JP2008181331A - Relation extraction method, relation extraction system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To obtain strength of a noticed relation of nodes from a graph structure data. <P>SOLUTION: In a relation extraction system 1, a relation extraction device 10 connected to a graph structure database 30, which stores data of graph structure composed of various nodes and various arcs, inputs a plurality of nodes desired to obtain strength of the relation, and obtains a coefficient for each of values of input node from a query graph pattern database 12 which stores many query graph patterns by using the number of nodes of a same concept which overlaps between subgraphs extracted from graph structure database 30 by using a query issuance section 11 which searches the query graph pattern relevant to an input node and a searched query graph pattern, and by using the number of nodes of same concept which overlaps between subgraphs extracted from the graph structure database 30. Thereby, the strength of the relation between input nodes can be obtained. Also the strength of the relation between nodes input can be obtained from a plurality of viewpoints by using a plurality of query graph patterns. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a graph mining technique for obtaining the strength of a relationship between nodes from data expressed in a graph structure.

  In recent years, development of a database system that expresses data having various structures as a graph and executes a search by matching a graph pattern from the data to obtain a search result has been progressing. Non-Patent Document 1 is a specification for searching RDF (Resource Description Framework) data by graph pattern matching. In the system based on the specification shown in Non-Patent Document 1, data having a matching structure is extracted from a large amount of data having a graph structure composed of various nodes and various arcs by using a query of the graph structure. , The matched part can be obtained as a subgraph. Patent Document 1 describes a technique for generating a query graph pattern having a graph structure based on information such as keywords and concepts. A query graph pattern is a graph for extracting data whose structure matches from data having a graph structure.

In addition, research on techniques such as graph mining that discovers characteristic relationships from data expressed as graphs is progressing. Non-Patent Document 2 describes a technique for obtaining a coefficient between a plurality of matters by obtaining an overlap of search results obtained by performing an information search on the Internet using keywords relating to a plurality of matters for which a characteristic relationship is desired. ing.
JP 2006-313501 A "SPARQL Query Language for RDF", [online], World Wide Web Consortium, [Search January 5, 2007], Internet <URL: http: // www. w3c. org / TR / rdf-sparql-query /> Yutaka Matsuo, 5 others, “Researcher Network Extraction Search System”, [online], Japanese Society for Artificial Intelligence, [Search January 5, 2007], Internet <URL: http: // www-kasm. nii. ac. jp / jsai2005 / schedule / pdf / 000137. pdf>

  However, Non-Patent Document 1 describes extracting a subgraph that matches a query graph pattern, but does not describe finding a characteristic relationship from the extraction result.

  The technique described in Non-Patent Document 2 uses information existing on the Internet as a search target for a plurality of matters for which a relationship is desired, and has a graph structure composed of various nodes and various arcs. It is not assumed that data having a search target is to be searched, and the method is not described.

  The present invention has been made in view of the above, and the problem is that a graph structure data composed of various nodes and various arcs can be used to determine the strength of the relationship between the nodes in question from a plurality of viewpoints. There is to ask from.

  In the relation extraction method according to the first aspect of the present invention, a plurality of nodes among the nodes stored in the graph structure database in which data having a graph structure composed of various nodes and various arcs is stored by the input means. Accepting the selection of the node and assigning the selected node to the variable node of the query graph pattern for extracting the data whose structure matches from the data having the graph structure for each selected node by the query issuing means The number of overlapping nodes by comparing the data acquired for each of the nodes selected in the step of acquiring data matching the query graph pattern from the graph structure database and the step of acquiring data by the coefficient calculation means And calculate the coefficient between the selected nodes based on the overlap number. Characterized in that it has Tsu and up, the.

  In the present invention, selection of a plurality of nodes from data having a graph structure composed of various nodes and various arcs is accepted, and each of the selected nodes is selected as a variable node of a query graph pattern. The data that matches the query graph pattern assigned to the selected node is extracted, and the coefficient is calculated based on the number of nodes that overlap between the extracted data for each selected node. Can be determined.

  In the above relationship extraction method, the step of acquiring data is characterized in that a query graph pattern including a variable node corresponding to the selected node is acquired from a query graph pattern database storing the query graph pattern.

  In the present invention, by acquiring a query graph pattern to be used from a query graph pattern storing a query graph pattern, various query graph patterns can be quickly acquired, and nodes selected from various viewpoints can be obtained. The strength of the relationship can be obtained.

  In the relation extraction method, the step of acquiring data generates a query graph pattern including a variable node corresponding to the selected node based on the selected node.

  In the present invention, since the query graph pattern to be used is generated based on the selected node, the flexible query graph pattern according to the selected node can be used.

  In the relationship extracting method, the accepting step inputs a query graph pattern including a variable node corresponding to the selected node.

  In the present invention, by inputting a query graph pattern to be used, it is possible to obtain the strength of the relationship between nodes from a desired viewpoint.

  The relation extracting method includes a step of displaying a relation between nodes selected based on the coefficient calculated in the calculating step as a graph structure.

  In the present invention, by displaying the strength of the relationship between nodes as a graph structure based on the calculated coefficient, the user can easily grasp the relationship between the selected nodes visually. .

  The relation extraction system according to the second aspect of the present invention is connected to a graph structure database in which data having a graph structure composed of various nodes and various arcs is stored, and among the nodes stored in the graph structure database. Input means for accepting selection of a plurality of nodes, and for each of the selected nodes, the selected node is substituted into a variable node of a query graph pattern for extracting data whose structure matches from data having a graph structure The query issuing means for acquiring data matching the query graph pattern from the graph structure database and the query issuing means for comparing the data acquired for each of the selected nodes to obtain the number of overlapping nodes, Coefficient calculating means for calculating a coefficient between nodes selected based on The features.

  The relation extraction system has a query graph pattern database storing a query graph pattern, and the query issuing means acquires a query graph pattern including a variable node corresponding to the selected node from the query graph pattern database. It is characterized by.

  In the relationship extraction system, the query issuing unit generates a query graph pattern including a variable node corresponding to the selected node based on the selected node.

  In the relationship extraction system, the input unit inputs a query graph pattern including a variable node corresponding to the selected node.

  The relation extraction system includes display means for displaying a relation between nodes selected based on the coefficient calculated by the coefficient calculation means as a graph structure.

  According to the present invention, it is possible to obtain the strength of the relationship between nodes of interest from a plurality of viewpoints from graph structure data composed of various nodes and various arcs.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 is a block diagram showing a configuration of a relationship extraction system 1 using a relationship extraction device 10 in the present embodiment. The relationship extraction system 1 shown in FIG. 1 includes a relationship extraction device 10, a user interface providing unit 20, and a graph structure database 30. The relationship extraction device 10 includes a user interface providing unit 20, a graph structure database 30. Is connected in an accessible state.

  The relationship extraction apparatus 10 includes a query issuing unit 11 that searches or generates a query graph pattern based on the node selected by the user interface providing unit 20, a query graph pattern database 12 that stores the query graph pattern, a query graph And a coefficient calculation unit 13 for calculating a coefficient representing the relationship from the search result using the pattern. The relationship extraction device 10 can be configured by a computer including an arithmetic processing device, a storage device, a memory, and the like, and the processing of each unit is executed by a program.

  The user interface providing unit 20 provides a GUI capable of selecting a node for which a relationship is desired, and displays the relationship of the selected node.

  The graph structure database 30 stores data having a graph structure composed of various nodes and various arcs as shown in FIG. In the figure, the label name of the arc for which nothing is displayed is “technical keyword”, and a part of the name space is also omitted. The graph structure data in the present embodiment is labeled effective graph data, and RDF defines a data model of data that can be expressed by a labeled effective graph.

  FIG. 3 is an explanatory diagram showing whether the graph structure data is generated from data managed in an existing relational database or the like and graphed. This figure shows a part of the data in FIG. 2, “Paper G”, its author “Taro Yamada”, its title “Introduction to B Technology”, and its keyword “B Technology”. Is represented by RDF / XML format data and an RDF graph. Here, the arc of the graph is called a property based on the RDF specification.

  Next, the operation of the relationship extraction system 1 configured as described above will be described with reference to the drawings.

  FIG. 4 is a sequence diagram showing the operation of the relationship extraction system 1 in the present embodiment. First, the user interface providing unit 20 generates a user operation GUI (S1). In the present embodiment, a GUI that prompts input of a keyword or the like that can identify the selected node is generated.

  Then, the generated GUI is presented to the user, and the user selects a plurality of nodes for which a relationship is desired (S2). The node selected by the user and the URI (Uniform Resource Identifier) which is the ID of the node are transmitted to the relationship extraction apparatus 10. Further, the search instruction content may be transmitted to the relationship extraction device 10. The search instruction content may directly specify a query graph pattern to be described later, or may be information useful for generating a query graph pattern. In the present embodiment, the relationship extraction system 1 that prompts the user to input and selects a node has been described. However, another device connected to the relationship extraction device 10 without the user interface providing unit 20 is provided. , A node may be selected and input without going through the user.

  The query issuing unit 11 searches for a query graph pattern from the query graph pattern database 12 based on the received input keyword value such as keyword or URI, search instruction content, and the like (S3). A large number of query graph patterns are stored in the query graph pattern database 12, and the query issuing unit 11 searches for query graph patterns related to input values. For example, it is assumed that specific technical names “A technology”, “B technology”, “C technology”, and “D technology” are input as keywords. In this case, a query graph pattern having a variable node “? Keyword” that can take a technical name from the query graph pattern database 12 is retrieved and read out.

  As an example, FIG. 5 and FIG. 6 show a query graph pattern having a variable node “? Keyword” that can take a technical name. FIG. 5 includes two variable nodes “? X” and “? Y” indicating the variable node “? Keyword” with the property “rm: technical keyword”, and the two variable nodes “? X” and “? Y”. Each indicates a variable node “? Person” with a property “rm: author”, and the variable node “? Person” has a property “rdf: type” that leads to the node “Person: person”. It is.

  FIG. 6 is a query graph pattern in which the variable node “? X” indicates the variable node “? Person” with the property “pj: person in charge” with respect to what is shown in FIG.

  In the present embodiment, the query graph pattern is extracted from the query graph patterns stored in the query graph pattern database 12 based on the input value of the user. For example, the query graph pattern database 12 is not provided and the patent graph A query graph pattern may be generated by analyzing data stored in the graph structure database 30 based on a user input value using the method shown in Document 1.

  Subsequently, the query issuing unit 11 requests the graph structure database 30 to return data (subgraph) that matches the pattern, using the query node obtained by substituting the input value for the variable node of the obtained query graph pattern as a query. (S4). For example, if the keywords “A technology”, “B technology”, “C technology”, “D technology” are input and the query graph pattern A shown in FIG. 5 is obtained, the variable node “? Four queries are created by substituting “A technology”, “B technology”, “C technology”, and “D technology” in the “keyword” portion.

  Then, the graph structure database 30 returns data that matches the query (S5). FIG. 7 is obtained by using a query graph pattern in which “technique B” is substituted for the variable node “? Keyword” of the query graph pattern A shown in FIG. 5 with respect to the graph structure shown in FIG. FIG. In FIG. 7, three structures that match the query are shown, and the values of the subgraph nodes corresponding to the variable nodes “? X”, “? Y”, “? Person” of the query graph pattern A are shown. Can be extracted. For example, there are three values corresponding to the variable node “? Person” in the pattern matching the above query: “Person: Taro Yamada”, “Person: Ichiro Tanaka”, “Person: Hanako Suzuki”.

  Subsequently, the result obtained by using the query is passed to the coefficient calculation unit 13 (S6), and the coefficient calculation unit 13 calculates the strength of the relationship between the input nodes based on those results ( S7). In the present embodiment, the coefficient is obtained by calculating the number of nodes of the same concept that overlap between data obtained by using a query in which the value of each input node is substituted. The data passed to the coefficient calculation unit 13 is the obtained graph structure data, the value of the variable node of the query graph pattern, or the like.

  As an example, a case where a value corresponding to the variable node “? Person” of the query graph pattern A is referred to when a coefficient is obtained is shown. The values corresponding to the variable node “? Person” obtained by substituting “A technology” are “Person: Sachiko Yamamoto”, “Person: Jiro Nakamura”, “Person: Taro Yamada”, and “B technology”. As described above, the values corresponding to the variable node “? Person” obtained by substituting are “Person: Taro Yamada”, “Person: Ichiro Tanaka”, and “Person: Hanako Suzuki”. The values corresponding to the variable node “? Person” obtained by substituting “C technology” are “Person: Taro Yamada” and “Person: Ichiro Tanaka”, and obtained by substituting “D technology”. The values corresponding to the variable node “? Person” are “Person: Ichiro Tanaka” and “Person: Hanako Suzuki”.

  Based on these results, when the value corresponding to the variable node “? Person” that is duplicated between the input nodes is obtained, “Person: Taro Yamada” is duplicated for “A technology” and “B technology”. “Person: Taro Yamada” overlaps for “A technology” and “C technology”, and there is no overlapping value for “A technology” and “D technology”. In addition, “Person: Taro Yamada” and “Person: Ichiro Tanaka” overlap for “B Technology” and “C Technology”, and “Person: Ichiro Tanaka” for “B Technology” and “D Technology”. “Person: Hanako Suzuki” overlaps, and “C Technology” and “D Technology” overlap “Ichiro Tanaka”.

When the number of nodes that overlap between the input nodes is obtained, the strength of the relationship is obtained using the Simpson coefficient expressed by the following equation (1).

  For example, when the coefficients of “A technology” and “B technology” are obtained, duplicate values among the values corresponding to the variable node “? Person” of the query result obtained by assigning each of “A technology” and “B technology” The number is 1, the number of values corresponding to the variable node “? Person” of the query result to which “A technology” is substituted is 3, and the variable node “? Person” of the query result to which “B technology” is substituted. Since the number of values corresponding to “3” is 3, the Simpson coefficient is 1 / min (3, 3) = 0.33. FIG. 8 is a diagram showing a table in which coefficients between input nodes are similarly obtained.

  FIG. 9 is obtained by substituting “A technology”, “B technology”, “C technology”, and “D technology” into the variable node “? Keyword” of the query graph pattern B shown in FIG. It is the figure which calculated | required the coefficient by the process shown to S7 from the result, and made it a table | surface.

  As shown in FIGS. 8 and 9, the results using the query graph pattern A are different from those using the query graph pattern B. This is because the graph pattern is different. When the relationship extraction system 1 presents the obtained coefficient to the user, if the query graph pattern is also presented, the user can determine from which point of view the coefficient is obtained. Judgment can be made. Query graph pattern A is the viewpoint of authors of two papers that use technology as a keyword, and query graph pattern B is the author of a paper that uses technology as a keyword and is also in charge of a project that uses technology as a keyword. It is a point of view.

  The obtained coefficient is sent to the user interface providing unit 20, and a display result screen is generated by the user interface providing unit 20 and presented to the user (S8). The image is displayed in a graph as shown in FIGS. As a result, the user can easily visually grasp the relationship between the designated nodes. Moreover, since a result can be obtained for each of a plurality of query graph patterns (viewpoints) by using a plurality of query graph patterns, the correlation between the query graph pattern (viewpoint) and the strength of the relationship between nodes may be shown. It becomes possible.

  Therefore, according to the present embodiment, the relationship extraction apparatus 10 connected to the graph structure database 30 that stores graph structure data composed of various nodes and various arcs is required to obtain a plurality of relationship strengths. A query issuing unit 11 for inputting a node and searching for a query graph pattern related to the node input from the query graph pattern database 12 storing a large number of query graph patterns, and a node input using the searched query graph pattern By calculating the coefficient using the number of nodes of the same concept that overlap between the subgraphs extracted from the graph structure database 30 for each of the values, the strength of the relationship between the input nodes can be obtained. Further, by using a plurality of query graph patterns, it is possible to obtain the strength of the relationship between nodes input from a plurality of viewpoints.

It is a block diagram which shows the structure of the relationship extraction system in one embodiment. It is explanatory drawing which shows the data in the graph structure database connected to the relationship extraction apparatus in one embodiment. FIG. 3 is a part of the data shown in FIG. 2, and is an explanatory diagram showing data in the original relational database and a state in which this data is expressed in RDF using an XML format and a graph format. It is a sequence diagram which shows operation | movement of the relationship extraction system shown in FIG. It is explanatory drawing which shows the query graph pattern utilized with the relationship extraction system shown in FIG. It is explanatory drawing which shows another query graph pattern utilized with the relationship extraction system shown in FIG. It is explanatory drawing which shows the state which showed the subgraph obtained by applying the query graph pattern shown in FIG. 5 with respect to the graph structure data shown in FIG. 2 with the thick line. FIG. 6 is an explanatory diagram showing a result of calculating coefficients by applying the query graph pattern shown in FIG. 5 to the graph structure data shown in FIG. 2. It is explanatory drawing which shows the result of having calculated the coefficient by applying the query graph pattern shown in FIG. 6 with respect to the graph structure data shown in FIG. It is explanatory drawing which represented the calculation result of the coefficient shown in FIG. 8 with the graph structure. It is explanatory drawing which represented the calculation result of the coefficient shown in FIG. 9 with the graph structure.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Relation extraction system 10 ... Relation extraction apparatus 11 ... Query issuing part 12 ... Query graph pattern database 13 ... Coefficient calculation part 20 ... User interface provision part 30 ... Graph structure database

Claims (10)

Receiving a selection of a plurality of nodes among nodes stored in a graph structure database storing data having a graph structure composed of various nodes and various arcs by an input means;
For each of the selected nodes, the selected node is substituted for a variable node of a query graph pattern for extracting data whose structure matches from the data having the graph structure by each query issuing means, and the query graph Obtaining data matching the pattern from the graph structure database;
The coefficient calculating means compares the data acquired for each of the selected nodes in the step of acquiring the data to determine the number of overlapping nodes, and calculates the coefficient between the selected nodes based on the overlap number. A calculating step;
A relationship extraction method characterized by comprising:   The relation according to claim 1, wherein the step of acquiring the data acquires a query graph pattern including a variable node corresponding to the selected node from a query graph pattern database storing a query graph pattern. Extraction method.   The relation extraction method according to claim 1, wherein the step of acquiring data generates a query graph pattern including a variable node corresponding to the selected node based on the selected node.   The relation extracting method according to claim 1, wherein the accepting step inputs a query graph pattern including a variable node corresponding to the selected node.   5. The relationship extraction method according to claim 1, further comprising a step of displaying a relationship between the selected nodes as a graph structure based on the coefficient calculated by the calculating step. Connected to a graph structure database that stores data with a graph structure consisting of various nodes and arcs,
Input means for accepting selection of a plurality of nodes among the nodes stored in the graph structure database;
For each of the selected nodes, data that matches the query graph pattern by substituting the selected node into a variable node of a query graph pattern for extracting data whose structure matches from the data having the graph structure Issuance means for obtaining from the graph structure database;
A coefficient calculation means for comparing the data obtained by the query issuing means for each of the selected nodes to determine the number of overlapping nodes, and calculating a coefficient between the selected nodes based on the overlap number;
A relationship extraction system characterized by comprising: A query graph pattern database storing query graph patterns;
The relation extraction system according to claim 6, wherein the query issuing unit acquires a query graph pattern including a variable node corresponding to the selected node from the query graph pattern database.   The relation extraction system according to claim 6, wherein the query issuing unit generates a query graph pattern including a variable node corresponding to the selected node based on the selected node.   The relation extraction system according to claim 6, wherein the input unit inputs a query graph pattern including a variable node corresponding to the selected node.   10. The relation extraction system according to claim 6, further comprising display means for displaying a relation between the selected nodes as a graph structure based on the coefficient calculated by the coefficient calculation means.

Images