JP2011248534A - Network analysis device, network analysis method and program for network analysis using graph pattern - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a network analysis device, a network analysis method and a program for network analysis permitting highly accurate analysis of a network represented in a labeled directed graph.SOLUTION: There are provided an analytical graph generating unit 31 that is connected to an RDF data memory device 10 in which RDF data of a labeled directed graph are stored, calculates a coefficient representing the state of superposition of result nodes identified out of the RDF data by substituting a plurality of designated node values out of the values of nodes matching a designated label for a key node of an analytical graph pattern having the same label as a designated label, and so generates an analytical graph, which is a non-labeled undirected graph, that this coefficient has an edge between nodes at or above a preset threshold, and a network analyzing unit 33 that performs network analysis of this analytical graph.

Description

  The present invention relates to a network analysis device using a graph pattern, a network analysis method, and a network analysis program for analyzing a network expressed by a directed graph with a label.

  Conventionally, there is a data mining technique as a technique for extracting information showing a characteristic tendency as information necessary for human decision making from a huge amount of data. In this data mining technique, characteristic numerical values calculated by analyzing the relationship between data mainly by a statistical method are used.

  However, this statistical method is not suitable for analyzing relevance from the graph structure data expressed as a network by nodes and the edges connecting between the nodes, in which various things are complicatedly related. Network analysis technology for mathematically analyzing the characteristics of structural data has attracted attention and has been actively studied.

  Typical examples of network analysis techniques for analyzing graph structure data include centrality analysis and clique extraction analysis.

  This centrality analysis is to analyze the central node by extracting nodes that have a strong influence from the graph structure data to be analyzed. Clique analysis is the graph structure to be analyzed. The faction of a node is analyzed by extracting a part of data formed by three or more closely related nodes from the data.

  By using these analysis techniques, it is possible to analyze graph structure data including a large amount of information.

  On the other hand, in recent years, Semantic Web, which is a technology that makes it possible to process a large amount of information on a computer by expressing a labeled directed graph on a network using a metadata description method called RDF (Resource Description Framework) among graph structure data. Is spreading.

  In this Semantic Web, more advanced graph structure data can be processed in consideration of the relationship and features used to describe the subject and object resources.

JP 2008-181331 A

  However, the above-described centrality analysis and clique analysis are intended for analysis of a network expressed by an unlabeled undirected graph (a graph composed of one type of node and one type of edge). There is a problem that it cannot be applied as it is to the analysis of a network expressed by (a graph composed of a plurality of types of nodes, the presence / absence of relationships between nodes, and edges indicating the types of relationships by directions).

  In order to perform a network analysis expressed by a directed graph with a label, by using the technique described in Patent Document 1, the same graph pattern is used for the nodes of the same label for which a correlation is to be obtained. It is conceivable to perform pattern matching and extract graph structure data of an unlabeled undirected graph composed of the same kind of edges as a result.

  However, the weights indicating the correlation between nodes are added to the edges of the graph structure data extracted here, and when used in network analysis as they are, edges with low correlation become noise and high-precision analysis can be performed. There was a problem that I could not.

  In view of the above, an object of the present invention is to provide a network analysis apparatus, a network analysis method, and a network analysis program using a graph pattern, which can analyze a network expressed by a directed graph with a label with high accuracy. And

  In order to solve the above problems, the network analysis device of the present invention includes a node including a value indicating a data element and a label indicating the type of the value, a label indicating a relationship between the nodes, and an edge including the direction of the relationship A graph structure data storage device in which graph structure data having a labeled directed graph structure is stored, one key node having a value designated from the outside as a node value, and a node that can take an arbitrary value A graph for acquiring an analysis graph pattern having a key node having the same label as a specified label from the graph pattern storage device in a network analysis device connected to the graph pattern storage device for storing the graph pattern configured by including the graph structure A pattern acquisition unit and an analysis graph acquired by the graph pattern acquisition unit A result node specifying unit for specifying a result node from the graph structure data stored in the graph structure data storage device by substituting the values of a plurality of specified nodes for the key node of the turn, and the specified node A coefficient indicating the degree of overlap of the result nodes specified by the result node specifying unit is calculated for any two nodes having a plurality of node values, and the calculated coefficients satisfy a preset condition. An analysis graph generation unit that generates an analysis graph of an unlabeled undirected graph composed of nodes having values of the plurality of nodes so as to have edges between the nodes, and an analysis graph generated by the analysis graph generation unit A network analysis unit that performs a network analysis using a predetermined analysis method and generates an analysis result, That.

  A network analysis device according to another aspect of the present invention includes a node including a value indicating a data element and a label indicating the type of the value, a label indicating a relationship between the nodes, and an edge including the direction of the relationship. In a network analyzer connected to a graph structure data storage device in which graph structure data having a labeled directed graph structure is stored, one key node having a value designated from the outside as a node value, and an arbitrary value A frequent graph pattern extraction unit that extracts a graph pattern composed of a graph structure including a node that can take a plurality of points, and a graph structure data stored in the graph structure data recording device that matches a predetermined number or more of the graph structure parts And a specified label among the graph patterns extracted by the frequent graph pattern extraction unit. A graph pattern acquisition unit that acquires a key node having the same label as an analysis graph pattern, and assigns the values of a plurality of specified nodes to the key nodes of the analysis graph pattern acquired by the graph pattern acquisition unit, respectively. The result node specifying unit for specifying a result node from the graph structure data stored in the graph structure data storage device, and the result for any two nodes having the values of the specified plurality of nodes A node having values of the plurality of nodes so that a coefficient indicating the overlapping degree of the result nodes specified by the node specifying unit is calculated, and the calculated coefficient has an edge between nodes satisfying a preset condition. An analysis graph generation unit that generates an analysis graph of an unlabeled undirected graph composed of: Relative analysis graph generated by the generating unit performs network analysis using a predetermined analysis method, characterized in that it comprises a network analyzer for generating analysis results.

  Further, the network analysis method of the present invention provides a label with a node composed of a value indicating a data element and a label indicating the type of the value, a label indicating a relationship between the nodes, and an edge including the direction of the relationship. It is composed of a graph structure data storage device storing graph structure data having a directed graph structure, a graph structure including one key node whose value is a value designated from the outside, and a node that can take an arbitrary value. A graph pattern acquisition step in which a computer connected to a graph pattern storage device that stores a graph pattern acquires an analysis graph pattern having a key node having the same label as the designated label from the graph pattern storage device, and the graph pattern acquisition In the key node of the analysis graph pattern obtained in the step, A result node specifying step of specifying a result node from the graph structure data stored in the graph structure data storage device by substituting the values of the plurality of specified nodes, respectively, For any two nodes having values, a coefficient indicating the degree of overlap of the result nodes specified by the result node specifying unit is calculated, and the calculated coefficient sets an edge between nodes satisfying a preset condition. An analysis graph generating step for generating an analysis graph of an unlabeled undirected graph composed of nodes having values of the plurality of nodes, and an analysis graph generated in the analysis graph generation step And a network analysis step for generating an analysis result by performing network analysis using the analysis method of That.

  A network analysis method according to another aspect of the present invention includes a node including a value indicating a data element and a label indicating the type of the value, a label indicating a relationship between the nodes, and an edge including the direction of the relationship. The computer connected to the graph structure data storage device storing the graph structure data having the structure of the directed graph with the label takes one key node having the value designated from the outside as the node value and an arbitrary value. A frequent graph pattern extracting step of extracting a graph pattern composed of a graph structure including a node to be obtained and a graph structure data stored in the graph structure data recording device with a predetermined number or more of the graph structure parts being matched; Of the graph patterns extracted by the frequent graph pattern extraction unit, the same label as the specified label is displayed. A graph pattern acquisition step of acquiring a key node having a key as an analysis graph pattern, and substituting the values of a plurality of specified nodes into the key node of the analysis graph pattern acquired by the graph pattern acquisition unit, respectively A result node specifying step for specifying a result node from the graph structure data stored in the graph structure data storage device, and the result node for any two nodes having values of the plurality of designated nodes A coefficient indicating the degree of overlap of the result nodes specified by the specifying unit is calculated, and from the nodes having the values of the plurality of nodes such that the calculated coefficient has an edge between nodes satisfying a preset condition. An analysis graph generation step for generating an analysis graph of the unlabeled undirected graph configured; Relative analysis chart analysis graph generated by the generating step performs a network analysis using a predetermined analysis method, and having a network analysis step of generating an analysis result.

  A network analysis program according to the present invention causes a computer to function as any one of the above-described network analysis apparatuses.

  According to the network analysis device, the network analysis method, and the network analysis program using the graph pattern of the present invention, the analysis of the network expressed by the directed graph with a label can be performed with high accuracy.

It is a block diagram which shows the structure of the network analysis system using the network analysis apparatus by 1st Embodiment of this invention. It is a sequence diagram which shows operation | movement of the network analysis system using the network analyzer by 1st Embodiment of this invention. It is an example of the RDF data memorize | stored in the RDF data storage device connected to the network analysis apparatus by 1st Embodiment and 2nd Embodiment of this invention. It is explanatory drawing which shows the example of the graph pattern for a search utilized with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is a flowchart which shows the operation | movement of the analysis graph production | generation process performed with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is explanatory drawing which shows a state when a result node is extracted in the analysis graph production | generation process performed with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is a table | surface which shows the example of the Simpson coefficient calculated in the analysis graph production | generation process performed with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is explanatory drawing which shows an example of the analysis graph produced | generated in the analysis graph production | generation process performed with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is a table | surface which shows the other example of the Simpson coefficient calculated in the analysis graph production | generation process performed with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is explanatory drawing which shows the other example of the analysis graph produced | generated in the analysis graph production | generation process performed with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is a table | surface which shows an example of the centrality calculated in the network analysis process performed with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is explanatory drawing which shows an example of the clique extracted in the network analysis process performed with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is a flowchart which shows the operation | movement of the periphery information search process performed with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is explanatory drawing which shows the example of the periphery information extracted by the periphery information search process performed with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is explanatory drawing which shows an example of the analysis result produced | generated with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is a flowchart which shows the operation | movement of the periphery information search process performed with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is explanatory drawing which shows an example of the analysis result produced | generated with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is a flowchart which shows the operation | movement of the periphery information search process performed with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is explanatory drawing which shows the example of the periphery information extracted by the periphery information search process performed with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is explanatory drawing which shows an example of the analysis result produced | generated with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is a flowchart which shows the operation | movement of the periphery information search process performed with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is explanatory drawing which shows an example of the analysis result produced | generated with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is explanatory drawing which shows an example of the analysis result produced | generated and displayed with the network analyzer by 1st Embodiment and 2nd Embodiment of this invention. It is a block diagram which shows the structure of the network analysis system using the network analyzer by 2nd Embodiment of this invention. It is a sequence diagram which shows operation | movement of the network analysis system using the network analyzer by 2nd Embodiment of this invention.

<< First Embodiment >>
<Configuration of Network Analysis System Using Network Analysis Device According to First Embodiment>
As shown in FIG. 1, the network analysis system 1 according to the first embodiment of the present invention is configured by connecting an RDF data storage device 10, a user terminal 20, a network analysis device 30, and a graph pattern storage device 40. ing.

  The RDF data storage device 10 stores RDF data in which data to be analyzed is expressed in RDF. The RDF data stored here is described in XML or the like, and each data is a node, and a data node that is a subject (subject) and a data node that is an object (object) for this subject. By representing the relationship (predicate) of each with an edge, it is expressed as a network.

  In the information expressed by the RDF data, each data element is indicated as a node value, and each node is provided with a label indicating the type of the node value. Each edge connecting nodes includes information on the direction from the data node having the subject node value to the data node having the object node value, and indicates the relationship between the subject and the object. A label indicating the type of edge is attached. A graph including information on the direction of the label and the edge is referred to as a directed graph with a label.

  The user terminal 20 is operated by a user, inputs information necessary for causing the network analysis device 30 to perform network analysis processing, and transmits the information to the network analysis device 30. Further, the user terminal 20 has a monitor (not shown), displays information transmitted from the network analysis device 30 and provides it to the user.

  The network analysis device 30 includes an analysis graph generation unit 31, an analysis graph storage unit 32, a network analysis unit 33, and a peripheral information search unit 34.

  The analysis graph generation unit 31 obtains a graph pattern corresponding to the label designated from the user terminal 20 by the user from the graph pattern storage device 40 from among the graphs stored in the graph pattern storage device 40 to obtain an analysis graph pattern. A subgraph that matches the analysis graph pattern is extracted from the RDF data stored in the RDF data storage device 10.

  Then, using the extracted subgraph, a subgraph expressed by an unlabeled undirected graph related to an analysis node set, which is a set of node values designated by the user from the user terminal 20, is generated as an analysis graph.

  The analysis graph storage unit 32 is configured by a memory that can be temporarily stored, for example, and stores the analysis graph generated by the analysis graph generation unit 31.

  The network analysis unit 33 performs network analysis for mathematically analyzing the characteristics of the analysis graph stored in the analysis graph storage unit 32 and generates an analysis result.

  The peripheral information search unit 34 acquires metadata about a characteristic node or edge in the analysis graph subjected to network analysis by the network analysis unit 33 as peripheral information, adds it to the analysis result of the network analysis, and the user terminal 20. Output to provide to.

  The graph pattern storage device 40 stores a plurality of search graph patterns generated in advance for detecting a graph structure portion (subgraph) having a specific structure. This search graph pattern has a graph structure including one key node having a value designated from the outside such as the user terminal 20 as a node value and a node capable of taking an arbitrary value.

<Operation of Network Analysis System Using Network Analysis Device According to First Embodiment>
Next, the operation of the network analysis system 1 according to the present embodiment will be described with reference to the sequence diagram of FIG.

  In the present embodiment, the RDF data storage device 10 describes, for example, RDF data expressed as a network as shown in FIG. 3 in XML and stores a plurality of data as analysis target data. Among the RDF data, those indicated by ellipses are nodes indicating the respective data, and the connection between the data node as the subject and the data node as the object for the subject is indicated by an arrow as an edge. .

  The information described in each node in FIG. 3 is the label and node value of each node, the left side of “:” indicates the label, and the right side indicates the node value. For example, for a node described as “person: Sachiko Yamamoto”, the label that is the type of the node value is “person”, and the node value that is the content of the stored data is “Sachiko Yamamoto”. In FIG. 3, in order to simplify the diagram, the label “technical term” is expressed as “technique”, the label “paper” is expressed as “discussion”, the label “project” is expressed as “p”, and the label “field” Is expressed as “minute”.

  In addition, the arrow at each edge indicates the direction from the data node that is the subject to the data node that is the object, and the label indicating the relationship between the subject and the object is distinguished by the type of arrow. .

  For example, the solid arrow is labeled “Author”, the two-dot dashed arrow is labeled “Project Manager”, the long dotted line is labeled “Thesis Keyword”, and the short dotted line is labeled “Project Keyword” Yes, it is shown that the dashed line is an edge whose label is “field”. The part connected from the node “Theory: Paper A” to the node “People: Sachiko Yamamoto” by the edge of the solid arrow heads from the main node “Theory: Paper A” to the object node “People: Sachiko Yamamoto”. This means that the relationship is “author”, that is, “the author of“ paper A ”is“ Sachiko Yamamoto ””.

  As shown in FIG. 3, the RDF data stored in the RDF data storage device 10 is mixed with nodes and edges with different labels, but the nodes with different labels and the edges with different labels are strictly distinguished. In this RDF data, a node or edge to which a certain label is added tends to be connected to a node or edge having a specific label, and there is a bias in the way of coupling the node or edge.

  When a plurality of RDF data to be analyzed is stored in the RDF data storage device 10 and the RDF data to be analyzed is designated by the user's operation of the user terminal 20, it is transmitted to the network analysis device 30. (S1).

  Information specifying RDF data transmitted from the user terminal 20 is acquired by the analysis graph generation unit 31 of the network analysis device 30.

  Next, an analysis target label that is a label of a desired analysis target node is designated from the user terminal 20 by a user operation (S2). Here, it is assumed that “technical term” is designated as the analysis target label.

  Information on the analysis target label specified by the user on the user terminal 20 is acquired by the analysis graph generation unit 31 of the network analysis device 30, and a graph pattern having a key node corresponding to the specified analysis target label is stored in the graph pattern storage device. 40 is required (S3).

  Then, in response to a request from the analysis graph generation unit 31, a graph pattern having a key node of the same label as the analysis target label specified by the user is extracted from the graph pattern storage device 40 and sent to the analysis graph generation unit 31. It is displayed on the user terminal 20 and provided to the user.

  Here, it is assumed that the user is provided with two pieces of information, that is, the graph pattern A shown in FIG. 4A and the graph pattern B shown in FIG.

  This graph pattern A is a result of setting the key node 101 with the label “technical information” and the node value as an arbitrary variable “? KeyNode”, and the node value as an arbitrary variable “? ResultNode” with the label “person”. The node 102 is connected to the key node 101 by the edge 103 of the label “paper keyword” and is connected to the result node 102 by the edge 104 of the label “author”, and the key node 101 is connected by the edge 106 of the label “paper keyword”. The result node 102 includes an intermediate node 108 connected to the edge 107 of the label “author”. The intermediate nodes 105 and 108 have no label or node value, and are indicated by arbitrary variables “? X1:? Y1” and “? X2:? Y2”, respectively.

  With this structure, the graph pattern A means that the relationship between the key node and the result node is “an author who has written two or more papers on technical terms”.

  The graph pattern B is a result of setting the key node 111 whose label is “technical information” and the node value is an arbitrary variable “? KeyNode” and the node value is “people” and the node value is an arbitrary variable “? ResultNode”. The node 112 is connected to the key node 111 by the edge 113 of the label “project keyword” and the result node 112 is connected to the edge 114 of the label “project person”. The key node 111 is connected to the key 116 by the edge 116 of the label “paper keyword”. The intermediate node 118 is connected to the result node 112 at the edge 117 of the label “author”. The intermediate nodes 115 and 118 have no label or node value, and are indicated by arbitrary variables “? X1:? Y1” and “? X2:? Y2”.

  This structure means that in the graph pattern B, the relationship between the key node and the result node is “the author of the paper on technical terms and also the person in charge of the technical term project”.

  Next, the graph pattern information specified by the user's operation from the graph patterns displayed on the user terminal 20 is transmitted to the network analysis device 30 as an analysis graph pattern (S4).

  Here, it is assumed that the graph pattern A among the graph patterns shown in FIG. 4 is designated and transmitted by the user's operation.

  In addition, among the node values corresponding to the analysis target label specified in step S2, information on a plurality of node values for which a mutual relationship is desired is specified by the user terminal 20 by the user operation as an analysis node set, It is transmitted to the network analyzer 30 (S5).

  Here, the analysis node set {“Technology A”, “Technology B”, “Technology C”, “Technology D”} is designated and transmitted by the user as a node value for which a mutual relationship is desired. And

  Further, threshold information of coefficients for determining that there is a relationship between the nodes indicated by the node values of the analysis node set specified in step S5 is specified by the user terminal 20 by the user's operation, and is sent to the network analysis device 30. It is transmitted (S6).

  Here, it is assumed that “0.5” is designated and transmitted as a threshold when it is determined whether there is a relationship between nodes based on the Simpson coefficient.

  When these pieces of information are transmitted to the network analysis device 30, an analysis graph generation process is executed in the analysis graph generation unit 31 (S7).

  The analysis graph generation process executed by the analysis graph generation unit 31 will be described with reference to the flowchart of FIG.

  First, one node value is selected as an analysis node value from the set of analysis nodes designated in step S5 (S21).

  Next, the analysis node value is substituted into the key node of the analysis graph pattern designated in step S4, thereby generating a matching query for obtaining a result node corresponding to the node value and the analysis graph pattern (S22). ).

  Here, it is assumed that a matching query has been generated in which the node value “B technology” selected in step S21 is assigned to the key node of the graph pattern A in FIG. 4A designated as the analysis graph pattern.

  Next, the generated matching query is used to perform matching processing on the RDF data stored in the RDF data storage device 10 (S23).

  Here, a matching query in which the node value “B technology” is assigned to the key node of the graph pattern A is used to perform matching processing on the RDF data, so that the node value “people” is indicated by a thick line in FIG. : Taro Yamada "," People: Ichiro Tanaka ", and" People: Hanako Suzuki "are obtained as result node values, and a result node set {" Taro Yamada "," Ichiro Tanaka "," Hanako Suzuki "} is generated.

  Then, it is determined whether or not matching processing has been performed for all the node values in the analysis node set (S24). When there is an unprocessed node value (“No” in S24), the process returns to step S21 and further. A matching process is performed.

  In the present embodiment, a result node set {“Sachiko Yamamoto”, “Jiro Nakamura” is obtained by a matching process executed by a matching query in which the node value “A technology” is assigned to the key node of the graph pattern A. ”,“ Taro Yamada ”} is generated, and the result node set {“ Taro Yamada ”,“ Ichiro Tanaka ”,“ Hanako Suzuki ”} is generated by the matching process executed by the matching query to which the node value“ B technology ”is assigned. The result node set {“Taro Yamada”, “Ichiro Tanaka”} is generated by the matching process executed by the matching query generated and assigned the node value “C technology”, and the node value “D technology” is substituted. It is assumed that a result node set {“Ichiro Tanaka”, “Hanako Suzuki”} is generated by the matching process executed by the matching query.

  When it is determined in step S24 that matching processing has been performed on all the node values in the analysis node set (“Yes” in S24), the overlapping degree of the node values in these analysis node sets is determined. Based on this, a value indicating the strength of the relationship between these nodes is calculated (S25).

As a value indicating the strength of the relationship between the nodes, a Simpson coefficient calculated by the following equation (1) is used in the present embodiment. This Simpson coefficient has a stronger relationship as the degree of overlapping of result nodes between nodes increases, and a higher value is calculated.

  An example of the Simpson coefficient between the nodes calculated by the above equation (1) is shown in FIG. In FIG. 7, the coefficient between the node having the node value “Technology A” and the node having the node value “Technology B” is “0.33”, and the node having the node value “Technology A” and the node value “C technology” The coefficient between the nodes of the node “0 technology” and the node of the node value “D technology” is “0”, and the node value “B technology” The coefficient between the node having the node value “C technology” and the node having the node value “D technology” is “1”. It is shown that the coefficient between the node with the node value “C technology” and the node with the node value “D technology” is “0.5”.

  Next, among the calculated Simpson coefficients between the nodes, a value higher than the threshold value “0.5” specified in step S8 is extracted, and the node of the node value in the analysis node set is extracted. A sub-graph expressed by an unlabeled undirected graph composed of the corresponding edges is generated as an analysis graph (S26).

  Here, between the node of the node value “A technology” and the node of the node value “C technology” from FIG. 7, between the node of the node value “B technology” and the node of the node value “C technology”, The node value “B technology” is configured to have an edge between the node of the node value “D technology” and the node value “C technology” and the node of the node value “D technology”. An analysis graph as shown in FIG. 8 is generated. An analysis graph calculated based on the graph pattern A is referred to as an analysis graph A.

  In step S4, the graph pattern B shown in FIG. 4B is designated by the user's operation. In steps S5 and S6, the analysis node set {“A technology”, “B technology”, “C technology”, FIG. 9 shows an example of Simpson coefficients calculated when “D technique”} and threshold information “0.5” are designated and the analysis graph generation process is performed.

  In FIG. 9, the coefficient between the node with the node value “A technology” and the node with the node value “B technology” is “0”, and the node with the node value “A technology” and the node with the node value “C technology” The coefficient between the node having the node value “A technology” and the node having the node value “D technology” is “0”, and the node having the node value “B technology” The coefficient between the node of the node value “C technology” is “0.5”, and the coefficient between the node of the node value “B technology” and the node of the node value “D technology” is “0.5”. It is shown that the coefficient between the node with the node value “C technology” and the node with the node value “D technology” is “0.5”.

  Then, among the calculated Simpson coefficients between the nodes, a value higher than the designated threshold value “0.5” is extracted, and the sub-graph expressed by the unlabeled undirected graph as described above is the analysis graph. Is generated as

  Here, between the node of the node value “B technology” and the node of the node value “C technology” from FIG. 9, between the node of the node value “B technology” and the node of the node value “D technology”, An analysis graph as illustrated in FIG. 10 is generated, which is configured to have an edge between the node having the node value “C technology” and the node having the node value “D technology”. An analysis graph calculated based on the graph pattern B is referred to as an analysis graph B.

  The analysis graph expressed by the unlabeled undirected graph generated in the present embodiment is configured by nodes having the same specified label, and is configured by edges obtained by applying the same rule by the pattern graph. Therefore, all edges can be regarded as edges having the same label, and this analysis graph can be regarded as an unlabeled graph. The edge direction of the analysis graph is regarded as an undirected graph because the semantic object is maintained between the two nodes connected to the edge by applying the same rule to the pattern graph. There is no problem.

  The generated analysis graph is stored in the analysis graph storage unit 32 (S8).

  When the user terminal 20 is operated by the user and a network analysis is requested in a state where the generated analysis graph is stored, this request is transmitted to the network analysis device 30 (S9).

  In the network analysis device 30, the network analysis request transmitted from the user terminal 20 is acquired by the network analysis unit 33, and the network analysis processing is executed for the stored analysis graph (S10).

  A case where (1) centrality analysis and (2) clique extraction analysis are performed as network analysis processing executed by the network analysis unit 33 in the present embodiment will be described.

(1) Network analysis processing by centrality analysis Centrality analysis is to analyze a central (high centrality) node by extracting nodes with strong influence from the graph structure data to be analyzed. The centrality is analyzed from various viewpoints such as the order center, the adjacent center, and the cutting center by an existing algorithm.

  Examples of existing algorithms for centrality analysis include degree-centric analysis algorithms that calculate the centrality from the number of connections of adjacent nodes, and nodes that are considered to be more influential as the number of connected nodes increases. There is a mediation center analysis algorithm that calculates the centrality from the probabilities that exist in multiple paths that connect between any two nodes, considering that the higher the degree of mediator role is, the more powerful the node is .

Of these, the centrality used in the centrality analysis of the order-centric analysis is calculated by the following equation (2).

  As an example of the centrality analysis, FIG. 11 shows the results when the analysis graph A of FIG. 8 is analyzed according to the order center analysis algorithm.

  As shown in FIG. 11, from the analysis graph A, the centrality (dci) of “Technology A” is “0.33”, and the centrality (dci) of “Technology B” is “0.66”. , It is calculated that the centrality (dci) of “C technology” is “1” and the centrality (dci) of “D technology” is “0.66”. As a result, the central node having the strongest influence in the analysis node set {“A technology”, “B technology”, “C technology”, “D technology”} regarding the label “technical term” is “C technology”. It can be seen that the node having the weakest influence is the “A technology”.

  In this way, by generating an analysis graph with an unlabeled undirected graph structure related to the same label in advance, centrality analysis regarding the label, which was impossible with the directed graph structure data with a plurality of labels mixed, is performed. It becomes possible.

(2) Network analysis processing by clique analysis Clique analysis is a process by extracting a part of data (clique) formed by three or more closely related nodes from the graph structure data to be analyzed. It analyzes the faction.

  This clique analysis has a drawback that the processing accuracy is reduced in graph structure data having a low network density.

  The network density is the density of edges included in the graph structure data, and has a value from “0” to “1”. When the network density is “1”, it shows complete graph structure data in which all the nodes are connected by edges. When the network density is “0”, there is no edge between the nodes and the nodes are completely isolated. The graph structure data is shown.

  Therefore, a data portion having a network density equal to or higher than a preset threshold is extracted as a clique, or a data portion in which all two nodes are connected at a preset distance N is extracted as an N-clique. This makes it possible to perform clique analysis.

  As an example of clique analysis, FIG. 12 shows a result of analyzing analysis graph A by extracting a clique having a network density equal to or higher than a preset threshold.

  In FIG. 12, the nodes and edges forming the extracted clique are indicated by bold lines. From the analysis graph A, the node “B technology”, the node “C technology”, the node “D technology”, and the node “B technology” are indicated. Cliques composed of an edge connecting node “C technology”, an edge connecting node “B technology” and node “D technology”, and an edge connecting node “C technology” and node “D technology” were extracted. It is shown that.

  From this analysis result, “B technology”, “C technology”, and “D technology” form a faction closely related to each other, and “A technology” is a set of analysis nodes of the label “technical term”. It can be seen that the relationship with other nodes is low and isolated.

  In this way, by generating an analysis graph with unlabeled undirected graph structure related to the same label in advance, clique analysis related to the label that was not possible with directed graph structure data with a plurality of labels mixed Is possible.

  Next, based on the results of the network analysis, elements satisfying a predetermined condition in the analysis graph, for example, a central node, an isolated node that does not belong to a clique, an edge that belongs to a different clique, or an inside of a clique A peripheral information search process for searching for metadata related to elements common to the plurality of nodes is executed in the peripheral information search unit 34 (S11).

  A case where the peripheral information search process is performed on the result of the network analysis process by centrality analysis and the case of (2) the result of the network analysis process by clique analysis will be described.

  FIG. 13 is a flowchart showing an operation when (1) peripheral information search processing is performed on the result of network analysis processing by centrality analysis. In this peripheral information search process, first, a node having a high centrality is extracted as an analysis node from the network analysis process result (S31, S32).

  This analysis node is extracted based on parameters such as a preset threshold value from the centrality of each node calculated as shown in FIG. 11, for example. This parameter is set in a network analysis request transmitted by the user's operation in step S9 or a setting file stored in advance when the system is executed.

  Specific “high-centrality node” extraction processing is as follows: 1) A threshold is set as a parameter, and a node with a centrality equal to or greater than this threshold is extracted. 2) Extraction percentage (n%) as a parameter Set the node and sort the nodes to be processed in descending order of centrality, and extract the top n% nodes from the sorted nodes, or the methods 1) and 2) There are methods to use together.

  A peripheral information search process is executed on the analysis node extracted in this way (S33).

  In this peripheral information search process, a node connected to the analysis node via one edge is searched from the original RDF data from which the analysis graph is extracted, and acquired as peripheral information.

  For example, in the network analysis result of the centrality analysis of the analysis graph A, “C technology” is extracted as a central node, so one edge is added to this “C technology” from the RDF data from which the analysis graph A is extracted. As shown in FIG. 14, the node “L field” connected to the edge “field”, the node “project 2” and the node “project 1” connected to the edge “project keyword”, and the edge “ Nodes “Paper G”, “Paper H”, and “Paper I” connected to “Paper” are acquired as peripheral information.

  Next, as shown in FIG. 15, the acquired peripheral information is added to the result of the network analysis process based on the centrality analysis, and an analysis result in which the peripheral information is linked is obtained (S34, S35).

  FIG. 16 is a flowchart showing the operation when the peripheral information search process is performed on the result of the network analysis process by (2) clique analysis. In this peripheral information search process, nodes that do not belong to any clique are first extracted as analysis nodes from the network analysis process result as described above (S41, S42). This “node that does not belong to any clique” is a node that is clearly identified by the above-described “network analysis processing by clique analysis”.

  A process similar to the peripheral information search process described in step S33 is executed on the extracted analysis node (S43).

  For example, since the “A technology” is extracted as a node that does not belong to any clique in the network analysis result by the clique analysis of the analysis graph A, the “A technology” is one of the RDF data from which the analysis graph A is extracted. When the nodes connected through the edge are searched, the node “L field” connected to the edge field and the nodes “paper A”, “paper B”, “paper C”, “paper D”, “ “Paper E” and “Paper F” are acquired as peripheral information.

  Next, as shown in FIG. 17, the acquired peripheral information is added to the result of performing the network analysis processing by clique analysis, and the analysis result in which the peripheral information is associated is obtained (S44, S45).

  In another example in which the peripheral information search process is performed on the result of the network analysis process by clique analysis, edges belonging to different cliques are extracted from the network analysis process result (S52), Related information between two nodes connected to both ends of the edge is searched as peripheral information (S53).

  In this peripheral information search process, the related information between the two nodes connected to both ends of the edge is obtained from the original RDF data from which the analysis graph is extracted and the result of the matching process using the analysis graph pattern specified in step S4. Is extracted.

  For example, in the network analysis result by the clique analysis of the analysis graph A, edges connected to the node “A technology” and the node “C technology” are extracted as edges belonging to different cliques.

  Then, using the analysis graph pattern A used for the analysis graph A, matching processing is performed on the RDF data using the nodes “A technology” and “C technology” as key nodes, respectively. As shown in FIG. The result node “Taro Yamada” obtained redundantly for one key node is acquired as a related node set.

  This is a matter related to "Technology A" and "Technology C", and presents that "Taro Yamada" was used as a keyword in two or more papers. " is there.

  Next, as shown in FIG. 20, the acquired related node set is added as the peripheral information to the result of the network analysis processing by clique analysis, and the analysis result in which the peripheral information is linked is obtained (S54, S55).

  In addition, in another example in which the peripheral information search process is performed on the result of the network analysis process by clique analysis, an analysis node set belonging to each clique is extracted from the network analysis process result (S62). ), Common information search processing regarding nodes included in the set of analysis nodes is performed (S63).

  For example, for the clique extracted as shown in FIG. 12, {"B technology", "C technology", "D technology"} is extracted as the set of analysis nodes to which it belongs. Then, as the common information search processing regarding these nodes, first, the graph pattern A used for clique extraction is used, and each node of the analysis node set is substituted for the key node, and matching processing is performed on the RDF data. Thus, a result node set is generated.

  For example, a result node set {“Taro Yamada”, “Ichiro Tanaka”, “Hanako Suzuki”} is generated by substituting “B technology” into a key node, and a result node set by substituting “C technology” into a key node. {“Taro Yamada”, “Ichiro Tanaka”} are generated, and a result node set {“Ichiro Tanaka”, “Hanako Suzuki”} is generated by substituting “D technology” into a key node.

  Next, for each node included in the analysis node set, the number of node values included in the result node set is set to the number of nodes included in the analysis node set (here, “B technology”, “C technology”, “D technology”). The score is calculated by dividing by 3).

  For example, for “B technology”, the score “1” is calculated by the number of nodes included in the result node set “3” ÷ the number of nodes included in the analysis node set “3”, and for “C technology”, the result node The score “0.67” is calculated by the number of nodes included in the set “2” ÷ the number of nodes included in the analysis node set “3”. For “D technology”, the number of nodes included in the result node set “2” ÷ Score “0.67” is calculated from the number of nodes “3” included in the analysis node set.

  Then, if necessary, remove nodes with a score equal to or lower than a preset threshold to filter out nodes with low commonality, or extract only nodes with a higher predetermined number of score values. Narrow nodes with high

  Here, if the threshold value of the score is “0.5”, “B technology”, “C technology”, and “D technology” exceed this threshold value, so all three are extracted as highly common nodes. Is done.

  Next, as shown in FIG. 22, the acquired highly common node is added as the peripheral information to the result of the network analysis processing by clique analysis, and the analysis result in which the peripheral information is linked is obtained ( S64, S65).

  When the peripheral information search process is performed by any one of these processes and an analysis result associated with the peripheral information is obtained, the analysis result is transmitted to the user terminal 20 and displayed (S12).

  An example of the analysis result displayed on the user terminal 20 is shown in FIG. In FIG. 23, regarding the cliques a to d, the extracted common elements are linked and displayed as the peripheral information.

  According to the present embodiment described above, an analysis graph of an unlabeled undirected graph that can be analyzed using an existing network analysis technology can be generated from a network represented by a directed graph with a label, and the accuracy can be achieved with a simple operation. High analysis processing.

  Further, in this embodiment, the analysis graph pattern or analysis target label specified by the user is used to generate an analysis graph of an unlabeled undirected graph narrowed down to one relationship, and the network analysis is performed. By changing to the analysis graph pattern, it is possible to generate an analysis graph with different meanings (relationships between nodes) of the edge of the analysis graph, and by changing the analysis target label, it consists of nodes with different labels An analysis graph can be generated, and analysis results can be obtained from various viewpoints.

  Further, by searching for and adding peripheral information related to information indicating the characteristics of the analysis graph, it is possible to provide a more accurate analysis result.

<< Second Embodiment >>
<Configuration of Network Analysis System Using Network Analysis Device According to Second Embodiment>
As shown in FIG. 24, the configuration of the network analysis system 2 according to the present embodiment does not include the graph pattern storage device 40. The network analysis device 30 includes a frequent graph pattern extraction unit 35, a frequent graph pattern storage unit 36, and Since the configuration is the same as the configuration of the network analysis system 1 according to the first embodiment, the description of the configuration units having the same functions is omitted.

  The frequent graph pattern extraction unit 35 extracts, from the RDF data stored in the RDF data storage device 10, a graph pattern having a specific structure that matches a predetermined number or more of graph structure parts (subgraphs) as a frequent graph pattern. .

  The frequent graph pattern storage unit 36 is constituted by a memory that can be temporarily stored, for example, and stores the frequent graph pattern extracted by the frequent graph pattern extraction unit 35.

  The analysis graph generation unit 31 sets a graph pattern designated by the user from the user terminal 20 from among those stored in the frequent graph pattern storage unit 36 as an analysis graph pattern, and subgraphs matching the analysis graph pattern are stored in the RDF data storage device 10. From the RDF data stored in.

  Then, using the extracted subgraph, a subgraph expressed by an unlabeled undirected graph related to an analysis node set, which is a set of node values designated by the user from the user terminal 20, is generated as an analysis graph.

<Operation of Network Analysis System Using Network Analysis Device According to Second Embodiment>
Next, the operation of the network analysis system 1 according to the present embodiment will be described with reference to the sequence diagram of FIG.

  As in the first embodiment, when a plurality of RDF data to be analyzed is stored in the RDF data storage device 10 and the RDF data to be analyzed is designated by the user's operation of the user terminal 20, the network analysis device 30 (S71).

  Information specifying the RDF data transmitted from the user terminal 20 is acquired by the frequent graph pattern extraction unit 35 of the network analysis device 30. The frequent graph pattern extraction unit 35 extracts, as a frequent graph pattern, a graph pattern having a specific structure that matches a predetermined number or more of graph structure parts (subgraphs) from the RDF data stored in the RDF data storage device 10. (S72). This graph pattern has a graph structure including one key node having a value designated from the outside such as the user terminal 20 as a node value and a node capable of taking an arbitrary value.

  Here, the threshold for determining whether or not the number of subgraphs is greater than or equal to the “predetermined number” may be set in a setting file or the like stored in advance when the system is executed, or in step S71. It may be set by a user operation when RDF data designation is performed.

  The extracted frequent graph pattern is stored in the frequent graph pattern storage unit 36 (S73).

  By extracting this frequent graph pattern, a subgraph having a high network density can be obtained as an analysis target, and there is a tendency that it is easy to obtain an analysis result that is useful and highly accurate in network analysis processing described later.

  Next, an analysis target label that is a label of a desired analysis target node is designated from the user terminal 20 by a user operation (S74). Here, it is assumed that “technical term” is designated as the analysis target label.

  Information on the analysis target label specified by the user on the user terminal 20 is acquired by the analysis graph generation unit 31 of the network analysis device 30, and the frequent graph pattern having key nodes corresponding to the specified analysis target label is the frequent graph pattern. The extraction unit 35 is requested (S75).

  Then, in response to a request from the analysis graph generation unit 31, a frequent graph pattern having a key node with the same label as the analysis target label specified by the user is extracted from the frequent graph pattern storage unit 36 and sent to the analysis graph generation unit 31. At the same time, it is displayed on the user terminal 20 and provided to the user.

  Since the subsequent processes in steps S76 to S84 are the same as the processes in steps S4 to S12 in the first embodiment, detailed description thereof is omitted.

  According to the present embodiment described above, among the plurality of search graph patterns generated in advance, those having a predetermined number or more of matching subgraphs in the corresponding RDF data are extracted as frequent graph patterns and used for processing. Therefore, a subgraph having a high network density can be obtained as an analysis target, and a highly accurate analysis result useful in network analysis processing can be obtained.

  In the first and second embodiments described above, the RDF data designation information, analysis target label designation information, analysis node set designation information, coefficient threshold designation information, and the like are input from the user terminal 20 as necessary. As described above, the processing may be executed by inputting these pieces of information in advance.

  Moreover, it is also possible to construct a network analysis program that causes the computer to function as a network analysis device by programming the functional configuration of the network analysis device in each of the above embodiments into a computer.

DESCRIPTION OF SYMBOLS 1, 2 ... Network analysis system 10 ... RDF data storage device 20 ... User terminal 30 ... Network analysis device 31 ... Analysis graph generation part 32 ... Analysis graph storage part 33 ... Network analysis part 34 ... Peripheral information search part 35 ... Frequent graph pattern Extraction unit 36 ... frequent graph pattern storage unit 40 ... graph pattern storage device

Claims (7)

Stores graph structure data with a labeled directed graph structure consisting of nodes consisting of a value indicating the data element and a label indicating the type of this value, a label indicating the relationship between the nodes, and an edge consisting of the direction of this relationship Graph structure data storage device, one key node having a value designated from the outside as a node value, and a graph pattern storage device storing a graph pattern including a graph structure including a node that can take an arbitrary value; In the network analyzer connected to
A graph pattern acquisition unit for acquiring an analysis graph pattern having a key node having the same label as the specified label from the graph pattern storage device;
By substituting the values of a plurality of designated nodes into the key nodes of the analysis graph pattern acquired by the graph pattern acquisition unit, the result nodes are selected from the graph structure data stored in the graph structure data storage device. A result node identification part to be identified;
A coefficient indicating the degree of overlap of the result nodes specified by the result node specifying unit is calculated for any two nodes having the values of the specified plurality of nodes, and the calculated coefficients are set in advance. An analysis graph generation unit that generates an analysis graph of an unlabeled undirected graph composed of nodes having values of the plurality of nodes so as to have an edge between nodes satisfying the condition;
A network analysis unit that performs a network analysis using a predetermined analysis technique on the analysis graph generated by the analysis graph generation unit, and generates an analysis result;
A network analysis apparatus comprising: Stores graph structure data with a labeled directed graph structure consisting of nodes consisting of a value indicating the data element and a label indicating the type of this value, a label indicating the relationship between the nodes, and an edge consisting of the direction of this relationship Network analyzer connected to the graph structure data storage device,
From the graph structure data stored in the graph structure data recording device, the graph pattern is composed of a graph structure including one key node having a value designated from the outside as a node value and a node capable of taking an arbitrary value. A frequent graph pattern extraction unit that extracts a matching number of graph structure parts of a predetermined number or more;
Of the graph patterns extracted by the frequent graph pattern extraction unit, a graph pattern acquisition unit that acquires a key node having the same label as the specified label as an analysis graph pattern;
By substituting the values of a plurality of designated nodes into the key nodes of the analysis graph pattern acquired by the graph pattern acquisition unit, the result nodes are selected from the graph structure data stored in the graph structure data storage device. A result node identification part to be identified;
A coefficient indicating the degree of overlap of the result nodes specified by the result node specifying unit is calculated for any two nodes having the values of the specified plurality of nodes, and the calculated coefficients are set in advance. An analysis graph generation unit that generates an analysis graph of an unlabeled undirected graph composed of nodes having values of the plurality of nodes so as to have an edge between nodes satisfying the condition;
A network analysis unit that performs a network analysis using a predetermined analysis technique on the analysis graph generated by the analysis graph generation unit, and generates an analysis result;
A network analysis apparatus comprising: A peripheral information acquisition unit that acquires, as peripheral information, metadata related to a node or an edge that satisfies a predetermined condition in an analysis graph in which network analysis has been performed by the network analysis unit;
The provisional information output part which adds the peripheral information acquired by the said peripheral information acquisition part to the analysis result produced | generated by the said network analysis part, and outputs as provision information is characterized by the above-mentioned. The network analyzer described. Stores graph structure data with a labeled directed graph structure consisting of nodes consisting of a value indicating the data element and a label indicating the type of this value, a label indicating the relationship between the nodes, and an edge consisting of the direction of this relationship Graph structure data storage device, one key node having a value designated from the outside as a node value, and a graph pattern storage device storing a graph pattern including a graph structure including a node that can take an arbitrary value; The computer connected to
A graph pattern acquisition step of acquiring an analysis graph pattern having a key node having the same label as the specified label from the graph pattern storage device;
By substituting the values of a plurality of specified nodes into the key nodes of the analysis graph pattern acquired in the graph pattern acquisition step, result nodes are selected from the graph structure data stored in the graph structure data storage device. A result node identification step to identify;
A coefficient indicating the degree of overlap of the result nodes specified by the result node specifying unit is calculated for any two nodes having the values of the specified plurality of nodes, and the calculated coefficients are set in advance. An analysis graph generation step of generating an analysis graph of an unlabeled undirected graph composed of nodes having values of the plurality of nodes so as to have an edge between nodes satisfying the condition;
A network analysis step for performing a network analysis on the analysis graph generated in the analysis graph generation step using a predetermined analysis technique and generating an analysis result;
A network analysis method characterized by comprising: Stores graph structure data with a labeled directed graph structure consisting of nodes consisting of a value indicating the data element and a label indicating the type of this value, a label indicating the relationship between the nodes, and an edge consisting of the direction of this relationship A computer connected to the graph data storage device
From the graph structure data stored in the graph structure data recording device, the graph pattern is composed of a graph structure including one key node having a value designated from the outside as a node value and a node capable of taking an arbitrary value. A frequent graph pattern extracting step for extracting a graph structure portion that matches a predetermined number or more,
Of the graph patterns extracted by the frequent graph pattern extraction unit, a graph pattern acquisition step of acquiring a key node having the same label as the specified label as an analysis graph pattern;
By substituting the values of a plurality of designated nodes into the key nodes of the analysis graph pattern acquired by the graph pattern acquisition unit, the result nodes are selected from the graph structure data stored in the graph structure data storage device. A result node identification step to identify;
A coefficient indicating the degree of overlap of the result nodes specified by the result node specifying unit is calculated for any two nodes having the values of the specified plurality of nodes, and the calculated coefficients are set in advance. An analysis graph generation step of generating an analysis graph of an unlabeled undirected graph composed of nodes having values of the plurality of nodes so as to have an edge between nodes satisfying the condition;
A network analysis step for performing a network analysis on the analysis graph generated in the analysis graph generation step using a predetermined analysis technique and generating an analysis result;
A network analysis method characterized by comprising: Peripheral information acquisition step for acquiring, as peripheral information, metadata related to nodes or edges that satisfy a predetermined condition in the analysis graph in which network analysis has been performed in the network analysis step;
6. The provision information output step of adding the peripheral information acquired in the peripheral information acquisition step to the analysis result generated in the network analysis step and outputting the information as provision information. The network analysis method described.   A network analysis program for causing a computer to function as the network analysis device according to any one of claims 1 to 3.