JP4628258B2 - Graph search device - Google Patents

Description

  The present invention relates to a graph search apparatus that can search a subgraph with a clear relationship between subgraphs searched from each keyword at high speed.

  A conventional graph search device (for example, the one described in Non-Patent Document 1) searches and displays a plurality of subgraphs matching a query from a certain graph, for example, as shown in FIG. One is searched from the keyword “XML”, and the other is searched from the keyword “Web service”.

  Further, a graph retrieval apparatus using graph mining called AGM (Apriori-based Graph Mining) described in Non-Patent Document 2 retrieves frequent subgraphs from labeled graphs.

  Further, a graph search apparatus using a simple shortest path algorithm searches a shortest path (subgraph in a broad sense) between nodes that match a keyword from a certain graph.

In link analysis (HITS algorithm), the relationship between subgraphs is expressed by performing link analysis at the time of retrieval.
SPARQL Query Language for RDF, http://www.w3.org/TR/rdf-sparql-puery/ Complete Mining of Freqent from Graphs, Akihiro Inokuchi, Takashi Washio and Hiroshi Motoda: Complete Mining of Freqent Patterns from Graphs: Mining Graph Data, Machine Learning, Vol.50, pp.321-354, (2003)

  However, in the subgraph as shown in FIG. 28, although both have nodes having the same instance “Person: Dentsu Taro”, the relationship between the subgraphs is not expressed, and thus the relationship is unclear.

  In Non-Patent Document 1, there is a description of searching for a subgraph that matches a query. In Non-Patent Document 2, there is a description of searching for a frequent subgraph, but in any document, the relationship between subgraphs is described. There is no description. Further, in the technique described in Non-Patent Document 2, many subgraphs are searched regardless of the keyword that the user wants, so it is not clear which is related to the user's interest.

  In addition, the graph search apparatus using the simple shortest path algorithm provides only the shortest path (subgraph) relationship and does not fully consider the knowledge around the useful subgraph. In other words, it is impossible to express other than the relationship between the shortest paths (subgraphs).

  In link analysis (HITS algorithm), since link analysis is performed at the time of search, a long time is required for the search.

  The present invention has been made in view of the above problems, and an object of the present invention is to provide a graph search device that can search a subgraph with a clear relationship between subgraphs searched from each keyword at high speed. is there.

  In order to solve the above-mentioned problem, the present invention of claim 1 is characterized in that a characteristic path included in a graph in which nodes having different instances are connected by an arc having a label and the class of the instance is defined. Pattern storage means for storing a pattern in advance, having a node having an instance at one end position, a node having a class defined at the other end position, and each arc having a label A first pattern which is a pattern having a node having an instance equal to the first keyword at one end position and a node having a specified class defined at the other end position from the pattern storage means Search for a node with an instance equal to the second keyword at one end, and with the specified class defined A pattern search unit that searches the pattern storage unit for a second pattern that is a pattern at one end position, a pattern merge unit that merges the first pattern and the second pattern at a node at the other end position, and A graph search device comprising: a graph search unit that searches the graph for a subgraph that matches the pattern merged by the pattern merge unit.

  The present invention of claim 2 searches the graph for a path having, at one end position, a node to which a tip of an arc including a predetermined content in a label is connected, and uses an instance of a node other than the node as a variable. The graph search device according to claim 1, further comprising: a pattern generation unit that generates a replacement pattern and stores the pattern in the pattern storage unit.

  The present invention of claim 3 is a storage means for storing a score of each label possessed by an arc of the graph or a score of each class defined in a node possessed by the graph, and labeling predetermined contents from the graph. A path having a node to which the tip of the arc included in is connected at one end position is searched, an instance of a node other than the node is replaced with a variable, and the score of the path is based on the score stored in the storage unit A pattern for calculating, searching for a subgraph matching the path from the graph, increasing the score according to the number of searched subgraphs, and storing a pattern having a score greater than a predetermined threshold in the pattern storage means The graph search device according to claim 1, comprising a generation unit.

  The invention according to claim 4 is a characteristic path pattern included in a graph in which nodes having different instances are connected by an arc having a label and the class of the instance is defined. A pattern storage means in which a pattern is pre-stored having nodes at both end positions, each arc having a label, and a node having an instance equal to the first keyword at one end position, and a second A first pattern which is a pattern having a node having an instance equal to the keyword at the other end position is retrieved from the pattern storage means, and a node having an instance equal to the first keyword is at one end position, A second pattern that is a pattern with a node having an instance equal to 3 keywords at the other end position A pattern search means for searching the pattern storage means, a pattern merge means for merging the first pattern and the second pattern at a common part included in each, and a pattern merged by the pattern merge means The graph search device includes a graph search means for searching for a subgraph to be searched from the graph.

  According to a fifth aspect of the present invention, a pattern which is a path having a node connected to the tip of an arc including a predetermined content in a label at both end positions is generated from the graph and stored in the pattern storage means. The graph search device according to claim 4, further comprising a pattern generation unit.

  The present invention of claim 6 is a storage means for storing a score of each label possessed by an arc of the graph or a score of each class defined in a node possessed by the graph, and labeling predetermined contents from the graph. The path having the node to which the tip of the arc included in is connected at both end positions is searched, the score of the path is calculated based on the score stored in the storage means, and the score is based on a predetermined threshold value. The graph search device according to claim 4, further comprising: a pattern generation unit that stores a large pattern in the pattern storage unit.

  The seventh aspect of the present invention provides a computer program for causing a computer to operate as the graph search device according to any one of the first to third aspects.

  The eighth aspect of the present invention provides a solution comprising a recording medium storing the computer program according to the seventh aspect.

  According to the graph search device of the present invention, there is a characteristic path pattern included in a graph in which nodes having different instances are connected by an arc having a label and the class of the instance is defined. A pattern storage means for storing a pattern in advance, a node having a node at one end position, a node having a class defined at the other end position, each arc having a label, and a first keyword A first pattern which is a pattern having a node having an equal instance at one end position and a node having a specified class defined at the other end position is retrieved from the pattern storage means, and a second keyword Has a node with an instance equal to at one end position and a node with the specified class defined at the other end position. The pattern search means for searching for the second pattern, which is a pattern, from the pattern storage means, the pattern merge means for merging the first pattern and the second pattern at the node at the other end position, and the pattern merge means for merging And a graph search means for searching the graph for a subgraph that matches the determined pattern, so that a subgraph in which the relationship between the subgraph searched from the first keyword and the subgraph searched from the second keyword is clear can be speeded up. You can search.

  Further, according to the graph search device of the present invention, there is a characteristic path pattern included in a graph in which nodes having different instances are connected by an arc having a label and the class of the instance is defined. A pattern storage means having a node with an instance at both end positions, each arc having a label, a pattern pre-stored, and a node with an instance equal to the first keyword at one end position A first pattern which is a pattern having a node having an instance equal to the second keyword at the other end position is retrieved from the pattern storage means, and a node having an instance equal to the first keyword is at one end position. A putter with a node at the other end position with an instance equal to the third keyword A pattern search means for searching for the second pattern from the pattern storage means, a pattern merge means for merging the first pattern and the second pattern at a common part included in each, and merging by the pattern merge means And a graph search means for searching from the graph for a subgraph that matches the specified pattern, a subgraph searched from the first keyword and the second keyword, and a subgraph searched from the first keyword and the third keyword. It is possible to search a subgraph having a clear relationship with

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. In addition, the same code | symbol is provided to the same thing and duplication description is abbreviate | omitted.

[First Embodiment]
FIG. 1 is a configuration diagram of a graph search apparatus according to the first embodiment.

  The graph search device 1 is connected to a user terminal 2, and a display device 3 is connected to the user terminal 2.

  The graph search device 1 is a graph database 11 in which a data group for displaying a graph G that is a directed graph is stored, and an input interface that generates an input interface that is displayed on the display device 3 when searching a subgraph of the graph. And an interface generation unit 12.

  FIG. 2 is a diagram illustrating a part of a graph G that can be displayed using all data groups stored in the graph database 11.

  Using all the data groups stored in the graph database 11, the graph G partially illustrated in FIG. 2, that is, nodes having different instances are connected by arcs having labels, and classes of the instances are defined. A graph G can be displayed. Conversely, a data group for displaying the graph G is stored in the graph database 11. Hereinafter, the data group is referred to as a graph G for convenience. Also, for convenience, graphs and subgraphs can be used to display any graphs, subgraphs (some graphs themselves or graphs included in them), paths (graphs without branches and closed loops), etc. , Path and so on.

  In the graph G, for example, nodes having instances such as “XML” and “B project” are connected by arcs having labels such as “rm: technical keyword” and “rm: author”. Further, in the graph G, a class “Person: person” or the like that is a concept such as an instance “Person: Taro Yamada” is defined in a node.

  FIG. 3 is a diagram illustrating a part of a data group stored in the graph database 11.

  For example, in the first data “<org: D project> <rm: technical keyword>“ XML ””, an arc from the node having the instance “org: D project” to the node of the instance “XML” is labeled “rm: technique”. It has shown that it has a keyword. For example, the last data “<person: Hanako Suzuki> <rdf: type> <Person: person>” indicates that the class of the node having the instance “person: Hanako Suzuki” is “person: person”. ing.

  Each data is actually expressed in RDF (described in the following document).

"Resource Description Framework (RDF) Model and Syntax Specification", Ora Lassia, Ralph R. Swick, [online], Internet <URL: http://www.w3.org/TR/1999/REC-rdf-syntax- 19990222 />
"RDF Vocabulary Description Language 1.0: RDF Schema", Dan Brickley, RVGuha, [online], Internet <URL: http://www.w3.org/TR/rdf-schema/>
Returning to FIG. 1, the graph search device 1 includes an arc score storage unit 13 used for generating a pattern used for searching for a subgraph.

  FIG. 4 is a diagram showing the contents stored in the arc score storage unit 13. The arc score storage unit 13 stores a score (arc score) of each label of the arc of the graph G. The instance of each node in the graph G is replaced with the class of that node, and only one of the nodes having the same instance (class) is left, and each arc score in the arc score storage unit 13 is stored in the graph thus formed. When applied, a part of it becomes as shown in FIG. 5, for example. In the graph shown in this figure, a node having an instance “org: Organization” is shown as an example of a node having a node instance class (a node class of the graph G) as an instance.

  Returning to FIG. 1, the graph search apparatus further includes a pattern generation unit 14 that generates a pattern, a pattern database 15 that stores the generated pattern, a pattern search unit 16 that searches for a pattern from the pattern database 15, and a pattern Is displayed on the display device 3, a pattern search unit 18 that searches for a subgraph of the graph G, a database interface 19 that relays the pattern search unit 16, the graph search unit 18, and each database. And an output interface generation unit 110 that generates an output interface.

  The graph search device 1 only needs to be able to transmit and receive (deliver) data in each unit (including a database). That is, each unit may be arranged on the same computer or may be distributed on a plurality of computers. In addition, a computer program that causes these computers to operate as a graph search device may be transmitted and received via a communication line. The computer program may be recorded on a recording medium such as a semiconductor memory, a magnetic disk, an optical disk, a magneto-optical disk, or a magnetic tape, and the recording medium may be distributed. The same applies to the second embodiment.

(Operation of the first embodiment)
FIG. 6 is a flowchart of the pattern generation unit 14. Here, when the pattern generation unit 14 handles the graph G, when acquiring a path from the graph G, when the pattern is stored in the pattern database 15, a query is transmitted to the database interface 19, but this is redundant. I will not mention it.

  First, the pattern generation unit 14 sets an initial value “1” of the node score, which indicates the importance level, to the node of the graph G (S1).

  Next, the pattern generation unit 14 calculates a node score for each node according to the formula of FIG. 7 (S3).

  Then, the initial value of the node score set in the graph G is updated with this node score (S5). Such calculation update is performed recursively until the node score converges. As a result, the node score indicates the importance.

  If a part of the graph in which each arc score of the arc score storage unit 13 is applied to the graph G is as shown in FIG. 8, for example, the node score of the node having the instance “XML” is 0.5. . The node score of the node having the instance “Person: Dentsu Saburo” is 0.9. The node score of the node having the instance “org: H3G” is 0.6.

  Next, from the graph G, the pattern generation unit 14 sets a node (keyword node) to which one end of an arc including a predetermined content (for example, “keyword” or “technical term”) as a label is connected to one end position. A path having a predetermined length or less is acquired (S7).

  Next, the pattern generation unit 14 calculates a path score of each path using the formula of FIG. 9 (S9).

  Next, the pattern generation unit 14 replaces the instance of a node other than the keyword node with a variable for each path (S11).

  Next, the pattern generation unit 14 classifies all the paths into path groups (consisting of the same path), and calculates a score for each path group using the formula of FIG. 10 (S13). Then, each path group whose score is equal to or less than a predetermined threshold is deleted (S15).

  Next, the pattern generation unit 14 deletes other than one path for each remaining path group (S17).

  Next, the pattern generation unit 14 assigns the score of the original path group to the remaining path (referred to as a pattern) as the pattern score, and stores it in the pattern database 15 (S19).

  Thereby, two patterns as shown in FIG. 11 are stored in the pattern database 15.

  The pattern is the same as the data group forming a part of the data group (graph G) stored in the graph database 11 and can be graphed as shown in this figure. Patterns are not displayed but are used to search the displayed graph. Since it is redundant to explain the actual pattern as a data group one by one, it will be explained for convenience with a graphed pattern as shown in FIG.

  In general, in a pattern, some nodes and arcs have instances and labels, and the rest do not. Variables are set for nodes and arcs that do not have instances or labels. Variables as shown in the figure? Followed by a word.

  In this pattern, a node at one end position has an instance, a variable is set in the other node, and each arc has a label.

  In general, in a pattern, if necessary, a class indicating a high-level concept of an instance of a node at the same position in a graph searched by the pattern is defined in advance for some nodes set with variables. The

  In this pattern, a class is defined for the node at the other end position.

  A subgraph retrieved from a certain graph by such a pattern has the following conditions.

  In other words, what is searched is (1) the graph or its subgraph, (2) having a pattern structure without excess or deficiency, (3) having instances or labels within the pattern without deficiency, In other words, nodes and arcs at positions equal to the positions of nodes and arcs with instances and labels in the pattern have instances and labels equal to the instances, and (4) in some cases, classes included in the pattern are defined A node at a position equal to the position of the node has an instance that is a subordinate concept of the class.

  Note that searching for a subgraph in this way by pattern is referred to as searching for a subgraph that matches the pattern.

  FIG. 12 is a diagram illustrating the storage contents of the class score storage unit used instead of the arc score storage unit 13 when the pattern generation unit 14 generates a pattern by another method. The class score storage unit stores a score (class score) of each class defined in a node of the graph G. Replace the instance of each node in the graph G with the class of that node, leave only one of the nodes having the same instance (class), and apply each class score in the class score storage unit to the graph thus formed. And a part of it is as shown in FIG. 14, for example.

  FIG. 13 is a flowchart when the pattern generation unit 14 generates a pattern by another method. For the same reason as in FIG. 6, the database interface 19 and the query will not be mentioned.

  The pattern generation unit 14 first obtains the graph G, replaces the instance of each node of the graph G with the class of the node, and leaves only one of the plurality of nodes having the same instance (class). Are deleted, and each class score of the class score storage unit is set (initial setting) in the graph formed as described above (S21). A part of it is as shown in FIG. 14, for example. In the graph shown in this figure, a node having an instance “org: Organization” is shown as an example of a node having a node instance class (a node class of the graph G) as an instance. In this way, the class score storage unit also stores the score for the class of the class, and here, the class score is also set for such a node.

  Next, for each class, that is, for each node of the graph illustrated in FIG. 14, a class score is calculated by the formula of FIG. 15 (S23). Then, the class score of the graph illustrated in FIG. 14 is updated with this class score (S25). Such calculation update is performed recursively until the class score converges. Thereby, a class score will show importance.

  Next, from the graph G, the pattern generation unit 14 sets a node (keyword node) to which one end of an arc including a predetermined content (for example, “keyword” or “technical term”) as a label is connected to one end position. A path having a predetermined length or less is acquired (S27).

  Next, the pattern generation unit 14 replaces the instance of a node other than the keyword node with a variable for each path (S29).

  Next, the pattern generation unit 14 classifies all paths into path groups (consisting of the same path), and deletes all but one path in each path group (S31).

  Next, the pattern generation unit 14 calculates the path score of each remaining path using the formula of FIG. 16 (S33). Although the class score after recursive calculation is used here, the class score (initial value) before recursive calculation may be used as it is.

  Next, the pattern generation unit 14 acquires a subgraph that matches each path from the graph G (S35).

  Note that S27 to S35 may be as follows.

  That is, in S27, a path having a length equal to or less than a predetermined length is acquired. In S29, an instance of a node of each path is replaced with a variable. In S31, all paths are grouped by path groups (consisting of the same path). In step S33, the path score of each remaining path is calculated according to the equation in FIG. 16, and in step S35, a node at one end position of each path is preliminarily stored. A predetermined instance (for example, “XML” or “Web service”) is set, and a subgraph that matches the path is acquired from the graph G.

  Next, the pattern generation unit 14 recalculates each path score according to the equation of FIG. 17 (S37). Next, the pattern generation unit 14 deletes a path whose path score is equal to or less than a predetermined threshold value from the remaining ones (S39).

  Next, the pattern generation unit 14 assigns the original path score to the remaining path (referred to as a pattern) as the pattern score, and stores it in the pattern database 15 (S41).

  Note that the recalculation of the path score employed in the flowchart of FIG. 13 may be employed in the flowchart of FIG. Further, the method of using the node score for the path score calculation adopted in the flowchart of FIG. 6 may be adopted in the flowchart of FIG.

  In any case, when a pattern including a pattern score is stored in the pattern database 15, a subgraph desired to be finally obtained can be searched.

  FIG. 18 is a sequence diagram when the sub graph is searched from the graph G. For the same reason as in FIG. 6 and the like, the description of acquiring the pattern does not refer to the database interface 19 and the query.

  In the graph search device 1, the input interface generation unit generates an input interface, transmits it to the user terminal 2 (S101), and displays it as shown in FIG. 19 (S103). The user terminal 2 transmits the keywords KW1, KW2 and the class CL specified by the operation here to the graph search device 1 (S105).

  Thereafter, the graph search device 1 uses these keywords and classes. Instead of using the keywords and classes transmitted from the user terminal 2, the graph search device 1 stores the keywords and classes in advance. May be used. The subsequent processing may be performed at a predetermined time. The same applies to the second embodiment.

  In the graph search device 1, the pattern search unit 16 has a node having an instance equal to the keyword KW1 at one end position and a node having a class CL defined at the other end position (first (first) pattern). Is acquired from the pattern database 15 (S107). The pattern search unit 16 has a pattern (referred to as a second pattern) having a node having an instance equal to the keyword KW2 at one end position and a node having a class CL defined at the other end position. 15 (S109).

  In the graph search device, one record including one or more keywords (for example, “XML”, “Web service”, “RDF”, and “Semantic Web”) and one index (for example, “technical keyword”) is recorded. The above-described index storage unit may be provided, and the following may be performed in S107.

  That is, in S107, a record including a keyword equal to the keyword KW1 is searched from this index storage unit, and a node having an instance equal to the index included in the record is at one end position, and the class CL is defined. A pattern (referred to as a first pattern) having the node at the other end position is acquired from the pattern database 15 (S107). The same applies to S109 and S407, S409, and S411 described below.

  Next, the pattern merge unit 17 merges the first pattern and the second pattern at the node at the other end position (S111).

  As a result, patterns as shown in FIG. 20 are generated.

  Next, the graph search unit 18 converts the pattern into a query and transmits it to the database interface 19 (S115), thereby acquiring a subgraph that matches the pattern from the graph G (S117). Thereby, subgraphs as shown in FIG. 21 are searched.

  Then, the graph search unit 18 calculates the subgraph score of each subgraph according to the formula of FIG. 22 (S119).

  Next, the graph search unit 18 sorts the subgraphs in the order of the subgraph scores (S121). Here, a plurality of subgraphs may be merged at a node having a common instance, and the sum of the subgraph scores of the plurality of subgraphs may be calculated as the score of the subgraph.

  Then, the output interface generation unit 110 generates an output interface in which each subgraph is arranged in descending order of the subgraph score, and transmits this to the user terminal 2 (S123) and displays it (S125).

  In the present embodiment, the case where there is one class specified by the operation on the user terminal 2 has been described. However, in the case where there are a plurality of classes, the above-described processing may be performed for each.

  Further, in the present embodiment, the case has been described where the number of keywords specified by the operation on the user terminal 2 is two, but in the case of three or more, the above-described processing may be performed for two of them.

  As described above, according to the first embodiment, a characteristic path pattern included in the graph G has a node having an instance at one end position and a class defined. A pattern storage means (pattern database 15) in which a pattern having a second character at the other end position is stored in advance, and a node (keyword node) having an instance equal to the first keyword (keyword KW1) at one end position, A first pattern as a pattern having a node in which the designated class (class CL) is defined at the other end position is retrieved from the pattern storage means (S107), and an instance equal to the second keyword (keyword KW2) is obtained. Have a node with one class at one end and a node with the specified class (class CL) defined at the other end. The second pattern, which is a pattern, is retrieved from the pattern storage means (S109), and the pattern retrieval means (pattern retrieval unit 16) merges the first pattern and the second pattern at the node at the other end position (S111). Since there is a pattern merging means (pattern merging section 17) and a graph searching means (graph searching section 18) that searches the graph G for a subgraph that matches the pattern merged by the pattern merging means (S117) A subgraph (illustrated in FIG. 21) in which the relationship between the subgraph retrieved from the keyword (illustrated in FIG. 28) and the subgraph retrieved from the second keyword (illustrated in FIG. 28) is clear can be retrieved at high speed. Here, since a pattern is not generated at the time of search, high-speed search can be performed. In this case, it is not necessary to specify a pattern.

  Further, a path having a node connected to the tip of an arc including a predetermined content in a label at one end position is searched from the graph G (S7), and instances of nodes other than the node are replaced with variables. A pattern is generated (S11) and stored in the pattern storage means (S19). Since the pattern generation means (pattern generation unit 14) is provided, more specifically, the score of each label of the arc of the graph G or the node of the graph G Storage means (arc score storage unit 13 or class score storage unit) in which the score of each class defined in 1 is stored, and a node connected to the tip of the arc including a predetermined content in the label from graph G (S7), the instance of a node other than the node is replaced with a variable (S29), and the path of the path is A is calculated based on the score stored in the storage means (S13), the subgraph matching the path is searched from the graph G (S35), and the score is increased according to the number of subgraphs searched (S37). Since the pattern storage means (pattern generation unit 14) stores a pattern having a score larger than a predetermined threshold in the pattern storage means (S41), the above-described pattern search can be performed.

[Second Embodiment]
FIG. 23 is a configuration diagram of the graph search device according to the second embodiment.

  The graph search device 1 </ b> A is connected to a user terminal 2, and a display device 3 is connected to the user terminal 2.

  The graph search device 1A includes a graph database 11, an arc score storage unit 13, a pattern generation unit 14A, a pattern database 15, a pattern search unit 16A, a pattern merge unit 17A, a graph search unit 18, and a database interface 19. And an output interface generation unit 110.

(Operation of Second Embodiment)
FIG. 24 is a flowchart of the pattern generation unit 14A. For the same reason as in FIG. 6 etc., the database interface 19 and the query will not be mentioned.

  First, the pattern generation unit 14A recursively performs steps S1, S3, and S5 (similar to FIG. 6). As a result, the node score of the graph G indicates the importance of the node.

  Next, from the graph G, the pattern generation unit 14A sets a node (keyword node) to which the tip of the arc including a predetermined content (for example, “keyword” or “technical term”) as a label is connected to both end positions. All the paths it has are acquired (S201).

  Note that the pattern generation unit 14A may acquire a path as follows.

  That is, the pattern generation unit 14A acquires a subgraph Gs related only to each instance for each combination of predetermined instances (for example, “XML” and “Web service”). Then, steps S1, S3 and S5 (same as FIG. 6) recursively performed on the graph G are performed on the subgraph Gs.

  In S201, all paths having these instances at both end positions are acquired.

  Since the sub graph Gs is smaller in scale than the graph G, high-speed processing is possible. Further, since the subgraph Gs accurately reflects the relevance of the keyword pair described later, it is possible to search for a subgraph that accurately reflects the relevance of the keyword pair (S419).

  In any case, a path is acquired in S201. In this path, a part other than the keyword node or a part included in the part is called an intermediate part.

  Next, the pattern generation unit 14A calculates the path score of each path using the formula of FIG. 9 (S203).

  Next, the pattern generation unit 14A deletes a path whose path score is equal to or less than a predetermined threshold value from the remaining paths (S205).

  Next, the pattern generation unit 14A assigns the original path score to the remaining path (referred to as a pattern) as the pattern score, and stores it in the pattern database 15 (S207).

  As a result, as shown in FIG. 25, each node has an instance, and each arc has a label, and a pattern is stored in the pattern database 15.

  As described above, when the pattern is stored in the pattern database 15 together with the score, the subgraph desired to be finally obtained can be searched.

  FIG. 26 is a sequence diagram when the sub graph is searched from the graph G. For the same reason as in FIG. 6 and the like, the description of acquiring the pattern does not refer to the database interface 19 and the query.

  In the graph search device 1A, the input interface generation unit 12 generates an input interface, transmits it to the user terminal 2 (S401), and displays it as shown in FIG. 19 (S403). The user terminal 2 transmits the keywords KW1, KW2, and KW3 specified by the operation here to the graph search device 1 (S405).

  In the graph search device 1A, a pattern (first pattern) in which the pattern search unit 16A has a node having an instance equal to the keyword KW1 at one end position and a node having an instance equal to the keyword KW2 at the other end position. Is obtained from the pattern database 15 (S407). The pattern search unit 16A also has a pattern (referred to as a second pattern) having a node having an instance equal to the keyword KW1 at one end position and a node having an instance equal to the keyword KW3 at the other end position. Obtained from the database 15 (S409). The pattern search unit 16A also has a pattern (referred to as a third pattern) having a node having an instance equal to the keyword KW2 at one end position and a node having an instance equal to the keyword KW3 at the other end position. Obtained from the database 15 (S411).

  Next, the pattern merging unit 17A merges the first pattern and the second pattern at the common part if there is a common part included in common (S413). The pattern merging unit 17A also performs the same for the first pattern and the third pattern, and the second pattern and the third pattern (S415). Thereby, a pattern as shown in FIG. 27 is generated.

  Next, the graph search unit 18 converts the pattern into a query and transmits it to the database interface 19 (S417), thereby acquiring a subgraph matching the pattern from the graph G (S419). As a result, the same subgraph as the pattern as shown in FIG. 27 is retrieved.

  Then, the graph search unit 18 calculates the subgraph score of each subgraph according to the formula of FIG. 22 (S421).

  Next, the graph search unit 18 sorts the subgraphs in the order of the subgraph scores (S423). Here, a plurality of subgraphs may be merged at a node having a common instance, and the sum of the subgraph scores of the plurality of subgraphs may be calculated as the score of the subgraph.

  Then, the output interface generation unit 110 generates an output interface in which each subgraph is arranged in descending order of the subgraph score, and transmits this to the user terminal 2 (S425) to display it (S427).

  In the present embodiment, the case where the number of keywords specified by the operation on the user terminal 2 is three has been described, but the same processing may be performed even when there are four or more keywords.

  As described above, according to the second embodiment, a characteristic path pattern included in the graph G has nodes having instances at both end positions, and each arc has a label. A pattern storage means (pattern database 15) in which a pattern is stored in advance and a node having an instance equal to the first keyword (keyword KW1) at one end position and equal to the second keyword (keyword KW2) A first pattern which is a pattern having a node having an instance at the other end position is retrieved from the pattern storage means (S407), a node having an instance equal to the first keyword is located at one end position, A pattern having a node having an instance equal to the keyword (keyword KW3) at the other end position A pattern is retrieved from the pattern storage means (S409), and the pattern retrieval means (pattern retrieval unit 16A) and the first pattern and the second pattern are merged at a common part included in each (S411) pattern merge means ( Since the pattern merging unit 17A) and the graph G are searched for a subgraph that matches the pattern merged by the pattern merging unit (S419), the first keyword and the second There is a relationship between a subgraph retrieved from the keyword (a subgraph equal to the pattern in FIG. 25A) and a subgraph retrieved from the first keyword and the third keyword (a subgraph equal to the pattern in FIG. 25A). It is possible to search a clear subgraph (illustrated in FIG. 27) at high speed. That. Here, since a pattern is not generated at the time of search, high-speed search can be performed. In this case, it is not necessary to specify a pattern.

  Further, from the graph G, a pattern that is a path having a node connected to the tip of an arc whose predetermined content is included in the label at both end positions is generated (S205), and stored in the pattern storage means (S207). Since the generation means is provided, more specifically, the storage means (the arc score storage unit 13 or the class score storage) in which the score of each label of the arc of the graph G or the score of each class defined in the node of the graph G is stored. Part) and the graph G are searched for a path having a node connected to the tip of the arc having a predetermined content in the label at both end positions (S201), and the score of the path is stored in the storage means. Calculation is performed based on the score (S203), and a pattern whose score is larger than a predetermined threshold value is stored in the pattern storage means (S207). Because and a turn generating means (pattern generation unit 14A), will allow the pattern search described above.

It is a lineblock diagram of a graph search device concerning a 1st embodiment. 6 is a diagram illustrating a part of a graph G. FIG. 6 is a diagram illustrating a part of a data group for displaying a graph G. FIG. It is the figure which showed a part of arc score memorize | stored in the arc score memory | storage part. It is the figure which applied a part of arc score to the graph. It is the figure which showed the flowchart when producing | generating a pattern. It is the figure which showed the calculation formula of a node score. It is a figure which shows the graph made into the example of node score calculation. It is the figure which showed the calculation formula of a pass score. It is the figure which showed the calculation formula of the score of a pass group. It is the figure which illustrated the produced | generated pattern. It is the figure which showed a part of class score memorize | stored in the class score memory | storage part. It is the figure which showed another flowchart which produces | generates a pattern. It is the figure which applied a part of class score to the graph. It is the figure which showed the calculation formula of a class score. It is the figure which showed the calculation formula of a pass score. It is the figure which showed the recalculation formula of the pass score. It is a sequence diagram of a 1st embodiment. It is the figure which illustrated the displayed interface for input. It is the figure which illustrated the merged pattern. It is the figure which illustrated the searched subgraph. It is the figure which showed the calculation formula of a subgraph score. It is a block diagram of the graph search apparatus which concerns on 2nd Embodiment. It is the figure which showed the flowchart when producing | generating a pattern. It is the figure which illustrated the produced | generated pattern. It is a sequence diagram of a 2nd embodiment. It is the figure which illustrated the merged pattern. It is the figure which illustrated the subgraph searched with the conventional graph search device.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1, 1A ... Graph search apparatus 2 ... User terminal 3 ... Display apparatus 11 ... Graph database 12 ... Input interface generation part 13 ... Arc score memory | storage part 14, 14A ... Pattern generation part 15 ... Pattern database 16, 16A ... Pattern search part 17, 17A ... Pattern merge unit 18 ... Graph search unit 19 ... Database interface 110 ... Interface generation unit for output

Claims (8)

A characteristic path pattern included in a graph in which nodes having different instances are connected by an arc having a label and the class of the instance is defined, and has a node having an instance at one end position. Pattern storing means for storing a pattern in advance, having a node having a class defined at the other end position, each arc having a label;
A first pattern which is a pattern having a node having an instance equal to the first keyword at one end position and a node having a specified class defined at the other end position is retrieved from the pattern storage means. A second pattern which is a pattern having a node having an instance equal to the second keyword at one end position and a node having a specified class defined at the other end position is retrieved from the pattern storage means. Pattern search means to perform,
Pattern merging means for merging the first pattern and the second pattern at a node at the other end position;
A graph search device comprising: a graph search unit that searches the graph for a subgraph that matches the pattern merged by the pattern merge unit. From the graph, search for a path having a node connected to the tip of an arc including a predetermined content in a label at one end position, and generate a pattern in which instances of nodes other than the node are replaced with variables, The graph search device according to claim 1, further comprising: a pattern generation unit that is stored in the pattern storage unit. Storage means for storing a score of each label of the arc of the graph or a score of each class defined in a node of the graph;
From the graph, search for a path having a node to which an arc tip including a predetermined content in a label is connected at one end position, replace an instance of a node other than the node with a variable, and calculate the score of the path Calculate based on the score stored in the storage means, search the graph for a subgraph that matches the path, increase the score according to the number of subgraphs searched, and the score is greater than a predetermined threshold The graph search device according to claim 1, further comprising: a pattern generation unit that stores a pattern in the pattern storage unit. A characteristic path pattern included in a graph in which nodes having different instances are connected by an arc having a label and the class of the instance is defined, and the node having the instance is present at both end positions. Pattern storing means for storing patterns in advance, each arc having a label;
A first pattern which is a pattern having a node having an instance equal to the first keyword at one end position and a node having an instance equal to the second keyword at the other end position is retrieved from the pattern storage means. The pattern storage means has a second pattern as a pattern having a node having an instance equal to the first keyword at one end position and a node having an instance equal to the third keyword at the other end position. Pattern search means to search from,
Pattern merging means for merging the first pattern and the second pattern at a common part included in common with each other;
A graph search device comprising: a graph search unit that searches the graph for a subgraph that matches the pattern merged by the pattern merge unit. Pattern generating means for generating, from the graph, a pattern that is a path having a node connected to the tip of an arc whose predetermined content is included in a label at both end positions, and storing the pattern in the pattern storage means. The graph search device according to claim 4. Storage means for storing a score of each label of the arc of the graph or a score of each class defined in a node of the graph;
From the graph, a path having a node connected to the tip of an arc including a predetermined content in a label at both end positions is searched, and a score of the path is calculated based on the score stored in the storage unit. The graph search device according to claim 4, further comprising: a pattern generation unit that stores a pattern having a score larger than a predetermined threshold in the pattern storage unit.   A computer program for operating a computer as the graph search device according to claim 1.   A recording medium in which the computer program according to claim 7 is stored.

Images