to claim 9 or 10, wherein a citation degree is calculated based on the calculated time distance. In the citation level calculation step,
Calculate the author distance that indicates the degree of relationship between authors who created, updated, or referenced content,
The citation relationship extraction method according to any one of claims 9 to 11, wherein a citation degree is calculated based on the calculated author distance. In the citation level calculation step,
Calculate the similarity between content,
The citation relationship extraction method according to claim 11, wherein a citation degree is calculated by integrating the calculated time distance, author distance, and similarity.   In the order relation estimation step, the access right level set for the content based on the access right level set for the content and the access right level set for the author. 11. The citation relationship extraction method according to claim 10, wherein if it is determined that the content is a content that can be a citation source, the order relationship between the content that can be a citation source and the content that can be a citation destination is estimated.   In the order relation estimation step, based on the content creation, update, or reference time, the content with the old creation, update, or reference time is estimated as the content that can be cited, and the content with the new creation, update, or reference time becomes the reference destination The citation relationship extraction method according to claim 10 or 14, wherein the order relationship between the content that can be a citation source and the content that can be a citation destination is estimated by estimating the content as a possible content.   The normalized time distance is calculated using a normalization parameter for normalizing a difference in creation, update, or reference time of content according to the content type in the quoting degree calculation step. The citation relation extraction method according to any one of the above. On the computer,
Based on the difference in creation, update, or reference time between contents and the degree of relationship between authors who created, updated, or referenced the contents, the degree of citation that indicates the likelihood of citations between the contents was calculated Citation level calculation process to
A citation relationship extraction program for executing a citation relationship extraction process for extracting a citation relationship between contents based on the calculated citation level. On the computer,
Execute an order relationship estimation process that estimates the order relationship between content that can be cited and content that can be cited.
The citation relationship extraction program according to claim 17, wherein the citation relationship extraction process executes a process of extracting a citation relationship between contents based on the calculated citation degree and the estimated order relationship.

Images