JP2012038023A - Query generation device and operation method therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a query generation device which can generate a query having versatility.SOLUTION: As representative paths have increased in number, the number of queries relevant to a group G13 has increased to a threshold value (1, for example) or more. Thus, when the numbers of queries for all groups have become equal to or higher than the threshold value, queries consisting solely of combined representative paths are retrieved.

Description

  The present invention relates to a query generation device and an operation method thereof.

  In recent years, the fields of utilization of the Semantic Web and Linked Data have expanded. With such a technology, disparate data sets held at a plurality of sites are combined, and data that could not be searched before can be searched.

  To obtain a word related to a certain keyword using such a data set, include the keyword in the query generated from the data set, and search the data set for data that matches the query. Extract the target word from the data.

  For consolidated datasets, many queries are generated, so the number of queries needs to be narrowed down.

  However, if the number of queries is simply narrowed down, the target word or the like can be obtained when a certain keyword is used, but it cannot be obtained when another keyword is used. In other words, the narrowed query lacks diversity.

  Even if diversity can be ensured, the obtained word or the like is not always reliable.

  Conventionally, among many query graph patterns, there is one that displays the one in line with the user's interest as the top ranking (Non-Patent Document 1).

  However, the problem of who performs the weighting to reflect the user's interests under what rule has not been solved.

  In addition, there is one that narrows down the number of query graph patterns by limiting the length of a path constituting the query graph pattern (Non-Patent Document 2).

  However, there is a possibility that information obtained only by a search using a query graph pattern composed of a long path cannot be obtained.

  In addition, information obtained by limiting the path length is highly related to the keyword, but is not necessarily reliable.

JP 2006-313501 A

“SemRank: ranking complex relationship search results on the semantic web”, [online], [searched July 13, 2010], Internet <URL: http://lsdis.cs.uga.edu/library/download/AMS05 -WWW2005.pdf> “What have Innsbruck and Leipzig in common? Extracting Semantics from Wiki Content”, [online], [searched July 13, 2010], Internet <URL: http://www.informatik.uni-leipzig.de/~ auer / publication / ExtractingSemantics.pdf>

  The present invention has been made in view of the above problems, and an object of the present invention is to provide a query generation device capable of generating a query having diversity and an operation method thereof.

  It is a further object to be able to generate queries that are suitable for obtaining reliable words and the like and have diversity.

  In order to solve the above-mentioned problem, the first aspect of the present invention is based on a data set obtained by merging two data sets composed of data having two or more nodes having an instance and a class indicating the instance. Obtaining data connecting a node having a predetermined class as indicating a keyword for searching data and a node having a predetermined class as indicating a target instance in the searched data; A query storage means for storing a query obtained by replacing an instance of data with a variable, and a node corresponding to a node common to the two data sets in a path consisting of only one route by a node included in the query A path generation means for dividing the path, a path storage means for storing the divided path, Path classifying means for classifying the path of the path storage unit into a group for each set of classes indicating instances of nodes at both ends of the path, and a predetermined search as an instance of the node at one node of the path of the path storage unit Data search means for substituting a word and searching for data that matches the path other than variables from the merged data set, and for each path in the group in each group, search by target path To obtain a set of instances of the other end node of the obtained data and a set of instances of the other end node of the data searched by the non-target path, and calculate the number of instances of the former set of the union of each set. A representative that calculates a value divided by the number of instances and selects a path corresponding to the value higher than a predetermined threshold as a representative path from the group Path merging means for merging paths representative of different groups, and for the merged path to connect the node corresponding to the keyword and the node corresponding to the target instance; and A representative path storage means for storing the merged path and a query including the merged path are searched from the query storage means, and the search is performed for each group set to which each path before merging belongs. First query search means for instructing the representative path selection means to adjust the threshold value, and the number of searched queries is determined in advance so that the number of queries obtained is equal to or greater than a predetermined number. And a second query search means for searching a query consisting only of the merged path from the query storage means. The query generation device that performs the processing is used as a solving means.

  For example, the second query search unit searches the query storage unit for a query including only two or more merged paths.

  The second aspect of the present invention shows a keyword for searching for data from a data set obtained by merging two data sets composed of data having two or more nodes each having an instance and a class indicating the instance. Obtain data connecting a node having a predetermined class as a thing and a node having a predetermined class to indicate a target instance in the retrieved data, and replace the data instance with a variable. An operation method of a query generation device comprising a query storage means for storing a received query, wherein a path consisting of only one route by a node included in the query corresponds to a node common to the two data sets A path generation step of dividing by a node and storing the divided path in a path storage means A path classification step of classifying paths of the path storage unit into groups for each class set indicating instances of nodes at both ends of the path, and a node at one end of the path of the path storage unit as an instance of the node A data search step of searching the merged data set for data that matches a path other than a variable, and each path in the group in each group as a target. To obtain the set of instances of the node at the other end of the data searched for and the set of instances of the node at the other end of the data searched by the non-target path, and add the number of instances of the former set to the sum of each set Calculate the value divided by the number of instances in the set and select the path corresponding to the value higher than the predetermined threshold from the group as the representative path The representative path selection step and the representative paths of different groups are merged so that the merged path connects the node corresponding to the keyword and the node corresponding to the target instance. A path merging step of storing the path in the representative path storage means, and a query including the merged path is searched from the query storage means, and the search is performed for each group set to which each path before merging belongs. A first query search step for instructing the representative path selection means to adjust a threshold value, and a number of searched queries so that the number of queries is equal to or greater than a predetermined number. A second query search step of searching for a query consisting only of the merged path from the query storage means. The feature generation method of the query generation device is used as the solution.

  For example, in the second query search step, a query including only two or more merged paths is searched from the query storage unit.

  According to the present invention, the number of retrieved queries is set to be equal to or more than a predetermined number for each group set to which the representative path before merging belongs, so that a query having diversity is obtained. be able to.

It is a block diagram of the data search apparatus containing the component of the query graph pattern generation apparatus which concerns on this Embodiment. It is a figure which shows the data which comprise the data set D1. It is the figure which replaced and represented the connection relationship between the RDF instances of the data set D3 with the relationship between classes. It is a figure which shows an example of a query. It is a figure which shows the part of each query after a division | segmentation. It is a figure which shows the mode of the division | segmentation of a query. It is a figure which shows the mode of the classification | category of the path | pass produced | generated from the data set D1. It is a figure which shows the mode of the classification | category of the path | pass produced | generated from the data set D2. It is a figure which shows a mode that the search word was included in the path | pass. It is a figure which shows the path | pass including a search word, and the data searched. It is a figure which shows a mode that the representative path | pass is selected from the group G13. It is a figure which shows a mode that a path | pass is merged. It is a figure which shows the result of the total of the 1st search by what merged the representative path | pass. It is a figure which shows the result of the total of the search by what merged arbitrary paths. It is a figure which shows the result of the total of the 2nd search by what merged the representative path | pass. It is a figure which shows the path | route H1, H2, and the query Q101 searched from these paths. It is a figure which shows a mode that a keyword is included in a query. The query Q101 including “Internet” also illustrates all the class values, properties, and directional characteristics. It is a figure which shows the example of the data searched from what includes the keyword in the query.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. Note that RDF (Resource Description Framework) technical terms are used as some terms.

  FIG. 1 is a block diagram of a data search apparatus including components of the query generation apparatus according to the present embodiment.

  The data search device 1 includes a data set storage unit 11 that stores the data set D1, a data set storage unit 12 that stores the data set D2, a data set merge unit 13 that merges (combines) the data sets D1 and D2, A data set storage unit 14 for storing the merged data set D3, a query generation unit 15 for generating a query graph pattern (referred to as a query for short) from the data set D3 based on parameters input from the outside, A query storage unit 16 that stores a query; a path generation unit 17 that generates a path from the query; a path storage unit 18 that stores a generated path; and a path classification unit 19 that classifies the path into a plurality of groups. A data search unit that searches the data set D3 for data that matches the path to which the search word in the data set D3 is applied. 01, a representative path selection unit 102 that selects a representative path from the group based on the retrieved data, a path merge unit 103 that merges the paths, a merged representative path, and a single representative that could not be merged A representative path storage unit 104 in which paths are stored, a path storage unit 105 in which arbitrary merged paths and a single arbitrary path that cannot be merged are stored, and paths in the representative path storage unit 104 and the path storage unit 105 A query search unit 106 that searches the query storage unit 16 for a query that includes a path that matches the query path, and a query search unit 107 that searches the query storage unit 16 for a query that includes only a plurality of paths in the representative path storage unit 104. A keyword is substituted for the query stored in the query storage unit 108 and the query storage unit 108 to match the query after the substitution. And a data retrieval unit 109 for retrieving data from a data set D3.

  FIG. 2 is a diagram illustrating data constituting the data set D1.

  As shown in FIGS. 2A and 2B, the minimum unit data includes information indicating one of two nodes (referred to as a node), information indicating the other (referred to as a node), and an arc connecting the nodes. It is the directed graph data with a label which consists of the information to show (it is called an arc).

  The node has a value of “instance” (“Internet”, “Graduation Research Project”, “Ichiro Tanaka”, etc.). The node has a value of a class (“technical term”, “research subject”, “faculty member”, etc.) indicating the type of instance. An instance of a node having the class “research project” is a “graduation research project” belonging to the type of “research project”. An instance of a node having the class “faculty and staff” is “Ichiro Tanaka” belonging to the type “faculty and staff”.

  Note that an instance of a node (keyword node) having the class “technical term” is a determination target of whether or not it matches a keyword input from the outside, and the keyword node is treated as a special node in the data search device 1. ing.

  The arc is directional, that is, it is directed to either node and has a property value. The direction of the arc is in accordance with the node relationship.

  The data in FIG. 2A is obtained by the fact that “there is a graduation research topic related to the Internet”, and the direction of the arc corresponds to the operation of deriving the “Internet” from the “gradation research problem”. The property is “technical term” because “there is a graduation research subject related to the Internet” can be rephrased as “the Internet is a technical term for the graduation research subject”.

  FIG. 2C is a result of merging (joining) the data of FIGS. 2A and 2B.

  In the data set D1, when two data have the same node (“Graduation Research Project” node), the two data are associated with each other at the node. For convenience, it is said that the two data are merged. FIG. 2C shows the data after merging.

  The data set D1 has a large number of data that can be merged, and the merge is repeated to form a mesh.

  The data set D1 is university-related data, for example.

  On the other hand, the data set D2 is, for example, SNS (Social Networking Service). Although not shown, the data set D2 also includes data similar to the data set D1 and has a mesh shape.

  Returning to FIG. 1, first, the data set merging unit 13 identifies nodes that exist in the data sets D1 and D2 and have the same class and instance, and merges the data sets D1 and D2 at the nodes. The data set merge unit 13 causes the data set storage unit 14 to store the merged data set D3.

  FIG. 3 is a diagram in which the connection relationship between the RDF instances of the data set D3 is replaced with the relationship between classes. In this figure, for example, there is an RDF triple in which an instance having a technical term as a class and an instance having a research subject as a class are connected by an arc.

  FIG. 3 shows the connection relationship between the RDF instances of the data set D3 merged with the class “technical terms” and the class “faculty and staff” replaced with the relationship between classes. Each node (merge node) is in each of the data sets D1 and D2.

  Returning to FIG. 1, the data search apparatus 1 inputs the keyword “Internet” and the parameter “faculty and staff” from the outside prior to examining the name of “faculty and staff” familiar with the keyword “Internet” (for example, “Ichiro Tanaka”). .

  The query generation unit 15 generates a query from the data set D <b> 3 based on the parameter “faculty / staff”, and stores the generated query in the query storage unit 16.

  Specifically, the query generation unit 15 first includes a special node, that is, a node having a class “technical term” (keyword node), a node having a class “faculty member” equal to a parameter (goal node), and these nodes. All the data consisting of one or more arcs to be connected or arcs and nodes connecting these nodes through one or more paths are acquired from the data set D3.

  Next, the query generation unit 15 replaces the instance of each node of each data with a variable corresponding to the class of the node. Note that the query generation unit 15 leaves the value and property indicating the directivity of the arc of the data without being replaced. All that is obtained is a query.

  Returning to FIG. 1, the path generation unit 17 generates as many paths as possible from all the queries stored in the query storage unit 16, and stores the generated paths in the path storage unit 18.

  FIG. 4 is a diagram illustrating an example of a query.

  Specifically, the path generation unit 17 first searches for a query connecting a keyword node and a goal node with a plurality of paths (consisting of arcs or arcs and nodes) as shown in FIG.

  In the query Q1, a node having a variable “? Technical term” corresponding to a node having the class “technical term” is a keyword node, and a node having a variable “? Teacher” corresponding to the class “teacher” is a goal node. . Similarly, other nodes have a variable consisting of “?” And a class value (such as “research subject”).

  FIG. 5 is a diagram showing a part of each query after division.

  Next, as shown in FIGS. 5A and 5B, the path generation unit 17 divides the corresponding query for each path. Each query after the division is a query that connects the keyword node and the goal node through one route.

  Next, the path generation unit 17 further divides the divided query and the query that originally connects the keyword node and the goal node with one route.

  FIG. 6 is a diagram illustrating how a query is divided.

  Specifically, as shown in FIG. 6, the path generation unit 17 is a node (intermediate node) other than a keyword node and a goal node, and a merge node (class “technical term” or “faculty member”). A query having a node (merge node) corresponding to (node having a node) is searched.

  Here, for example, when the data sets D1 and D2 are merged, the path generation unit 17 searches for a query having a node stored as a merge node as an intermediate node.

  Alternatively, for example, if the identification information (name or the like) indicating the corresponding data set is given to the node of each data set in advance, the path generation unit 17 uses a node having both identification information, that is, a merge node as an intermediate node. Find the query you have.

  The node having the variable “? Faculty staff” in FIG. 6 is a merge node obtained by replacing the instance of the merge node having the class “faculty staff” in FIG. 3 with the variable “? Faculty staff”.

  Next, as illustrated in FIG. 6, the path generation unit 17 divides the corresponding query at the merge node. The divided query is a query that connects the keyword node and the merge node through one route.

  If there is a merge node in the divided query, the path generation unit 17 further divides the query by the merge node. The query that originally did not have an intermediate node and merge node and the query after splitting are all paths.

  Returning to FIG. 1, the path classification unit 19 then classifies all the paths in the path storage unit 18 into a plurality of groups.

  Specifically, the path classification unit 19 first classifies all the paths into a group of paths generated from the data set D1 and a group of paths generated from the data set D2.

  Next, the path classification unit 19 classifies the paths of each group according to combinations of class values possessed by nodes at both ends.

  FIG. 7 is a diagram showing a state of classification of paths generated from the data set D1.

  In the path generated from the data set D1, the value of the class of the nodes at both ends is either “technical term” or “faculty and staff”, so the path has a group G11 whose class values at both ends are both “technical term”. The class values at both ends are classified into a group G12 in which “faculty and staff” are included, and the group G13 in which the value at one end is “technical term” and the value at the other end is “faculty and staff”.

  The group G11 includes paths G111 and 112. The group G12 includes paths G121 and 122. The group G13 includes paths G131, 132, and 133.

  FIG. 8 is a diagram showing a state of classification of paths generated from the data set D2.

  Similarly, the paths generated from the data set D2 are also classified into three groups G21, G22, and G23.

  The group G21 includes paths G211, 212, and 213. The group G22 includes paths G221 and 222. The group G23 includes paths G231, 232, and 233.

  Therefore, the paths are classified into 6 groups.

  The path classification unit 19 assigns the ID of the group to which the path belongs to the path.

  Returning to FIG. 1, the data search unit 101 applies the search word in the data set D3 to the path of the path storage unit 18, and searches the data set D3 for data that matches the path after application.

  First, for each class value of the end node in the path, the data search unit 101 obtains one or more (for convenience, 2) instances of a node having the same class as the value so as not to be duplicated from the data set D3. To do. The acquired instance is hereinafter referred to as a search word.

  In the example of FIGS. 7 and 8, there are two class values, “technical term” and “faculty / staff”. Therefore, the data search unit 101 uses two search words (hereinafter referred to as search words SW11 and SW12) from “technical term”. ) And two search words (hereinafter referred to as search words SW21 and SW22) are acquired from the “faculty and staff”.

  The data search unit 101 first performs the following for each group.

  FIG. 9 is a diagram illustrating a state in which a search word is included in the path.

  That is, as shown in FIG. 9, the data search unit 101 generates a path including one of the search words SW11 and SW12 as one end instance for the group G11. In this way, the paths P11 and P12 are generated from the path G111, the paths P21 and P22 are generated from the path G112, and the data search unit 101 performs the same processing for the other groups G12 to G23.

  FIG. 10 is a diagram illustrating a path including a search word and data to be searched.

  Next, the data search unit 101 searches the data set D3 for all data that matches all of the paths P11 other than variables (such as “? Research Project”) as shown in FIG. In FIG. 10, the search word SW11 is “patent”. FIG. 10 also shows all of the values of class, properties, and directivity that are omitted in FIG. In addition, the data search unit 101 performs the same search for the other paths P12, 21, and 22.

  In addition, the data search unit 101 performs the same processing for the other groups G12 to G23.

  Then, the data search unit 101 transmits the searched data to the representative path selection unit 102.

  Returning to FIG. 1, the representative path selection unit 102 selects a representative path from the group based on the data.

  The representative path selection unit 102 calculates a set M of values for each path belonging to the group in each group including a plurality of paths. The set M is a set of characteristic values m indicating characteristics that the target path (target path) has for each path (non-target path) that is not the target of the same group.

  For example, in the group G11, the set M targeted by the path G111 includes only the characteristic value m indicating the characteristic that the path G111 has with respect to the path G112. Further, the set M targeted by the path G112 includes only the characteristic value m indicating the characteristic of the path G112 with respect to the path G111.

  For example, in the group G13, the set M that is the target of the path G131 includes the characteristic value m indicating the characteristic that the path G131 has for the path G132, and the characteristic value m that indicates the characteristic that the path G131 has for the path G133. The set M targeted by the path G132 includes a characteristic value m indicating a characteristic of the path G132 with respect to the path G131, and a characteristic value m indicating a characteristic of the path G132 with respect to the path G133. The set M targeted by the path G133 includes a characteristic value m indicating a characteristic that the path G133 has for the path G132, and a characteristic value m indicating a characteristic that the path G133 has for the path G131.

The representative path selection unit 102 calculates such a characteristic value m by the following equation (1).

However, ki is the search word Q1 (ki) used for the corresponding group is a set of data searched from the target path after application of the search word ki,
Q2 (ki) is a set of data searched from the non-target path after application of the search word ki,
Q1 (ki) ∪Q2 (ki) is the union of the set Q1 (ki) and the set Q2 (ki),
| Q1 (ki) ∪Q2 (ki) | is the number of data included in the union Q1 (ki) ∪Q2 (ki)
| Q1 (ki) | is the number of data included in the set Q1 (ki),
Σ is the sum of the values obtained by dividing the number of data | Q1 (ki) | by the number of data | Q1 (ki) ∪Q2 (ki) |
| K | is the number of search words included in the set M of each search word ki.

  That is, the characteristic value m is a value indicating how much a set of data retrieved from the non-target path can be covered by a set of data retrieved from the target path.

  Next, for example, in the group G11, the representative path selection unit 102 selects a set M having a characteristic value m larger than a predetermined threshold from two sets, the path G111 being the target set M and the path G112 being the target set M. In the group G11, a path corresponding to the selected set M is selected as a representative.

  For example, in the group G13, the representative path selection unit 102 selects only the characteristic value m that is larger than the threshold value from three, the path G131 being the target set M, the path G132 being the target set M, and the path G133 being the target set M. And a group G13 selects a path corresponding to the selected set M as a representative.

  FIG. 11 is a diagram illustrating a state in which a representative path is selected from the group G13.

  The set M targeted for the path G131 includes a characteristic value 29.3% for the path G132 and a characteristic value 64.7% for the path G133.

  The set M targeted for the path G132 includes a characteristic value 71.0% for the path G131 and a characteristic value 82.3% for the path G133.

  The set M targeted for the path G133 includes a characteristic value 36.0% for the path G131 and a characteristic value 20.3% for the path G132.

  Therefore, when the threshold is 70%, the target set M for the path G132 is selected, and the path G132 is selected as a representative.

  The representative path selection unit 102 assigns a flag indicating that to the representative path.

  The representative path selection unit 102 may obtain an average or maximum value of the characteristic values m in each set M, and may select a path corresponding to the largest value as a representative path.

  Returning to FIG. 1, the path merge unit 103 first has two representative paths extracted from the path storage unit 18 having the same class of node at one end of one path and the class of node at one end of the other path. If the class of the node at the other end of one path is “technical term” and the class of the node at the other end of the other path is “faculty / staff”, merge the two paths with the nodes of the same class, The representative path storage unit 104 stores all of the merged paths and single paths whose original class is “technical terms” and whose other node class is “faculty”. .

  FIG. 12 is a diagram illustrating how paths are merged.

  For example, as shown in FIG. 12, if the path G131 and the path G122 are representative, the path merge unit 103 merges these with nodes having the class “faculty and staff” to generate a path H1.

  For example, the classes at both ends of the path of the group G11 are “technical terms” and the classes at both ends of the path of the group G12 are “faculty and staff”. Therefore, these paths cannot be merged and are stored in the representative path storage unit 104. Is not remembered.

  For example, since the classes at both ends of the path of the group G11 are “technical terms”, these paths are not stored in the representative path storage unit 104 even as independent paths.

  Returning to FIG. 1, the path merge unit 103 performs the same processing for an arbitrary path regardless of whether or not it is a representative.

  The path merge unit 103 takes two arbitrary paths extracted from the path storage unit 18 as one end node class of one path and one end node class of the other path and the other end of one path. If the class of the node is “technical term” and the class of the node at the other end of the other path is “faculty and staff”, then the two paths are merged with nodes of the same class, and this is done in all possible combinations, All of the merged paths and the single paths whose original one-end class is “technical term” and whose other-end node class is “faculty and staff” are stored in the path storage unit 105. The merging method is the same as the merging of representative paths.

  Next, the query search unit 106 searches the query storage unit 16 for a query including a path that matches the path in the representative path storage unit 104, and the number of searched queries is the number of the group or group to which the corresponding path belongs. Aggregate for each combination.

  FIG. 13 is a diagram illustrating a result of a total of the first search by merging representative paths.

  As shown in FIG. 13, for example, the paths of the group G11 cannot be merged, and therefore the search itself is not performed for the group of the groups G11. This is indicated by crosses at the intersections of G11 columns and G11 rows.

  Further, the path of the group G11 and the path of the group G12 cannot be merged, and therefore the search itself is not performed for the group G11 and G12. This is indicated by crosses at the intersections of G11 columns and G12 rows.

  Also, since the class at both ends of the path of the group G11 is “technical term”, the intersection of the column of G11 and the bottom row is indicated by a cross.

  On the other hand, the path of the group G11 and the path of the group G13 can be merged, and the number of retrieved queries is “2”.

  Further, the path of the group G12 and the path of the group G13 can be merged, and the number of retrieved queries is “0”.

  FIG. 13 shows the presence / absence of a search for other groups or combinations of groups, and the number of queries searched.

  Further, the query search unit 106 searches the query storage unit 16 for a query including a path that matches the path stored in the path storage unit 105, and determines the number of searched queries for each group or combination of groups to which the corresponding path belongs. To sum up.

  FIG. 14 is a diagram illustrating a result of totaling search by merging arbitrary paths.

  FIG. 14 shows that the path of the group G12 and the path of the group G13 can be merged, and the number of retrieved queries is “41”.

  As shown in FIG. 13, the query search unit 106 specifies a group at a place where the number of searched queries is zero. In FIG. 13, there are two places where the number of queries is 0, and since the group G13 is common, the group G13 is specified.

  Further, if the number of queries in FIG. 14 corresponding to a place where the number of searched queries is 0 is equal to or greater than a predetermined threshold (for example, 1), the query search unit 106 identifies the specified group to the representative path selection unit 102. For G13, the threshold used to select the representative path is lowered, and an instruction is given to select the representative path again in the same manner.

  In FIG. 13, there are two places where the number of queries is 0, and in FIG. 14, the number of queries at the corresponding places is 41 and 15. For example, if the threshold is 1, the query search unit 106 represents the representative path selection unit. An instruction is issued to 102.

  For example, in FIG. 10, the representative path selection unit 102 changes the threshold value to, for example, 25%, and the paths G131 and 132 are selected as representatives. That is, the number of representatives increases to two.

  When the representative path is reselected, the path merge unit 103 merges the representative paths again.

  When the representative paths are merged again, the query search unit 106 performs aggregation again.

  FIG. 15 is a diagram showing the result of the second search totaling by merging representative paths.

  As the number of representative paths increases, the number of queries at locations related to the group G13 increases, which is equal to or greater than a threshold (for example, 1).

  Returning to FIG. 1, the query search unit 107 searches the query storage unit 16 for a query including only a plurality of paths in the representative path storage unit 104 when the number of queries is equal to or greater than the threshold value. The processed query is stored in the query storage unit 108.

  FIG. 16 is a diagram illustrating paths H1 and H2 and a query Q101 searched from these paths.

  The query Q101 is a query that connects a keyword node and a goal node with a plurality of paths (consisting of arcs or arcs and nodes). One path corresponds to the path H1, and the other path corresponds to the path H2. There may be three or more routes.

  The query search unit 107 searches all the queries that satisfy such conditions so as to search the query Q101 from the paths H1 and H2.

  Returning to FIG. 1, next, the data search unit 109 substitutes the keyword “Internet” for the query in the query storage unit 108, and searches the data set D3 for data that matches the query after the substitution.

  Specifically, the data search unit 109 reads the query, includes, for example, “Internet” as an instance of the keyword node in the query, and searches the data set D3 for data that matches the query after substitution.

  FIG. 17 is a diagram illustrating how keywords are included in a query.

  As shown in FIG. 17, the data search unit 109 includes, for example, “Internet” as an instance in the keyword node of the query Q101.

  FIG. 18 illustrates all of the class values, properties, and directivity for the query Q101 including “Internet”.

  FIG. 19 is a diagram illustrating an example of data retrieved from a query including a keyword.

  The data search unit 109 searches the data set D3 for data such as data D203 in FIG. 19 that matches all of the queries Q101 in FIG. 18 except for variables (such as “? Paper”). This data indicates that the name of “faculty and staff” who is familiar with “Internet” is “Ichiro Tanaka”.

  Therefore, the data search unit 109 outputs the goal node instance to the outside.

  As a result, the name of the “faculty and staff” who is familiar with “Internet” is “Ichiro Tanaka”.

  As described above, the data search device 1 is merged with the data set merging means (13) for merging two data sets composed of data having two or more nodes having an instance and a class indicating the instance. A data set storage means (14) for storing the data set, a node (keyword node) having a class predetermined as indicating a keyword for searching for data from the data set, and the searched data Query generation means (15) for acquiring data connecting a node (goal node) having a predetermined class as indicating a target instance, and generating a query in which the instance of the data is replaced with a variable, the query Is stored in the query storage means (16) for storing A path generation means (17) for dividing a path consisting of only one route by a node at a node corresponding to a node common to the two data sets, and a path storage means (18) for storing the divided paths. ), A path classification means (19) for classifying the paths of the path storage unit into groups for each class set indicating instances of nodes at both ends of the path, and a node at one end of the path of the path storage unit A predetermined search word is substituted as an instance of the node, data search means (101) for searching the merged data set for data that matches the path other than variables, and each group in the group A set of instances of the node at the other end of the data searched for by the target path and the other of the data searched by the non-target path And a value obtained by dividing the number of instances of the former set by the number of instances of the union of each set is calculated by Equation (1), and corresponds to a value higher than a predetermined threshold value. Representative path selection means (102) for selecting a path from the group as a representative path, and merging the representative paths of different groups, and the merged path corresponds to the node corresponding to the keyword and the target instance. A path merging unit (103) for connecting nodes, a representative path storage unit (104) for storing the merged path, and a query including the merged path from the query storage unit are searched. Then, for each set of groups to which each path before merging belongs, the number of retrieved queries is equal to or greater than a predetermined number. First query search means (106) for instructing the representative path selection means to adjust the threshold value, and if the number of searched queries exceeds a predetermined number, from the query storage unit Second query search means (107) for searching for a query consisting only of the merged path.

  In other words, since the number of retrieved queries is greater than or equal to a predetermined number for each group set to which the representative path before merging belongs, it is possible to obtain diverse queries.

  In addition, since the second query search unit searches the query storage unit for a query including only two or more merged paths, a query suitable for obtaining a target instance having reliability can be obtained. Can do.

  The computer program for causing the computer to function as the query graph pattern generation device according to the present embodiment can be recorded on a computer-readable recording medium such as a semiconductor memory, a magnetic disk, an optical disk, a magneto-optical disk, or a magnetic tape, Further, it can be widely distributed by being transmitted via a communication network such as the Internet.

DESCRIPTION OF SYMBOLS 1 ... Data search device 11 ... Data set storage part 12 ... Data set storage part 13 ... Data set merge part 14 ... Data set storage part 15 ... Query generation part 16 ... Query storage part 17 ... Path generation part 18 ... Path storage part 19 ... path classification unit 101 ... data search unit 102 ... representative path selection unit 103 ... path merge unit 104 ... representative path storage unit 105 ... path storage unit 106 ... query search unit 107 ... query search unit 108 ... query storage unit 109 ... data search unit

Claims (5)

A class predetermined to indicate a keyword for retrieving data from a data set obtained by merging two data sets composed of an instance and data having two or more nodes having a class indicating the instance. A query obtained by acquiring data connecting a node having a node and a node having a predetermined class to indicate a target instance in the retrieved data and substituting the instance of the data with a variable is stored. Query storage means;
Path generation means for dividing a path consisting of only one route by a node included in the query at a node corresponding to a node common to the two data sets;
Path storage means for storing the divided paths;
Path classification means for classifying the paths of the path storage unit into groups for each class set indicating instances of nodes at both ends of the path;
Data search means for substituting a predetermined search word as an instance of the node into one end node of the path of the path storage unit, and searching the merged data set for data that matches the path other than variables; ,
For each path in the group in each group, a set of instances of the node at the other end of the data searched by the target path and an instance of the node at the other end of the data searched by the non-target path Get the set, calculate the value obtained by dividing the number of instances of the former set by the number of instances of the union of each set, and select the path corresponding to the value higher than the predetermined threshold from the group as the representative path Representative path selection means to
Path merging means for merging paths of representatives of different groups so that the merged path connects the node corresponding to the keyword and the node corresponding to the target instance;
Representative path storage means for storing the merged path;
A query including the merged path is retrieved from the query storage means, and the number of retrieved queries is greater than or equal to a predetermined number for each group set to which each path before merging belongs. First query search means for instructing the representative path selection means to adjust a threshold;
A second query search means for searching a query consisting only of the merged path from the query storage means when the number of searched queries is equal to or greater than a predetermined number. Generator. The query generation apparatus according to claim 1, wherein the second query search unit searches the query storage unit for a query including only two or more merged paths. A class predetermined to indicate a keyword for retrieving data from a data set obtained by merging two data sets composed of an instance and data having two or more nodes having a class indicating the instance. A query obtained by acquiring data connecting a node having a node and a node having a predetermined class to indicate a target instance in the retrieved data and substituting the instance of the data with a variable is stored. An operation method of a query generation device including query storage means,
A path generation step of dividing a path consisting of only one route by a node included in the query at a node corresponding to a node common to the two data sets, and storing the divided path in a path storage unit;
A path classification step of classifying the paths of the path storage unit into groups for each set of classes indicating instances of nodes at both ends of the path;
A data search step of substituting a predetermined search word as an instance of the node into a node at one end of the path of the path storage unit, and searching the merged data set for data that matches the path other than variables; ,
For each path in the group in each group, a set of instances of the node at the other end of the data searched by the target path and an instance of the node at the other end of the data searched by the non-target path Get the set, calculate the value obtained by dividing the number of instances of the former set by the number of instances of the union of each set, and select the path corresponding to the value higher than the predetermined threshold from the group as the representative path A representative path selection process,
The representative paths of different groups are merged so that the merged path connects the node corresponding to the keyword and the node corresponding to the target instance, and the merged path is stored as representative path storage means. Path merge process to be stored in
A query including the merged path is retrieved from the query storage means, and the number of retrieved queries is greater than or equal to a predetermined number for each group set to which each path before merging belongs. A first query search step for instructing the representative path selection means to adjust a threshold;
A second query search step of searching for a query consisting only of the merged path from the query storage means when the number of searched queries exceeds a predetermined number. How the generator works.   4. The operation method of the query generation device according to claim 3, wherein the second query search step searches the query storage unit for a query including only two or more merged paths.   A computer program for causing a computer to function as the query generation device according to claim 1.

Images