JP5208193B2 - Related word graph creation device, related word graph creation method, related word providing device, related word providing method, and program - Google Patents

Description

  The present invention relates to a related word graph creating device, a related word graph creating method, a related word providing device, a related word providing method, and a program.

  Query suggestion that provides related terms related to keywords used for search is performed. For example, Patent Literature 1 describes a server that presents recommended keywords in accordance with a word used by a user, the number of times the word is used, a category of a search keyword, and the like. A thesaurus can be considered as a dictionary for searching related words related to keywords. Non-Patent Document 1 describes a thesaurus search process.

JP 2010-3134 A

Yoji Kiyota, Takeshi Abekawa, Satoshi Yoshida, Goro Tamura, Satoshi Sakai, Hidetaka Masuda, 2008, "Development of a general-purpose thesaurus search library", [online], The 14th Annual Conference of the Language Processing Society, Internet <URL: http : //www.r.dl.itc.u-tokyo.ac.jp/%7Ekiyota/paper/2008/NLP_2008_PD1-3-kiyota.pdf>

  The thesaurus defines a hierarchy of concepts, but contains few proper nouns and technical terms. Therefore, it is conceivable to use a topic graph obtained by analyzing a corpus containing a large amount of text, but the topic graph contains many proper nouns and technical terms, but the concept hierarchy is not defined. Therefore, there is a need to combine a topic graph with a thesaurus. However, the words included in the thesaurus and the words included in the topic graph hardly matched, and it was difficult to combine the topic graph with the thesaurus.

  The present invention has been made in view of such a background, and a related word graph creation device capable of combining a thesaurus defining a concept hierarchy with a topic graph defining a relationship between words included in a corpus, An object is to provide a word graph creation method, a related word providing device, a related word providing method, and a program.

  A main invention of the present invention for solving the above problems is an apparatus for creating a related word graph that defines a relationship between words, a thesaurus storage unit that stores a thesaurus that defines a concept hierarchy of words, and a corpus A topic graph storage unit that stores a topic graph that defines a relationship between words included in the category graph, a category graph storage unit that stores a category graph that defines a hierarchy of categories used for directory search, and a first graph in the category graph The thesaurus, the category graph, and the topic graph are combined so that words included in the category and the thesaurus are linked and words included in the second category and the topic graph in the category graph are linked. , Graph to create the related word graph of graph structure And further comprising a forming unit.

  According to the related word graph creating apparatus of the present invention, a thesaurus and a category graph are linked, and a category graph and a topic graph are linked, whereby one related word graph in which all the thesaurus, the category graph, and the topic graph are combined. Can be created. Therefore, the related word graph includes both a thesaurus that is often composed of common words and a topic graph that contains many fixed nouns, specific words, technical terms, etc. Can consist of a wide range of words.

  In the related word graph creation device of the present invention, the graph creation unit includes a word included in the thesaurus that matches the name of the first category in the category graph, and a lower hierarchy of the first category. The topic graph may be combined with the thesaurus so that a word included in the topic graph that matches the name of the second category to which the user belongs belongs is linked.

  In this case, the thesaurus and the topic graph can be combined so as to be directly linked by using the category graph. Therefore, general terms included in the thesaurus can be linked to fixed nouns, specific words, technical terms, etc. included in the topic graph, so links should be provided between various words in the related term graph. Can do.

  In the related word graph creation device of the present invention, the thesaurus, the topic graph, and the category graph are directed graphs, and the graph creation unit connects the edges for each edge between two categories in the category graph. A first word that matches the name of the first category of the two categories is included in the thesaurus, and a second word that matches the name of the second category of the two categories When the topic graph is included in the topic graph, the topic graph may be combined with the thesaurus so that an edge connecting the first and second words is created.

  In the related word graph creating device of the present invention, the thesaurus storage unit stores node information including a word corresponding to the node for each node constituting the thesaurus, and an edge between the nodes constituting the thesaurus For each, edge information including two pieces of identification information indicating the two nodes is stored, and the topic graph storage unit stores the node information including a word corresponding to the node for each of the nodes constituting the topic graph. Storing the edge information including two pieces of the identification information for each edge between the nodes constituting the topic graph, and the category graph storage unit configured to store the node information for each node constituting the category graph. Storing the node information including the word indicating the category corresponding to, and constructing the category graph The edge information including two pieces of the identification information is stored for each edge between the nodes, and the graph creation unit stores the node information and the edge information stored in the thesaurus storage unit in the related word graph storage Each of the edge information stored in the category graph storage unit, and the node information and the edge information stored in the topic graph storage unit are registered in the related word graph storage unit. The word set in the first node information corresponding to the first identification information included in the edge information is one of the words set in the second node information stored in the thesaurus storage unit. And the word set in the third node information corresponding to the second identification information included in the edge information is the topic The identification information indicating the second node information and the identification information indicating the fourth node information when the word matches the word set in the fourth node information stored in the rough storage unit. The set edge information may be registered in the related word graph.

  According to another aspect of the present invention, there is provided a related word providing apparatus for providing related words related to an input word, the first category in the category graph defining a hierarchy of categories used for directory search, Words included in a thesaurus that defines the concept hierarchy of words are linked, and words included in a topic graph that defines a relationship between the second category in the category graph and words included in the corpus are linked. A related word graph storage unit that stores a related word graph having a graph structure, created by combining the thesaurus, the category graph, and the topic graph, a word input unit that receives an input of a first word, and the related item In the related word graph stored in the word graph storage unit, one or more edges are extracted from the first word. The second word is connected to the associated word acquisition unit that acquires, as the related word, and further comprising a related word output unit which outputs the related word.

  According to the related word providing apparatus of the present invention, the thesaurus and the category graph are linked, and the category graph and the topic graph are linked to create one related word graph combining all the thesaurus, the category graph, and the topic graph. From the related word graph, the related word (second word) linked to the keyword (first word) can be presented. Therefore, the related term graph includes both a thesaurus that is often composed of common words and a topic graph that contains many fixed nouns, specific words, technical terms, etc. The related words can be searched from, and the possibility that the keywords are included in the related word graph can be increased, so that the related words can be surely presented.

  Another aspect of the present invention is a related word providing device that provides related words related to an input word, and is included in a corpus and a thesaurus storage unit that stores a thesaurus that defines a concept hierarchy of words. A topic graph storage unit that stores a topic graph that defines a relationship between words, a category graph storage unit that stores a category graph that defines a hierarchy of categories used for directory search, the first category in the category graph, and A graph structure combining the thesaurus, the category graph, and the topic graph such that words included in the thesaurus are linked and words included in the second category and the topic graph in the category graph are linked A graph creation unit for creating the related word graph of In the related word graph storage unit that stores the word graph, the word input unit that receives the input of the first word, and the related word graph stored in the related word graph storage unit, one from the first word Suppose that the related word acquisition part which acquires the 2nd word connected via the above edge as the said related word, and the related word output part which outputs the said related word are provided.

  According to the related word providing apparatus of the present invention, the thesaurus and the category graph are linked, and the category graph and the topic graph are linked to create one related word graph combining all the thesaurus, the category graph, and the topic graph. From the related word graph, the related word (second word) linked to the keyword (first word) can be presented. Therefore, the related word graph includes both a thesaurus that is often composed of common words and a topic graph that contains many fixed nouns, specific words, technical terms, etc. The related word (second word) linked to the keyword (first word) can be easily presented. The related word graph includes both a thesaurus that is often composed of common words and a topic graph that contains many fixed nouns, specific words, technical terms, etc. It is possible to search for words, increase the possibility that keywords are included in the related word graph, and reliably present related words.

  Another aspect of the present invention is a method for creating a related word graph that defines relationships between words, wherein a computer stores a thesaurus that defines a concept hierarchy of words in a memory and is included in the corpus. A topic graph defining a relationship between words is stored in the memory, a category graph defining a hierarchy of categories used for directory search is stored in the memory, and included in the first category and the thesaurus in the category graph The related words in the graph structure by combining the thesaurus, the category graph, and the topic graph so that the words included in the category graph and the second category in the category graph and the words included in the topic graph are linked. A graph will be created.

  Another aspect of the present invention is a method for providing a related word related to an input word, the concept of the first category and word in a category graph defining a hierarchy of categories used for directory search. Words included in a thesaurus that defines a hierarchy are linked, and words included in a topic graph that defines a relationship between the second category in the category graph and words included in the corpus are linked. A computer that stores a related word graph having a graph structure, created by combining the thesaurus, the category graph, and the topic graph, receives an input of a first word and stores the related word stored in the related word graph storage unit. In a word graph, a second word connected from the first word via one or more edges, Obtained as serial related word, and outputting said related words.

  Another aspect of the present invention is a method for providing related words related to an input word, in which a computer stores a thesaurus defining a concept hierarchy of words in a memory, and between words included in a corpus. A topic graph that defines the relationship of each of the categories, a category graph that defines a hierarchy of categories used for a directory search is stored in the memory, and the words included in the first category and the thesaurus in the category graph The related word graph of the graph structure is combined by combining the thesaurus, the category graph, and the topic graph so that the words included in the second category and the topic graph in the category graph are linked. The related word graph is created and stored in the memory, and the first word is entered. In the related word graph stored in the memory, the second word connected from the first word via one or more edges is acquired as the related word, and the acquired relation A word is output.

  Another aspect of the present invention is a program for creating a related word graph that defines a relationship between words, and a program for creating a related word graph that defines a relationship between words. Storing a thesaurus defining a concept hierarchy of words in a computer, storing a topic graph defining a relationship between words included in the corpus in the memory, and a category used for directory search. A category graph defining a hierarchy is stored in the memory, and the first category in the category graph and words included in the thesaurus are linked and included in the second category and the topic graph in the category graph. The thesaurus, the category graph, and the Combining the topic graph, and the step of creating the related word graph graph structure, to execution.

  According to another aspect of the present invention, there is provided a program for providing a related word related to an input word, the first graph in a category graph defining a hierarchy of categories used for directory search on a computer. A category and a word included in a thesaurus that defines a concept hierarchy of words are linked, and a second category in the category graph and a word included in a topic graph that defines a relationship between words included in the corpus are linked. Storing a related word graph of a graph structure created by combining the thesaurus, the category graph, and the topic graph in a memory, receiving an input of a first word, and storing in the memory One or more edges from the first word in the related word graph being A second word which is connected via a step of acquiring, as the related word, and be executed, and a step of outputting the related word.

  According to another aspect of the present invention, there is provided a program for providing a related word related to an input word, the step of storing a thesaurus defining a concept hierarchy of the word in a computer, and a corpus. Storing in the memory a topic graph defining a relationship between words contained therein, storing in the memory a category graph defining a hierarchy of categories used for directory searches, and the first in the category graph Combining the thesaurus, the category graph, and the topic graph such that words included in the category and the thesaurus are linked, and words included in the second category and the topic graph in the category graph are linked, The step of creating the related word graph of the graph structure One or more of the first word in the related word graph stored in the memory, a step of storing the related word graph in the memory, a step of receiving an input of a first word, The second word connected via the edge is acquired as the related word, and the step of outputting the acquired related word is executed.

  Other problems and solutions to be disclosed by the present application will be made clear by the embodiments of the invention and the drawings.

  According to the present invention, it is possible to combine a topic graph that defines a relationship between words included in a corpus with a thesaurus that defines a concept hierarchy.

4 is a diagram for explaining a related word graph 10. FIG. It is a figure which shows the structural example of a computer system. 2 is a diagram illustrating a hardware configuration of a search device 20. FIG. 3 is a diagram illustrating a software configuration of the search device 20. FIG. 2 is a diagram illustrating a configuration example of a thesaurus 11. FIG. 3 is a diagram illustrating a configuration example of a topic graph 12. FIG. 6 is a diagram illustrating a configuration example of a category graph 13. FIG. 4 is a diagram illustrating a configuration example of a related word graph 10. FIG. It is a figure which shows the flow of a merge process. It is a figure which shows the flow of the process which registers the node information and edge information based on the thesaurus. It is a figure which shows the flow of the process which registers the node information and edge information based on the topic graph. It is a figure which shows the flow of the process which registers the node information and edge information based on the category graph. FIG. 10 is a diagram showing a flow of edge addition processing based on a category graph 13. It is a figure which shows the state with which the thesaurus 11 and the topic graph 12 were linked. It is a figure which shows an example of the screen 40 which inputs a keyword. It is a figure which shows the flow of an acquisition process of a related word when a keyword is comprised only from one token. It is a figure for demonstrating a synonym. It is a figure which shows the flow of an acquisition process of a synonym. It is a figure for demonstrating a low-order word. It is a figure which shows the flow of the process which acquires a low-order word. It is a figure for demonstrating a lower association word. It is a figure which shows the flow of the process which acquires a lower association word. It is a figure for demonstrating a high-order association word. It is a figure which shows the flow of the process which acquires a high-order association word. It is a figure for demonstrating an isotopic association word. It is a figure which shows the flow of the process which acquires a peer association word. It is a figure which shows the flow of the acquisition process of a related word when a keyword is comprised from the some token. It is a figure for demonstrating a common synonym. It is a figure which shows the flow of the process which acquires a common synonym. It is a figure for demonstrating a common low-order word. It is a figure which shows the flow of a process which acquires a common low-order word. It is a figure for demonstrating a mediation word. It is a figure which shows the flow of the process which acquires a mediation word. It is a figure for demonstrating the optimal mediation word. It is a figure which shows the flow of the process which acquires the optimal mediation word. It is a figure which shows an example of the screen 41 which outputs a search result and a related word. It is a figure which shows a part of related word graph 10 used for a search process.

  Hereinafter, the search device 20 according to the present embodiment will be described. In response to a search request (referred to as “query”), the search device 20 according to the present embodiment searches for information corresponding to a search keyword set in the query, and a word related to the search keyword (hereinafter, “ So-called query suggestion is provided. In addition, the search device 20 of the present embodiment creates a dictionary (hereinafter referred to as “related word graph 10”) that manages the relationship between words, which is used for query suggestion.

== Related Word Graph 10 ==
The related word graph 10 is created by defining a relationship between words created from a corpus such as a Web page or a newspaper article in a thesaurus 11 that defines a concept hierarchy of words (a relationship between upper and lower concepts between words). This is done by connecting (merging) graphs (hereinafter referred to as “topic graph 12”). Usually, the thesaurus 11 is mainly composed of general words, and the topic graph 12 created from the corpus includes more specific words and proper nouns (referred to as “topics”) than general words. Therefore, since there are few overlapping words in the thesaurus 11 and the topic graph 12, it is difficult to simply merge the topic graph 12 into the thesaurus 11. Therefore, in the search device 20 of the present embodiment, the topic graph 12 is merged with the thesaurus 11 by using a directed graph (hereinafter referred to as “category graph 13”) that defines a category hierarchy used for Web page directory search. Then, the related word graph 10 is created. FIG. 1 is a diagram for explaining a related word graph 10. Also,
The thesaurus 11 manages the vertical relationship of concepts using a directed graph and manages words for expressing each concept. In the thesaurus 11, the concept (also called “class”) is managed as a numerical value. The concept can be a value such as “# 3”. A word expressing a concept (referred to as an “instance”) is managed as a character string. The instance may be a character string such as “music”. In FIG. 1, concepts managed by the thesaurus 11 are represented by black triangles, and words are represented by white triangles. Of the edges (also referred to as “links”) between nodes, the vertical relationship is represented by solid arrows, and the relationship between concepts and instances is represented by dashed-dotted arrows. In the graph structure constituting the thesaurus 11, an instance is a node (referred to as a “leaf node” or simply “leaf”) having no outgoing edge at the end. The concept is a node other than a leaf node. Hereinafter, nodes other than leaf nodes are also referred to as “concept nodes”. In the example of FIG. 1, the node 111 of “# 1” and the node 112 of “# 3” are both concepts, and the superordinate concept of the node 112 is the node 111. The “sound” leaf node 114 is an instance of the node 111.

  In the category graph 13, a hierarchy of category names is defined by the directed graph, and the category name becomes each node. In FIG. 1, a category name (hereinafter referred to as “category name”) in the category graph 13 is represented by a square. In the example of FIG. 1, since the edge is directed from the node 131 corresponding to the category name “Rock” to the node 132 corresponding to the category name “Rock Artist”, “Rock” is higher than “Rock Artist”. It turns out that it becomes a category. The nodes that make up the category graph 13 are also called “redirects” to distinguish them from the nodes that make up the thesaurus 11.

  The topic graph 12 defines the relationship between topics by a directed graph. In the topic graph 12, words become nodes. In FIG. 1, topics in the topic graph 12 are represented by diamonds. The relationship between words in the topic graph 12 may be a relationship between words obtained by analyzing a corpus such as a relationship between a subject and a predicate, a distance between words, and a frequency of coexistence in the same document. Such relevance that does not have a vertical relationship is represented by a dashed arrow in FIG. In the example of FIG. 1, a dashed arrow link 123 is drawn from the topic 121 to the topic 122, and it can be seen that there is a relationship from “Rock Artist” to “○ Mountain × Huo”. The nodes constituting the topic graph 12 are also referred to as “redirects” in order to distinguish them from the thesaurus 11 nodes.

  As will be described later, when the word of the node included in the thesaurus 11 and the category name of the node included in the category graph 13 match, an edge is formed from the node of the thesaurus 11 to the node below the node of the category graph. Created. For example, in FIG. 1, the “lock” leaf node 116 of the thesaurus 11 and the “lock” redirect 131 of the category graph 13 in FIG. 1 match, and in this case, the leaf node 116 is connected to the subordinate of the redirect 131. An edge 141 is created in the redirect 132 that has been redirected, and similarly, an edge is created from the leaf node 116 to another redirect connected to the lower level of the redirect 131.

  Further, when the category name of the node included in the category graph matches the word of the node included in the topic graph, the category graph 13 is connected to the node connected by the edge from the node of the topic graph 12. A new edge is created from the node. For example, in FIG. 1, the redirect 132 of “Rock Artist” in the category graph 13 matches the redirect 121 of “Rock Artist” in the topic graph 12, and the redirect 121 is connected to the redirect 122 connected at the edge from the redirect 121. , A new edge 142 is created from the redirect 132.

  In this way, the “rock” node 116 included in the thesaurus 11 to the “rock artist” node 121 and the “mountain × husband” node 122 in the topic graph 12 are connected by edges. The thesaurus 11 and the topic graph 12 are merged via the category graph 13.

  Further, in the present embodiment, as will be described later, the topic graph 12 is directly merged into the thesaurus 11 by creating an edge 143 between the “lock” node 116 included in the thesaurus 11 and the redirect 121. I am also doing so.

  The created related word graph 10 is used for query suggestion that provides a word related to a keyword as a related word. The keyword may be a phrase composed of a plurality of words. Each word that makes up a keyword is called a “token”. As will be described later, the search device 20 searches the related word graph 10 for a node that matches the token, and outputs nodes such as the parent, child, sibling, ancestor, and descendant of the searched node as related words.

== System configuration ==
Hereinafter, a computer system including the search device 20 that creates the related term graph 10 as described above will be described. FIG. 2 is a diagram illustrating a configuration example of a computer system including the search device 20 according to the present embodiment. In the present embodiment, the search device 20 is connected to the communication network 21, receives a query including a keyword from the terminal 22, and searches for a document corresponding to the keyword included in the received query. The terminal 22 is, for example, a computer such as a personal computer or a mobile phone terminal, and the communication network 21 is, for example, the Internet or a local area network. Further, the search device 20 realizes a so-called keyword suggestion function that presents related words related to a keyword when responding to a search result.

  FIG. 3 is a diagram illustrating a hardware configuration of the search device 20. The search device 20 includes a CPU 201, a memory 202, a storage device 203, a communication interface 204, an input device 205, and an output device 206. The storage device 203 is, for example, a hard disk drive or a flash memory that stores various data and programs. The CPU 201 implements various functions by reading a program stored in the storage device 203 into the memory 202 and executing it. The communication interface 204 is an adapter for connecting to the communication network 21, for example, an adapter for connecting to Ethernet (registered trademark), a modem for connecting to a public telephone line network, and a wireless for connecting to a mobile phone line network. It is an interface for communication equipment. The input device 205 is, for example, a keyboard, a mouse, a microphone, or the like that accepts data input. The output device 206 is, for example, a display, a printer, or a speaker that outputs data.

  FIG. 4 is a diagram illustrating a software configuration of the search device 20. The search device 20 includes a data acquisition unit 211, a graph merge unit 212, a search query reception unit 213, a search processing unit 214, a related word acquisition unit 215, a related word output unit 216, a thesaurus database 231, a topic graph database 232, and a category graph database. 233 and a related word graph database 234. The data acquisition unit 211, the graph merge unit 212, the search query reception unit 213, the search processing unit 214, the related word acquisition unit 215, and the related word output unit 216 store programs stored in the storage device 203 in the memory by the CPU 201. The thesaurus database 231, the topic graph database 232, the category graph database 233, and the related term graph database 234 are realized as a part of a storage area provided by the memory 202 or the storage device 203.

  The thesaurus database 231 stores the thesaurus 11. FIG. 5 is a diagram illustrating a configuration example of the thesaurus 11 stored in the thesaurus database 231. In the thesaurus database 231, the thesaurus 11 includes information relating to nodes constituting the graph (hereinafter referred to as “node information”) and information relating to edges connecting the nodes (hereinafter referred to as “edge information”). The node information included in the thesaurus 11 includes information for identifying the node (hereinafter referred to as “node ID”), a word (name) corresponding to the node, and the type of the node. The node type is “node” or “leaf”, the node information corresponding to the concept node is set to “node”, and the node information corresponding to the leaf node is set to “leaf”. . In the node information corresponding to the concept node, a unique value assigned to the concept is set as the name. In the present embodiment, it is assumed that a character string obtained by adding a node ID to “#” is set as the name. Among the two nodes connected to the edge, the edge information included in the thesaurus 11 includes a node ID (ID) indicating a higher node and a node ID (child ID) indicating a lower node, a weight assigned to the edge, As well as the relationship between the two nodes to which the edge connects. The direction of the edge is from the node indicated by the edge information ID to the node indicated by the child ID of the edge information. The relationship of edge information is “instance” or “lower”. For example, in the example of FIG. 1, “instance” is set for the edge 115 having an instance relationship from the concept node 111 of “# 1” to the leaf node 114 of “sound”. For example, in the example of FIG. 1, “lower” is set for the edge 113 indicating the hierarchy between two concept nodes, such as the concept node 111 of “# 1” to the concept node 112 of “# 3”.

  The topic graph database 232 stores the topic graph 12. FIG. 6 is a diagram illustrating a configuration example of the topic graph 12 stored in the topic graph database 232. In the topic graph database 232, similarly to the thesaurus 11, the topic graph 12 includes node information for each node constituting the graph and edge information for each edge connecting the two nodes. The node information included in the topic graph 12 includes only a node ID indicating a node and a word (name) corresponding to the node. For example, in the example of FIG. 1, for the redirect 121 of “Rock Artist”, the node information indicating the redirect 121 is set in the ID column, and the node information in which “Rock Artist” is set in the name column is registered in the topic graph database 232. The The edge information included in the topic graph 12 includes an ID indicating the first node, a child ID indicating the second node, and an edge assigned to the edge from the first node toward the second node. Weights included. For example, in the example of FIG. 1, the node ID indicating the redirect 121 is set in the ID column for the edge 123 from the redirect 121 of “Rock Artist” to the redirect 122 of “Oyama × Huo”, and the node ID indicating the redirect 122 is Edge information is created in which the weight of the edge 123 set in the child ID column and assigned when the corpus is analyzed is set in the weight column, and this edge information is registered in the topic graph database 232.

  The category graph database 233 stores the category graph 13. FIG. 7 is a diagram illustrating a configuration example of the category graph 13 stored in the category graph database 233. Like the thesaurus 11 and the topic graph 12, the category graph 13 includes node information for each node constituting the graph and edge information for each edge connecting the two nodes. The node information included in the category graph 13 includes a node ID indicating a node and a category name (name). For example, in the example of FIG. 1, the node ID indicating the node 131 is set in the ID column and the node information in which “lock” is set in the name column is registered in the category database 233. The edge information included in the category graph 13 includes an ID indicating a higher category and a child ID indicating a lower category among the two categories connected by the edge. For example, in the example of FIG. 1, for the edge 133 from the “rock” node 131 to the “rock artist” node 132, the node ID indicating the node 131 is set in the ID column, and the node ID indicating the node 132 is the child ID column. Is registered in the category graph database 233.

  The data acquisition unit 211 acquires node information and edge information from the thesaurus database 231, topic graph database 232, and category graph database 233. In the present embodiment, the thesaurus database 231, the topic graph database 232, and the category graph database 233 are all included in the search device 20, but may be managed by a separate database server. Further, for example, the data acquisition unit 211 accesses the editor computer of the thesaurus 11, the computer of the corpus analyst, the search server that performs directory search, etc., acquires the data of the graph structure, and acquires the data of the acquired graph structure Node information and edge information may be created based on the data.

  The related word graph database 234 stores the related word graph 10. FIG. 8 is a diagram illustrating a configuration example of the related word graph 10 stored in the related word graph database 234. Similar to the thesaurus 11, topic graph 12, and category graph 13 described above, the related term graph 10 includes node information about each node constituting the graph and edge information about an edge connecting two nodes. The node information included in the related word graph 10 includes a node ID indicating a node, a name, and a node type. In the node information type column, “node” or “leaf” is set for the nodes included in the thesaurus 11, and “redirect” is set for the nodes included in the topic graph 12 and the category graph 13. The edge information included in the related word graph 10 includes an ID indicating a connection source node of a directed edge, a child ID indicating a connection destination node, a weight assigned to the edge, and two nodes to which the edge is connected. The relationship is included. Either “instance”, “subordinate” or “related” is set in the relationship of the edge information.

  The graph merge unit 212 merges the topic graph 12 with the thesaurus 11 using the category graph 13. In the present embodiment, the graph merging unit 212 creates an edge that connects the leaf node of the thesaurus 11 and the node of the category graph 13, creates an edge that connects the node of the category graph 13 and the node of the topic graph 12, By creating an edge that connects the leaf node of the thesaurus 11 and the node of the topic graph 12, the topic graph 12 is merged with the thesaurus 11 to create the related word graph 10. Details of the merge processing by the graph merge unit 212 will be described later. The graph merge unit 212 registers the created related word graph 10 in the related word graph database 234.

  The search query receiving unit 213 receives a query transmitted from the terminal 22. As described above, the query includes a keyword.

  The search processing unit 214 searches for a document corresponding to the keyword included in the query. The document search process by the search processing unit 214 is assumed to be a search process by a general search engine.

  The related word acquisition unit 215 acquires a related word corresponding to the keyword included in the query from the related word graph 10 stored in the related word graph database 234. Details of related word acquisition processing by the related word acquisition unit 215 will be described later.

  The related word output unit 216 outputs the related word acquired by the related word acquisition unit 215. In the present embodiment, it is assumed that the related word output unit 216 transmits the related word to the terminal 1 together with the search processing result by the search processing unit 214.

== Merge processing ==
Hereinafter, the process of merging the topic graph 12 with the thesaurus 11 using the category graph 13 will be described.
FIG. 9 is a diagram showing the flow of merge processing. The graph merge unit 212 registers node information and edge information in the related word graph database 234 based on the thesaurus 11 (S301), and registers node information and edge information in the related word graph database 234 based on the topic graph 12 ( In S302, by registering node information and edge information in the related word graph database 234 based on the category graph 13 (S303), the related word graph 10 is registered in the related word graph database 234.

  FIG. 10 is a diagram showing a flow of processing for registering node information and edge information based on the thesaurus 11. The graph merge unit 212 reads node information from the thesaurus database 231 (S321), reads edge information from the thesaurus database 231 (S322), and registers the read node information and edge information in the related word graph database 234 as they are (S323). . That is, the node information and edge information registered in the thesaurus database 231 are registered in the related word graph database 234.

FIG. 11 is a diagram showing a flow of processing for registering node information and edge information based on the topic graph 12.
The graph merge unit 212 sets x as the maximum value of the ID of the node information registered in the related word graph database 234 (S341). The graph merge unit 212 reads out edge information from the topic graph database 232 to obtain an edge list (S342), and adds x to the edge information ID and child ID for each edge information included in the edge list (S343). Then, the “related” relationship is added to the edge information (S344). The graph merge unit 212 reads the node information from the topic graph database 232 to obtain a node list (S345), adds x to the node information ID for each node information included in the node list (S346), and adds the node information to the node information. The type of “redirect” is added (S347).

  The graph merge unit 212 performs the following processing on each node information included in the node list. The graph merge unit 212 sets the node information to N (S348), and searches the related word graph database 234 for node information that matches the name of N as matching information (S349). If there is matching information (S350: YES), the graph merging unit 212 searches the edge list for edge information E that matches the ID of ID N (S351), and sets each of the searched edge information E as matching information. Edge information including an ID included, a child ID included in E, a weight included in E, and a “lower” relationship is created and written into the related word graph database 234 (S352).

  After performing the above processing for each node information, the graph merge unit 212 writes the node information included in the node list to the related word graph database 234 (S353), and the edge information included in the edge list is related word graph database 234. (S354).

  As described above, the node information and edge information stored in the topic graph database 232 are registered in the related word graph database 234. If the topic also exists as a leaf node of the thesaurus 11, edge information indicating a “lower” relationship from the leaf node of the thesaurus 11 toward another topic related to the topic is related word graph database. 234 is registered.

FIG. 12 is a diagram showing a flow of processing for registering node information and edge information based on the category graph 13.
The graph merge unit 212 sets x as the maximum value of the ID of the node information registered in the related word graph database 234 (S361), reads the edge information from the category graph database 233, and sets it as an edge list (S362). For each piece of included edge information, x is added to each of the edge information ID and child ID (S363), and a "lower" relationship is added to the edge information (S364). The graph merge unit 212 reads the node information from the category graph database 233 to obtain a node list (S365), adds x to the node information ID for each node information included in the node list (S366), and adds the node information to the node information. The type of “redirect” is added (S367). The graph merge unit 212 writes the node information included in the node list into the related word graph database 234 (S368), and writes the edge information included in the edge list into the related word graph database 234 (S369).

  The graph merge unit 212 sets the node information to N for each node information included in the node list (S370), and performs an edge addition process shown in FIG. 13 (S371).

  The graph merge unit 212 receives node information whose name matches the name of N and whose type is “leaf” (that is, node information corresponding to the leaf node of the thesaurus 11 corresponding to the name of N). The related word graph database 234 is searched as matching information (S381). If there is matching information (S382: YES), the graph merging unit 212 searches the related word graph database 234 for edge information E that matches the ID of ID N (S383), and for each of the searched edge information E, Edge information including the ID included in the matching information, the child ID included in E, the weight included in E, and the “lower” relationship is created and written in the related word graph database 234 (S384). Thereby, when there is a word of the thesaurus 11 that matches the category name in the category graph 13, an edge is created from the word to a lower category of the category. For example, as shown in FIG. 14, when the word “lock” corresponding to the leaf node 116 of the thesaurus 11 and the category name “lock” corresponding to the redirect 131 of the category graph 13 match, an edge 141 is added. Will be.

  The graph merge unit 212 receives node information whose name matches the name of N and whose type is “redirect” (that is, node information corresponding to redirection of the topic graph 12 corresponding to the name of N). If the matching information is searched from the related word graph database 234 (S385), and if there is matching information (S386: YES), the edge information E that matches the child ID N is searched from the related word graph database 234 (S387). For each of the searched edge information E, an edge information including an ID included in E, an ID included in matching information, a weight included in E, and a “lower” relationship is created, and the related word graph database 334 is created. Write (S388). Thereby, when there is a topic in the topic graph 12 that matches the category name in the category graph 13, an edge from the category higher in the category name to the topic is created. For example, as shown in FIG. 14, when the category name “rock artist” corresponding to the redirect 132 in the category graph 13 matches the word “rock artist” corresponding to the redirect 121 in the topic graph 12, the edge 144 Will be added. Further, the edge 143 from the leaf node 116 added to the upper level of the redirect 132 in the above step S384 to the redirect 121 is added.

  In addition, the graph merge unit 212 searches the related word graph database 234 for edge information E whose ID matches the ID of the matching information (S389), and sets the ID and E included in N for each of the searched edge information E. Edge information including the included ID, the weight included in E, and the “lower” relationship is created and written into the related word graph database 334 (S390). As a result, when there is a topic in the topic graph 12 that matches the category name in the category graph 13, an edge is created from the category name to the previous topic connected by the edge from the topic. For example, as shown in FIG. 14, when the category name “Rock Artist” corresponding to the redirect 132 in the category graph 13 matches the word “Rock Artist” corresponding to the redirect 121 in the topic graph 12, an edge 142 is added. Will be.

  As described above, the topic graph 12 is connected to the thesaurus 11 by the edge 137 from the leaf node 116 of the thesaurus 11 to the redirect 122 of the topic graph 12.

  The graph merging unit 212 repeats the above processing for each node information, thereby registering the node information and edge information stored in the category graph database 233 in the related word graph database 234 and also the thesaurus 11 that matches the category name. The topic of the topic graph 12 that matches the word or category name can be linked.

  As described above, if there is a category name included in the category graph 13 that matches the word of the leaf node of the thesaurus 11, an edge from the word to the category name is created, and the topic graph If there is a match for 12 topics, an edge from the category name to that topic is created, and an edge from the leaf node of the thesaurus 11 to the topic is also created.

== Query suggestion process ==
Next, query suggestion processing for presenting related terms related to keywords included in a query will be described. FIG. 15 is a diagram illustrating an example of a screen 40 that is displayed on the terminal 22 and allows a user to enter a keyword. When the terminal 22 accesses the search device 20, the search processing unit 214 transmits screen data for displaying the screen 40 to the terminal 22, and the terminal 22 displays the screen 40 based on the screen data. When the user inputs a keyword in the keyword input field 401 and presses the search button 402, a query including the keyword input in the input field 401 is transmitted from the terminal 22 to the search device 20. The search processing unit 214 of the search device 20 searches for a document corresponding to the keyword included in the query. Note that the document search processing by the search processing unit 214 is the same as that performed by a general search engine, and a description thereof will be omitted.

  Further, the related word acquisition unit 215 acquires a related word related to the keyword included in the query from the related word graph 10 in response to the reception of the query. In the present embodiment, different processing is performed depending on whether the number of tokens constituting the keyword is one or two or more. FIG. 16 is a diagram showing a flow of related word acquisition processing when a keyword is composed of only one token. In the example of FIG. 15, “Rock Artist” is set as the keyword, and the token constituting the keyword is only one “Rock Artist”, so the processing of FIG. 16 is performed.

  The related word acquisition unit 215 creates an empty related word list (S401), and searches the related word graph database 234 for node information whose name matches the token included in the query (S402). When the node information is searched (S403: YES), the related word acquisition unit 215 performs a synonym acquisition process for each node information (S404), performs a low-level word acquisition process (S405), and generates a low-level association word. Is acquired (S406), a higher association word acquisition process is performed (S407), and a peer association word acquisition process is performed (S408).

== Synonyms ==
A token synonym is another leaf node that is in an instance relationship from a conceptual node above the token when the token is a leaf node of the thesaurus 11. As shown in FIG. 17, when two leaf nodes 604 other than the token 601 are connected in an instance relationship from the upper node 603 connected by the token 601 and the edge 602, the synonym of the token 601 is Leaf node 604.

FIG. 18 is a diagram illustrating a flow of synonym acquisition processing. When the type of node information is “leaf” (S421: YES), the related word acquisition unit 215 searches the related word graph database 234 for edge information whose child ID matches the node information ID as parent edge information. (S422). The related word acquisition unit 215 performs the following processing for each of the searched parent edge information. That is, the related word acquisition unit 215 acquires, as parent node information, node information whose ID matches the ID of the parent edge information from the related word graph database 234 (S423), and the type of the acquired parent node information is “node”. If so (S424: YES), edge information whose ID matches the ID of the parent node information is further searched from the related word graph database 234 (S425). Here, the case where the child ID of the edge information matches the ID of the node information is excluded. The related word acquisition unit 215 acquires, as child node information, node information whose ID matches the child ID of the edge information from the related word graph database 234 for each searched edge information (S426), and acquires the acquired child node information. If the type is “leaf” (S427: YES), the child node information name is added as a synonym to the related word list (S428).
As described above, another leaf node that is subordinate to the upper node of the leaf node of the thesaurus 11 and is the same as that of the leaf node is acquired as a synonym.

== Narrow word ==
The token narrower word is a redirect connected to a leaf node in the related word graph 10 when the token is a leaf node of the thesaurus 11. As shown in FIG. 19, when the token is a leaf node 611 of the thesaurus 11, even if another leaf node 613 is connected to the leaf node 611 by the edge 612, the leaf node 613 is not a low-order word. The redirect 614 connected to the leaf node 611 is a narrower term.
FIG. 20 is a diagram showing a flow of processing for acquiring a narrower word. When the type of the node information is “leaf” (S441: YES), the related word acquisition unit 215 searches the related word graph database 234 for edge information whose ID matches the ID of the node information as child edge information ( S442). The related word acquisition unit 215 acquires, as child node information, node information whose ID matches the child ID of the child edge information from the related word graph database 234 for each searched child edge information (S443). If the type of node information is “redirect” (S444: YES), the name of the child node information is added to the related word list as a lower term (S445).
As described above, the redirect node connected to the lower level of the leaf node of the thesaurus 11, that is, the topic name of the topic graph 12 or the category name of the category graph 13 connected to the lower level of the leaf node of the thesaurus 11 is the lower term. Get as.
Note that a leaf node connected to a lower level of a leaf node may be acquired as a lower term. Further, a leaf node connected to a lower level of the leaf node may be acquired as a lower order association word described later.

== Lower association words ==
The token lower-level association word is a lower-level redirect connected to the redirect at the edge in the related word graph 10 when the token is a redirect. As shown in FIG. 21, when the token 621 is a redirect, even if the leaf node 623 is connected to the lower level by the edge 622, the leaf node 623 does not become the lower-level association word of the token 621 and The connected redirect 624 becomes a lower-order associative word.
FIG. 22 is a diagram illustrating a flow of processing for acquiring a lower association word. When the type of the node information is “redirect” (S461: YES), the related word acquisition unit 215 searches the related word graph database 234 for edge information whose ID matches the ID of the node information as child edge information ( S462). The related word acquisition unit 215 acquires, as child node information, node information whose ID matches the child ID of the child edge information from the related word graph database 234 for each searched child edge information (S463). If the type is “redirect” (S464: YES), the name of the child node information is added to the related word list as a lower association word (S465).
As described above, the redirect connected to the lower level of the redirect of the topic graph 12 or the category graph 13 is acquired as the lower association word.

== Higher association words ==
The high-level associated word of a token is a high-level redirect connected to the redirect at the edge in the related word graph 10 when the token is a redirect. As shown in FIG. 23, when the token 631 is a redirect, even if the leaf node 623 is connected to the higher level by the edge 632, the leaf node 633 is not a higher level association word of the token 631, and is higher in the token 631. The connected redirect 634 becomes a higher-level association word.
FIG. 24 is a diagram illustrating a flow of processing for acquiring a higher association word. When the type of node information is “redirect” (S481: YES), the related word acquisition unit 215 searches the related word graph database 234 for edge information whose child ID matches the node information ID as parent edge information. (S482). The related word acquisition unit 215 acquires, as parent node information, node information whose ID matches the ID of the parent edge information from the related word graph database 234 for each searched parent edge information (S483). If the type is “redirect” (S484: YES), the name of the parent node information is added to the related word list as a higher association word (S485).
As described above, the redirect connected to the top of the redirect of the topic graph 12 or the category graph 13 is acquired as a high-level association word.

== Similarity word ==
When the token is a redirect, the token associative word is connected to a lower level of the upper node (concept node, leaf node, or redirect) connected to the redirect in the related word graph 10. It is a redirect. As shown in FIG. 25, the redirect 644 connected to the lower level of the node 643 connected to the higher level of the token 641 is the peer association word.
FIG. 26 is a diagram showing a flow of processing for acquiring a peer association word. When the type of the node information is “redirect” (S501: YES), the related word acquisition unit 215 searches the related word graph database 234 for edge information whose child ID matches the node information ID as the parent edge information. (S502) The following processing is performed for each searched parent edge information. That is, the related word acquisition unit 215 searches the related word graph database 234 for edge information whose ID matches the ID of the parent edge information as child edge information (S503). Here, the child edge information whose child ID matches the node information ID is excluded. The related word acquisition unit 215 acquires, as child node information, node information whose ID matches the child ID of the child edge information from the related word graph database 234 for each searched child edge information (S504). If the type is “redirect” (S505: YES), the name of the child node information is added to the related word list as an associative word (S506).
As described above, another redirect that is a lower level of the upper node of the redirect is acquired as a peer association word.

== Related word acquisition processing of multiple tokens ==
When the keyword is composed of a plurality of tokens, in the present embodiment, processing different from the processing in FIGS. 16 to 26 described above is performed. FIG. 27 is a diagram showing a flow of related word acquisition processing when a keyword is composed of a plurality of tokens. For each token constituting the keyword, the related word acquisition unit 215 searches the related word graph database 234 for node information whose name matches the token (S701), and if one or more node information is searched ( (S702: YES), common synonym acquisition processing is performed (S703), common subordinate word acquisition processing is performed (S704), mediation word acquisition processing is performed (S705), and optimal mediation word acquisition processing is performed (S706). ). Hereinafter, common synonyms, common subordinate terms, mediation words, and optimal mediation words will be described, and each acquisition process will be described.

== Common synonyms ==
A common synonym of a plurality of tokens is a node (a leaf node or a redirect) connected to a lower level of an upper node connected to all tokens in the related word graph 10. As shown in FIG. 28, when there is an upper node 805 connected to both tokens 801 and 802 by edges 803 and 804, the node 806 connected to the lower side of the node 805 is connected to the tokens 801 and 802. Become a common synonym.
FIG. 29 is a diagram illustrating a flow of processing for acquiring a common synonym. The related word acquisition unit 215 performs the following processing using each of the node information as the first node information.
The related word acquisition unit 215 searches the related word graph database 234 for edge information whose child ID matches the ID of the first node information as parent edge information (S721), and performs the following processing for each of the searched parent edge information I do.
The related word acquisition unit 215 uses each of the other node information as the second node information, and from the related word graph database 234, the child ID matches the ID of the second node information, and the ID is the ID of the parent edge information. The matching edge information is searched as child edge information (S722), and if the child edge information cannot be searched (S723: NO), the processing of steps S724 to S727 is skipped.
When the related word acquisition unit 215 can search the child edge information for all the second node information (S723: YES), the related word acquisition unit 215 obtains the edge information whose ID matches the ID of the parent edge information from the related word graph database 234 as the child edge. Information is retrieved (S724), and for each retrieved child edge information, node information whose ID matches the child ID of the edge information is obtained from the related word graph database 234 as child node information (S725). If the type is “leaf” or “redirect” (S726: YES), the name of the child node information is added to the related word list (S727).
By repeating the above processing for each parent edge information and each first node information, leaf nodes and redirects connected to the lower order of nodes connected to the higher order of all tokens are acquired as common synonyms.

== Common narrower terms ==
A common low-order word of a plurality of tokens is a node (leaf node or redirect) connected to the low-order of all tokens in the related word graph 10. As shown in FIG. 30, when the token 821 has nodes 823 and 824 connected to the lower level and the token 822 has only the node 824 connected to the lower level, the nodes 824 connected to all the lower levels of the tokens 821 and 822 have Becomes a common narrower term.
FIG. 31 is a diagram showing a flow of processing for acquiring a common narrower word. The related word acquisition unit 215 performs the following processing using each of the node information as the first node information.
The related word acquisition unit 215 searches the related word graph database 234 for edge information whose ID matches the ID of the first node information as first child edge information (S741), and for each of the searched first child edge information. The following processing is performed.
The related word acquisition unit 215 uses each of the other node information as the second node information, and from the related word graph database 234, the ID matches the ID of the second node information, and the child ID is the first child edge information. Edge information that matches the child ID is searched as second child edge information (S742), and if the second child edge information cannot be searched (S743: NO), the processing of steps S744 to S746 is skipped.
The related word acquisition unit 215, when the second child edge information can be searched for all the second node information (S743: YES), the node whose ID matches the ID of the first child edge information from the related word graph database 234 Information is acquired as child node information (S744), and if the type of child node information is “leaf” or “redirect” (S745: YES), the name of the child node information is added to the related term list (S746).
By repeating the above processing for each child edge information and each first node information, leaf nodes and redirects connected to the lower order of all tokens are acquired as common narrower terms.

== Medium ==
An intermediary word of a plurality of tokens is a node (leaf node or redirect) existing on a path connecting two tokens in the related word graph 10. As shown in FIG. 32, a node 833 existing on the path connecting the token 831 to the token 832 serves as an intermediary word. In FIG. 32, for the tokens 831 and 832, the node 834 is not an intermediary word.
FIG. 33 is a diagram showing a flow of processing for acquiring a mediation word. The related word acquisition unit 215 performs the following processing for each combination of two pieces of node information.
The related word acquisition unit 215 searches the related word graph database 234 for a path from the first node corresponding to one of the two pieces of node information to the second node corresponding to the other (S761). The route from the first node to the first node is also searched (S762). A route between two nodes is searched by a route search process in a general graph. The related word acquisition unit 215 specifies a node (hereinafter, referred to as “routed node”) that passes through the route for each of the searched routes (S763), and the node information corresponding to the specified routed node is stored in the related word graph database. It acquires from 234 (S764). If the type of the route node information is “leaf” or “redirect” for each of the obtained route node information (S765: YES), the related word acquisition unit 215 uses the name of the route node information as an intermediary word to list related words. (S766).

  By performing the above processing for each combination of two pieces of node information, a leaf node or a redirect existing on a path connecting two tokens included in the keyword is acquired as an intermediary word.

== Optimal mediator ==
An optimal mediation word of a plurality of tokens is a node (a leaf node or a redirect) that exists on an optimal path among paths connecting two tokens in the related word graph 10. In the present embodiment, the optimum route is determined as the optimum route with the largest edge weight average value. The optimum route may be determined according to the cost assigned to the edge, for example, or a route with a small number of nodes on the route may be determined as the optimum route. Further, it may be determined according to the weights and costs assigned to the nodes and edges. As shown in FIG. 34, the route connecting token 841 and token 842 includes route abc and route defg, and the average weight of route abc is (50 + 100 + 60) ÷ 3 = 70, and route defg Since the average value of the weights is (50 + 50 + 50 + 50) ÷ 4 = 50, the optimum route is the route abc, and the nodes 843 and 844 included in the route abc are optimum mediation words.
FIG. 35 is a diagram for explaining the flow of processing for acquiring the optimum mediation word. The related word acquisition unit 215 performs the following processing for each combination of two pieces of node information.
The related word acquisition unit 215 searches the related word graph database 234 for a route from the first node corresponding to one of the two pieces of node information to the second node corresponding to the other (S781). A route from the first node to the first node is also searched (S782). A route between two nodes is searched by a route search process in a general graph. The related word acquisition unit 215 identifies an edge included in the route (hereinafter referred to as “routed edge”) for each of the retrieved routes (S783), and associates edge information corresponding to the identified routed edge with the related word graph database. A search is made from 234, and an average value of the weights of the searched edge information is obtained (S784). The related word acquisition unit 215 determines the route having the largest average weight value as the optimum route (S785), identifies the transit node via the optimum route (S786), and associates the node information corresponding to the identified transit node. Obtained from the word graph database 234 (S787). If the type of the route node information is “leaf” or “redirect” for each of the obtained route node information (S788: YES), the related word acquisition unit 215 uses the name of the route node information as the optimum mediation word. It adds to the list (S789).

  By performing the above processing for each combination of two pieces of node information, a leaf node or a redirect that exists on the optimum route with the smallest total weight among the routes connecting two tokens included in the keyword mediates. Obtained as a word.

  The related word acquisition unit 215 of the search device 20 acquires related words related to the keyword from the related graph 10 as described above. The related word output unit 216 adds a related word to the search result obtained by the search processing unit 214 and transmits the result to the terminal 22. FIG. 36 is a diagram illustrating an example of a screen 41 for displaying the search result. FIG. 37 is a diagram showing a part of the related term graph 10 used for the search processing. On the screen 41, related words 412 to 415 are displayed together with the search result 411. 37, in the related word graph 10 in the example of FIG. 37, “○ mountain ××” of the node 122 connected by the edge 421 from the node 132 of the keyword “rock artist” input in the input field 401 of the screen 40 described above. "Husband" was acquired as a lower association word. The related word 413 is obtained by acquiring “○ Saki ΔΔ” connected from the node 132 of “Rock Artist” at the edge 422 as a lower association word. The related word 414 is obtained by acquiring “lock” of the node 116 connected to the node 132 of “rock artist” by the edge 423 as a high-level association word. The related word 415 is obtained by acquiring the “hit chart” of the node 134 connected by the edge 425 from the node 131 connected by the edge 424 to the node 132 of the “rock artist” as an associative word. . In this way, in the example of FIG. 37, the screen 41 displays the lower association words, the upper association words, and the peer association words. Similarly, when a synonym or a lower term is searched, it can be displayed as a related word. Further, when a plurality of tokens are set for the keyword, a common synonym, a common subordinate word, and an intermediary word can be acquired as related words from the related word graph 10 and displayed on the screen 41. When any of the related words 412 to 415 is clicked on the screen 41, a query including the clicked related word as a keyword is transmitted from the terminal 22 to the search device 20, and the above-described search processing and related word acquisition processing are performed. Is done.

  As described above, in the search device 20 of this embodiment, the topic graph 12 can be merged with the thesaurus 11 by using the category graph 13. By combining the thesaurus 11 and the topic graph 12, it is possible to create a related term graph 10 composed of a wide range of words such as general words, proper nouns, and technical terms. Therefore, the possibility that the token matches the word included in the related word graph 10 can be increased. Thereby, the related word relevant to a token can be provided reliably.

  Further, since the related word graph 10 has a graph structure, related words connected via an edge can be easily obtained starting from a node that matches the keyword included in the query. Therefore, the search device 20 of the present embodiment can provide query suggestions quickly.

  In addition, the search device 20 of the present embodiment includes a synonym, a subordinate word, a subordinate association word, a high association word, a peer association word, a common synonym, a common subword, a mediation word, and an optimal mediation word (hereinafter referred to as “synonym etc.”). Etc.)), and related words that evaluate the relevance of the token from various viewpoints. Therefore, the user can easily select a search keyword to be newly refined with reference to related words.

  Although the present embodiment has been described above, the above embodiment is intended to facilitate understanding of the present invention and is not intended to limit the present invention. The present invention can be changed and improved without departing from the gist thereof, and the present invention includes equivalents thereof.

  For example, in the present embodiment, it is assumed that a query is transmitted from the terminal 22 to the search device 20 via the communication network 21, but the search device 20 may directly receive a keyword input from the user. Further, not only at the time of document search, but also various related words related to the keyword can be presented.

  In the present embodiment, the thesaurus 11, the topic graph 12, and the category graph 13 are all directed graphs. However, the thesaurus 11, the topic graph 12, and the category graph 13 may be a tree structure that does not have a circular structure, or may be an undirected graph. Good. When an undirected graph is employed, in the process of searching for edge information using an ID or child ID as a key, both edge information searching using the ID as a key and edge information searching using the child ID as a key are performed. Like that.

  In this embodiment, synonyms and the like are presented as related words for the tokens constituting the keywords included in the query. However, the present invention is not limited to this, and in the related word graph 10, the node corresponding to the token is the starting point. The various nodes connected at the edge can be presented as related words. For example, an upper node (higher word) connected to the token at the edge may be presented as a related word. Further, upper or lower nodes connected by two or more edges starting from the token may be presented as upper words or lower words.

  In the present embodiment, all nodes corresponding to synonyms and the like are acquired as related words. However, a predetermined number is determined according to the weight of the edge connecting the node corresponding to the synonym and the token and the token. You may make it present only.

10 Related Word Graph 11 Thesaurus 12 Topic Graph 13 Category Graph 20 Search Device 21 Communication Network 22 Terminal 201 CPU
202 Memory 203 Storage Device 204 Communication Interface 205 Input Device 206 Output Device 211 Data Acquisition Unit 212 Graph Merging Unit 213 Search Query Reception Unit 214 Search Processing Unit 215 Related Word Acquisition Unit 216 Related Word Output Unit 231 Thesaurus Database 232 Topic Graph Database 233 Category Graph Database 234 Related Words Graph Database

Claims (12)

A device for creating a related term graph that defines the relationship between words,
A thesaurus storage unit that stores a thesaurus that defines a conceptual hierarchy of words;
A topic graph storage unit that stores a topic graph that defines a relationship between words included in the corpus;
A category graph storage unit that stores a category graph that defines a hierarchy of categories used for directory searches;
The thesaurus, the category graph, such that words included in the first category and the thesaurus in the category graph are linked, and words included in the second category and the topic graph in the category graph are linked. And a graph creation unit that creates the related word graph having a graph structure by combining the topic graphs,
A related word graph creating apparatus comprising: The related word graph creation device according to claim 1,
The graph creation unit matches a word included in the thesaurus that matches the name of the first category in the category graph and a name of the second category belonging to a lower hierarchy of the first category. , Combining the topic graph with the thesaurus so that words included in the topic graph are linked;
Related word graph creation device characterized by The related word graph creation device according to claim 1,
The thesaurus, the topic graph, and the category graph are directed graphs,
The graph creation unit includes, for each edge between two categories in the category graph, a first word that matches the name of the first category among the two categories to which the edge is connected included in the thesaurus. And an edge connecting the first and second words is created when the topic graph includes a second word that matches the name of the second category of the two categories. Combining the topic graph with the thesaurus,
Related word graph creation device characterized by The related word graph creation device according to claim 3,
The thesaurus storage unit stores node information including a word corresponding to the node for each node constituting the thesaurus, and two identifications indicating two nodes for each edge between the nodes constituting the thesaurus Memorize edge information including information,
The topic graph storage unit stores the node information including a word corresponding to the node for each node constituting the topic graph, and the two identifications for each edge between the nodes constituting the topic graph. Storing the edge information including information;
The category graph storage unit stores the node information including a word indicating the category corresponding to the node for each node constituting the category graph, and for each edge between the nodes constituting the category graph. Storing the edge information including two pieces of the identification information;
The graph creation unit
Register the node information and the edge information stored in the thesaurus storage unit in the related word graph storage unit,
Register the node information and the edge information stored in the topic graph storage unit in the related word graph storage unit,
For each of the edge information stored in the category graph storage unit, a word set in the first node information corresponding to the first identification information included in the edge information is stored in the thesaurus storage unit. The word set in the second node information corresponding to the second identification information included in the edge information matches the word set in the second node information stored, and The identification information indicating the second node information and the identification information indicating the third node information when the word matches the word set in the third node information stored in the topic graph storage unit; Registering the edge information set to the related word graph;
Related word graph creation device characterized by A related word providing device for providing related words related to an input word,
The first category in the category graph that defines the hierarchy of categories used for directory search is linked to the words included in the thesaurus that defines the conceptual hierarchy of words, and is included in the second category and corpus in the category graph A related word for storing a related word graph having a graph structure created by combining the thesaurus, the category graph, and the topic graph so that the words included in the topic graph that define the relationship between the words are linked. A graph storage unit;
A word input unit for receiving input of the first word;
In the related word graph stored in the related word graph storage unit, a related word acquisition unit that acquires, as the related word, a second word connected from the first word via one or more edges. When,
A related word output unit for outputting the related word;
A related word providing device comprising: A related word providing device for providing related words related to an input word,
A thesaurus storage unit that stores a thesaurus that defines a conceptual hierarchy of words;
A topic graph storage unit that stores a topic graph that defines a relationship between words included in the corpus;
A category graph storage unit that stores a category graph that defines a hierarchy of categories used for directory searches;
The thesaurus, the category graph, such that words included in the first category and the thesaurus in the category graph are linked, and words included in the second category and the topic graph in the category graph are linked. And a graph creation unit that creates the related word graph having a graph structure by combining the topic graphs,
A related word graph storage unit for storing the related word graph;
A word input unit for receiving input of the first word;
In the related word graph stored in the related word graph storage unit, a related word acquisition unit that acquires, as the related word, a second word connected from the first word via one or more edges. When,
A related word output unit for outputting the related word;
A related word providing device comprising: A method for creating a related term graph that defines relationships between words,
Computer
Store a thesaurus that defines the conceptual hierarchy of words in memory;
Storing in the memory a topic graph defining relationships between words contained in the corpus;
Storing in the memory a category graph defining a hierarchy of categories used for directory searches;
The thesaurus, the category graph, such that words included in the first category and the thesaurus in the category graph are linked, and words included in the second category and the topic graph in the category graph are linked. And combining the topic graphs to create the related word graph in a graph structure,
A related word graph creation method characterized by A method of providing related terms related to an input word,
The first category in the category graph that defines the hierarchy of categories used for directory search is linked to the words included in the thesaurus that defines the conceptual hierarchy of words, and is included in the second category and corpus in the category graph A computer that stores a related word graph having a graph structure created by combining the thesaurus, the category graph, and the topic graph so that words included in a topic graph that defines a relationship between words are linked to each other ,
Accepts the input of the first word,
In the related word graph stored in the related word graph storage unit, a second word connected from the first word via one or more edges is acquired as the related word,
Outputting the related word;
A related word providing method characterized by A method of providing related terms related to an input word,
Computer
Store a thesaurus that defines the conceptual hierarchy of words in memory;
Storing in the memory a topic graph defining relationships between words contained in the corpus;
Storing in the memory a category graph defining a hierarchy of categories used for directory searches;
The thesaurus, the category graph, such that words included in the first category and the thesaurus in the category graph are linked, and words included in the second category and the topic graph in the category graph are linked. And the topic graph are combined to create the related word graph having a graph structure,
Storing the related word graph in the memory;
Accepts the input of the first word,
In the related word graph stored in the memory, a second word connected from the first word via one or more edges is acquired as the related word,
Outputting the acquired related terms;
A related word providing method characterized by A program for creating a related term graph that defines relationships between words,
On the computer,
Storing in a memory a thesaurus defining a conceptual hierarchy of words;
Storing in the memory a topic graph defining relationships between words included in the corpus;
Storing in the memory a category graph defining a hierarchy of categories used for directory searches;
The thesaurus, the category graph, such that words included in the first category and the thesaurus in the category graph are linked, and words included in the second category and the topic graph in the category graph are linked. And combining the topic graphs to create the related word graph in a graph structure;
A program for running A program for providing related words related to an input word,
On the computer,
The first category in the category graph that defines the hierarchy of categories used for directory search is linked to the words included in the thesaurus that defines the conceptual hierarchy of words, and is included in the second category and corpus in the category graph A related word graph having a graph structure created by combining the thesaurus, the category graph, and the topic graph is stored in a memory so that words included in a topic graph that defines a relationship between words are linked to each other. Steps,
Receiving an input of a first word;
Obtaining, as the related word, a second word connected from the first word via one or more edges in the related word graph stored in the memory;
Outputting the related terms;
A program for running A program for providing related words related to an input word,
On the computer,
Storing in a memory a thesaurus defining a conceptual hierarchy of words;
Storing in the memory a topic graph defining relationships between words included in the corpus;
Storing in the memory a category graph defining a hierarchy of categories used for directory searches;
The thesaurus, the category graph, such that words included in the first category and the thesaurus in the category graph are linked, and words included in the second category and the topic graph in the category graph are linked. And combining the topic graphs to create the related word graph in a graph structure;
Storing the related word graph in the memory;
Receiving an input of a first word;
Obtaining, as the related word, a second word connected from the first word via one or more edges in the related word graph stored in the memory;
Outputting the acquired related terms;
A program for running

Images