Genre generation apparatus and genre generation program

Abstract

To provide an apparatus and a program for generating a genre on a scale desired by a user while comprehensively utilizing repository information.SOLUTION: A genre generation apparatus 1 includes: a title extraction unit 10 and a hierarchy extraction unit 20 for extracting data from a repository; an initial setting unit 30 for setting an initial state tree including a root node; a growth processing unit 40 for repeatedly executing a growth operation of generating child nodes and assigning lower hierarchy classification names thereto; a title assignment unit 50 for assigning a title belonging to each classification name to each node; and a merge processing unit 60 for executing merge processing. The growth processing unit 40 executes the generation operation in a state where the classification names already assigned to the other nodes and the classification names including prescribed excluded words are excluded from the lower hierarchy classification names.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an apparatus and a program for generating genres for classifying people and things. [Background technology]

[0002] Conventionally, it has been common to use categories in data processing, and their usefulness has been widely recognized. Here, a category is a hierarchical classification name for classifying people and things. For example, Patent Document 1 describes a face recognition device that can identify a person from a video including a face by limiting the category to which the person belongs. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2021-136038 A Summary of the Invention [Problem to be solved by the invention]

[0004] Category information can be extracted, for example, from repositories that satisfy certain conditions. On the other hand, in classification, it is preferable that the same classification name is used only once without duplication. In the following, "genre" is used to mean a classification name that is unique. When generating genres from a repository, if the information to be read is limited to avoid duplication, the repository information cannot be used comprehensively. Also, if the genres are set in detail, the scale of classification becomes large, which makes it difficult for some users to use.

[0005] The present invention has been made to solve the above problems, and has an object to provide an apparatus and a program for generating genres on a scale desired by a user while comprehensively utilizing repository information. [Means for solving the problem]

[0006] In order to solve such problems, a genre generation device according to an embodiment is a genre generation device that generates genre data for classifying people and things in the form of a tree-structured graph from a repository having information on people and things and the hierarchical classifications to which they belong, and is equipped with a title extraction unit that extracts title data including titles, which are the names of people and things, and category names to which the titles belong, from the repository, a hierarchy extraction unit that extracts hierarchical data including the category names and category names of a lower hierarchy than the category names, an initial setting unit that sets an initial state tree including a root node and in which the category names are assigned to each node, a growth processing unit that repeats a generation operation of generating child nodes and assigning category names of the lower hierarchy to them, starting from leaf nodes of the initial state tree and moving from nodes of a shallower hierarchy to nodes, and a title assignment unit that assigns titles belonging to category names to each node, and the growth processing unit performs the generation operation by excluding category names already assigned to other nodes and category names including predetermined excluded words from the category names of the lower hierarchy.

[0007] Moreover, the genre generation program according to the embodiment causes a computer to function as the genre generation device according to the embodiment. Effect of the Invention

[0008] According to the present invention, it is possible to provide an apparatus and a program for generating genres on a scale desired by a user while comprehensively utilizing repository information. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a block diagram illustrating a genre generation device according to an embodiment. [Figure 2A] 11 is a table illustrating an example of a data structure of title data. [Figure 2B] 13 is an example of title data. [Figure 3A] 1 is a table illustrating an example of a data structure of hierarchical data. [Figure 3B] 1 is an example of hierarchical data. [Figure 4] 1 is an example of an initial state tree. [Diagram 5] 11 is a tree structure graph illustrating the processing in the growth processing unit. [Figure 6] Here is an example of a big genre tree with assigned titles. [Figure 7] 11 is a tree structure graph illustrating a process in a merge processing unit. [Figure 8] 13 is an example of a genre tree after a merge process. [Figure 9] 1 is an example of a tree structure graph for calculating a similarity. [Figure 10] 1 is a flowchart illustrating an operation of the genre generation device according to the embodiment. [Figure 11] 13 is a flowchart illustrating an operation of a growth processing unit. [Figure 12] 13 is a flowchart illustrating an operation of a merge processing unit. [Figure 13A] 1 is an example of a tree structure graph. [Figure 13B] FIG. 13B is an enlarged view of a portion of FIG. 13A. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] [Genre generation device] The genre generation device 1 according to the embodiment is a device that extracts data from a repository and generates genre data in the form of a tree-structured graph. The repository is an electronic collection of a wide range of information, and can use external information. The condition of a repository is that it has information on people and things and the hierarchical classification to which they belong. In the following, we will explain the case of using Wikipedia (registered trademark) as a repository as an example. Wikipedia is an online encyclopedia that contains a huge number of articles. Each article has a "title" and has a hierarchical "category" that classifies the articles.

[0011] Wikipedia has a restriction that different articles cannot be created with the same title, and each article lists the category to which it belongs. An article may belong to one category or multiple categories. Each category has a page that lists the titles belonging to that category, the categories at a higher level, and the categories at a lower level. For this reason, Wikipedia meets the conditions of a repository that can be used by the genre generation device 1. A category at a higher level is a category one level above which that category is classified. A category at a lower level is a category one level below which that category is classified. In the genre generation device 1, the name of a person or thing is called a "title." A title corresponds to a title in Wikipedia. Moreover, a name of a hierarchical classification is called a "classification name." A classification name corresponds to a category in Wikipedia.

[0012] Here, we will explain the tree-structured graph. As shown in Fig. 13A and Fig. 13B, the tree-structured graph is composed of nodes and edges, and can represent the hierarchical relationships between the nodes. The top layer is the root node RN, and the further away from the root node RN, the lower the layer and the deeper the layer. The layer depth D can be counted by the number of edges from the root node RN to that node. The number of layers from the deepest layer to the root node RN is the height Ht of the tree-structured graph. The string of numbers in parentheses is the coordinate in tuple format, which will be described later. Also, the higher-level node connected by an edge is the parent node, and the lower-level node is the child node. Note that for one node, there may be zero, one or multiple child nodes, and one parent node. Also, a node with zero child nodes may be described as a leaf node. The genre generation device 1 assigns one classification name to one node of the tree structure graph. Also, the same classification name is not assigned to different nodes. In other words, the same classification name is not set at different positions in the classification. This ensures that classification names are not duplicated.

[0013] In this specification, a genre is a hierarchical classification of people or things, in which the same classification name is not assigned to different positions in the classification. A genre tree is a tree-structured graph in which one classification name is assigned to one node and the same classification name is not assigned to different nodes. As described later, the genre generation device 1 assigns to each node of the genre tree a title that belongs to the classification name of the node.

[0014] As illustrated in Fig. 1, the genre generation device 1 is a device that performs information processing, and includes a control unit 110 and a storage unit 120. The control unit 110 exchanges data with the storage unit 120 and the outside, and performs processing to generate genres. The control unit 110 includes a title extraction unit 10, a hierarchical extraction unit 20, an initial setting unit 30, a growth processing unit 40, and a title assignment unit 50, and further includes a merge processing unit 60 and a coordinate assignment unit 70 here. The control unit 110 can also include a group setting unit 55, a search unit 81, and a similarity calculation unit 82.

[0015] The storage unit 120 stores data to be processed by the control unit 110, data resulting from the processing, data input from outside, etc. The storage unit 120 includes an extracted data storage unit 121 and a genre tree storage unit 122. Data can be input and output between the genre generation device 1 and the outside via an input line 130 and an output line 140 .

[0016] Moreover, the genre generation device 1 can be connected to an external repository 300 via a line 200. The line 200 is a communication line such as the Internet or a LAN (Local Area Network). The repository 300 here is Wikipedia. Hereinafter, each component of the category generation device 1 will be described with reference to the drawings. As a specific example, a case where a category is generated for a sports player (person) will be described.

[0017] (Title extraction part) The title extraction unit 10 is a means for extracting title data including titles, which are names of people or things, and the names of categories to which the titles belong, from the repository 300. The extracted title data is stored in the extracted data storage unit 121. As shown in the example of FIG. 2A, the title data can be represented as a table that associates a "title" with its "category name." The "category name" is the name of the category to which the title belongs, and if the title belongs to multiple categories, all of them are extracted and listed. The listed "category names" can be made into unique data by managing them in a set structure. The title data needs to include at least a pair of the title and the category name to which the title belongs. The title data makes it possible to identify the category name to which the title belongs.

[0018] The title can be extracted by selecting a title according to the object to be classified. If the object to be classified is an athlete, the title of the person can be selected and extracted. For example, if a predetermined phrase is included in the information to which a title is given, it can be determined that it is the title of a person. In the case of Wikipedia, for example, it can be determined that the title is the title of a person if the title belongs to a category that includes the character strings "year of birth" or "year of death" or to the category "living people." As shown in Fig. 2B, the titles that were determined to be people and extracted include "Player P1" to "Player P6." Players P1 and P2 belong to the category of "baseball players of American teams," players P3 and P4 belong to the category of "players of team A," and players P5 and P6 belong to the category of "players of team B."

[0019] (Hierarchy extraction part) The hierarchy extraction unit 20 is a means for extracting hierarchical data including a classification name and a classification name of a lower hierarchy from the repository 300. The extracted hierarchical data is stored in the extracted data storage unit 121. As shown in the example of FIG. 3A, hierarchical data can be represented as a table that associates a "classification name" with its "subclassification name at a lower level." A subclassification name at a lower level is one level lower than the classification names classified under that classification name, and if there are multiple subclassification names, they are all extracted and listed. The listed "subclassification names at a lower level" can be made into unique data by managing them in a set structure. Hierarchical data needs to include at least a pair of a classification name and a classification name at a lower level than that classification name. The hierarchical data makes it possible to identify the classification names at a lower level.

[0020] As shown in Fig. 3B, in the hierarchy below the category "baseball players of each team," there are multiple categories including "baseball players of American teams," "baseball players of Japanese teams," and "baseball players of Cuban teams." In addition, in the hierarchy below the category "baseball players of American teams," there are multiple categories including "players of team A," "players of team B," and "players of team C."

[0021] (Initial setting section) The initial setting unit 30 is a means for setting an initial state tree ITR including a root node RN and with a classification name assigned to each node. The initial state tree ITR is set by creating a tree structure graph and assigning a classification name to each node. The initial setting unit 30 sets the initial state tree ITR by accepting preset contents or input from an input line 130. The initial state tree ITR set by the initial setting unit 30 is stored in the genre tree storage unit 122. It is preferable that a classification name is assigned to the initial state tree ITR based on a hierarchical relationship. The initial state tree ITR is the initial state of the tree structure graph, and the tree structure graph starts growing from the leaf nodes of the initial state tree ITR. The initial setting unit 30 can set the height of the initial state tree ITR and the number of branches of each node for a range including the root node RN. The initial state tree ITR may be only the root node RN, and in that case, the root node RN becomes a leaf node. Then, the initial setting unit 30 assigns one classification name to each node.

[0022] As shown in Fig. 4, the initial state tree ITR has a structure in which the root node RN has five child nodes. The categories of "sports player" are assigned to the root node RN, and the child nodes are "soccer player", "baseball player", "rugby football player", "basketball player", and "tennis player". Figs. 4, 6, and 8 are examples of screens displayed on a display or the like, in which nodes are displayed as rectangles and categories assigned to the nodes are written inside the rectangles.

[0023] It is preferable that the classification name assigned to the leaf node of the initial state tree ITR is a classification name included in the hierarchical data. Here, the category such as "soccer player" assigned to the leaf node exists in Wikipedia and is included in the hierarchical data. On the other hand, the hierarchy between the root node RN and the leaf node may be omitted. In Wikipedia, there is a category called "sports players by sport" in the hierarchy below "sports players", and the hierarchy below that is categories such as "soccer players". Here, the hierarchy of "sports players by sport" is omitted. However, categories such as "soccer players" are assigned to nodes in the same hierarchy below the hierarchy of "sports players", and the assignment is based on a hierarchical relationship.

[0024] The structure of the initial state tree ITR and the allocation of classification names to each node can be preset. However, if the preset classification name is not included in the hierarchical data, the growth process using the hierarchical data cannot be performed. Therefore, it is preferable to make it possible to select the classification name to be assigned from the classification names included in the hierarchical data. The initial setting unit 30 presents the user with options for classification names based on the hierarchical data, accepts the result of the selection from the options, and can set the initial state tree ITR from the root node RN to the hierarchy desired by the user according to the user's selection. When presenting options to the user and accepting the selection result, the initial setting unit 30 refers to the extracted data storage unit 121, outputs the category names included in the hierarchical data from the output line 140, and presents the options to the user. Then, the initial setting unit 30 accepts the user's selection result inputted from the input line 130.

[0025] In the example of FIG. 4, in assigning categories to five child nodes, the initial setting unit 30 first presents the user with options of which sports to select. The initial setting unit 30 presents the options based on the hierarchical data. Next, the initial setting unit 30 accepts, for example, five categories such as "soccer player" as a result of the selection. At this time, the number of child nodes may not be set in advance, and child nodes may be provided according to the number of sports to be selected. In addition, by further performing the presentation and reception, categories such as "US soccer player" and "Spanish soccer player" may be provided under "soccer player". The presentation of options may start from the root node RN. The initial setting unit 30 can set the initial state tree ITR up to the hierarchy desired by the user by repeating the presentation and reception.

[0026] (Growth Processing Section) The growing processing unit 40 is a means for repeatedly performing a generation operation, starting from the leaf nodes of the initial state tree ITR and moving from the nodes in the shallowest hierarchy to generate child nodes and assign classification names in the lower hierarchy. The growing processing unit 40 refers to the hierarchical data in the extracted data storage unit 121 to identify classification names in the lower hierarchy and perform the generation operation. The growing processing unit 40 grows the genre tree by repeatedly performing the generation operation. The output from the growing processing unit 40 is a big genre tree BTR, which is stored in the genre tree storage unit 122. The big genre tree BTR is a genre tree that has completed processing in the growing processing unit 40. The growth processing unit 40 performs work for each node based on the hierarchical data. The node on which work is performed is referred to as a working node. Work is performed from nodes in shallower layers, and when there are no more nodes in that layer that have not been worked on (hereinafter referred to as unworked nodes), it moves to a lower layer. A layer in which a working node is located is referred to as a processing layer. As illustrated in FIG. 5, in the growth process, for example, in node N02, which is a working node, two child nodes N04 and N05 are generated and assigned classification names C04 and C05. At this time, node N03 is an unworked node. When the processing of the layers of nodes N02 and N03 is completed, the processing layer moves in the direction of arrow A1.

[0027] If the classification name of the working node has a lower-level classification name that satisfies the condition, the growth processing unit 40 generates a child node and assigns it. If there are multiple lower-level classification names that satisfy the condition, the growth processing unit 40 generates child nodes of the same number and assigns classification names to them. If there are no lower-level classification names that satisfy the condition, no child nodes are generated. The conditions are that the classification name has not been assigned to another node and that the classification name does not include a predetermined excluded word. In other words, the growing processing unit 40 performs the generation work by removing the classification names that have already been assigned to another node and the classification names that include a predetermined excluded word from the classification names of the lower hierarchy. Including an excluded word means that the classification name includes the character string of the excluded word.

[0028] In Wikipedia, the same category may exist as a lower hierarchical level of another category. In this case, if the category is assigned according to the hierarchical data, the category already assigned will be assigned to another node. The growth processing unit 40 performs the generation work excluding the category (classification name) already assigned to another node, thereby generating a genre tree in which the same category is not assigned to different nodes. The predetermined excluded words may be, for example, words expressing functions in the repository or names of people. In the case of Wikipedia, the excluded words may be, for example, "category", "subject", "template", "stub", or titles determined by the title extraction unit 10 to be people.

[0029] The growth processing unit 40 repeats the generation work until there are no nodes to move. A state where there are no nodes to move means that there are no unprocessed nodes in the processing hierarchy and there are no unprocessed hierarchies. The growth processing unit 40 generates a child node if there is a classification name in a lower hierarchy that satisfies the conditions, so an unprocessed hierarchy occurs in the hierarchy one level below. However, if no child node has been generated for any of the working nodes in a certain processing hierarchy, there is no unprocessed hierarchy and there are no nodes to move. And there are no classification names to assign.

[0030] The growth processing unit 40 continues the process in this manner until all nodes no longer have any lower-level category names that satisfy the conditions. The generated big genre tree BTR is grown from the initially set genre tree to cover all the category name information in the repository, and has a hierarchical structure that covers all the genre fields that the user wants to generate. The growth processing unit 40 can generate a big genre tree BTR that covers all the hierarchical classification name information of the repository without duplication, with the classification name assigned to the root node RN at the top. The big genre tree BTR has the largest height Ht among the genre trees grown from the set initial state tree ITR.

[0031] (Title assignment section) The title assignment unit 50 is a means for assigning titles belonging to a category name to each node. One category name is assigned to each node of the big genre tree. The title assignment unit 50 refers to the title data in the extracted data storage unit 121 to identify the category name to which the title belongs. Then, it reads the big genre tree BTR in the genre tree storage unit 122 and assigns titles to each node. The output from the title assignment unit 50 is the big genre tree BTR to which the titles are assigned, and is stored in the genre tree storage unit 122. Note that if a title belongs to multiple category names, the same title will be assigned to multiple nodes. 6, the title of a person belonging to the category of each node of the big genre tree BTR is assigned to the node by the title assignment unit 50. For example, the node of the category "United States baseball team players" is assigned the titles "Player P1," "Player P2," ..., which belong to the category.

[0032] (Group setting section) The group setting unit 55 is a means for grouping titles not assigned to the tree structure graph with a designated title designated by the user from among the titles assigned to the tree structure graph, and assigning the titles to the same node as the designated title. The titles received from the user are input from an input line 130. The group setting unit 55 creates group setting data in which the designated title and a title to be grouped with the designated title are paired. The group setting unit 55 can also read the big genre tree BTR to which titles are assigned in the genre tree storage unit 122, and assign the titles to be grouped. The output from the group setting unit 55 is the group setting data and the big genre tree BTR to which the titles to be grouped are assigned, and these are stored in the genre tree storage unit 122. For example, a new athlete may not have information in the repository and cannot be assigned to the big genre tree BTR. The group setting unit 55 can group a person who has no information in the repository with one of the people who the user considers to be close among the people who have been assigned to the big genre tree BTR, and assign the person to the same node as that person.

[0033] The group setting unit 55 accepts user designation of people the user wants to add to and assign to the genre tree and people to be grouped, and performs grouping. The group setting unit 55 may request the user to designate these people. The group setting unit 55 can perform grouping after the title assignment by the title assignment unit 50. However, in the case of a person whose information is not available in the repository and who is known in advance not to be included in the title data, grouping may be performed in advance. When performing grouping in advance, the group setting unit 55 may create group setting data, and the title assignment unit 50 may assign titles to the big genre tree BTR by referring to the group setting data together with the title data.

[0034] (Merge processing section) The merge processing unit 60 is a means for assigning the category name of the node at the deepest level and the title belonging to it to the parent node and deleting the node at the deepest level. The merge processing unit 60 reads the big genre tree BTR to which titles have been assigned from the genre tree storage unit 122, and repeatedly performs merging processing so that the tree structure graph reaches a predetermined height. The merge processing unit 60 degenerates the big genre tree BTR to which titles have been assigned to each node into a genre tree RTR of a predetermined height and outputs it. The genre tree RTR is stored in the genre tree storage unit 122.

[0035] As shown in Fig. 7, in the merge process, for example, for nodes N04 and N05 in the deepest hierarchy, the classification names C04 and C05 and the titles W04 and W05 belonging to them are assigned to node N02, which is the parent node. For nodes N06 and N07 in the deepest hierarchy, the classification names C06 and C07 and the titles W06 and W07 belonging to them are assigned to node N03, which is the parent node. Then, nodes N04-N07 are deleted. The processing hierarchy moves in the direction of arrow A2. Note that the classification names C04-C07 assigned to the nodes N04-N07 to be deleted may be deleted without being assigned to the parent node.

[0036] As a result of the processing in the merge processing unit 60, as shown in Fig. 8, in the node of the category "Baseball players of American teams", child nodes are deleted, and the titles "Player P3", "Player P4", "Player P5", and "Player P6" that were assigned to the child nodes are assigned in addition to the original titles "Player P1" and "Player P2". The height Ht of the genre tree RTR is lower than the big genre tree BTR. Note that here, the categories "Team A players" and "Team B players" that were assigned to the child nodes are deleted without being assigned to the parent node. The merge processing unit 60 can generate a genre tree RTR of a predetermined size while covering all the information in the repository.

[0037] The predetermined height of the tree structure graph in the processing of the merge processing unit 60 can be a fixed value. On the other hand, if the user can specify the height in accordance with the processing in the merge processing unit 60, convenience can be improved. Therefore, it is preferable that the merge processing unit 60 accepts the height of the tree structure graph specified by the user and sets the height as the predetermined height. The merge processing unit 60 accepts the height specification input from the input line 130. Also, it is possible to request the user to specify the height of the tree structure graph. The height specification by the user may be performed when setting the initial state tree ITR.

[0038] (Coordinate assignment part) The coordinate assignment unit 70 is a means for assigning coordinates in the form of a tuple to nodes of a tree structure graph. The coordinate assignment unit 70 reads the genre tree RTR in the genre tree storage unit 122, assigns coordinates, and outputs the genre tree RTR with the assigned coordinates. The genre tree RTR with the assigned coordinates is stored in the genre tree storage unit 122 and output from the output line 140. By assigning coordinates in tuple format, the node position of each genre in the genre tree RTR can be managed by coordinates. Also, by expressing the nodes of the genre tree in tuple format, it is possible to efficiently calculate the similarity between genres, which will be described later. Furthermore, the distance between nodes can be calculated with a simple algorithm, and it can also be applied to the grouping process of items when registering them in a database.

[0039] Coordinates in tuple format are vector coordinates such as (0,0,...). Coordinates are assigned by increasing the degree of child nodes by one to (0,0,...,X), and changing the value of X to 0, 1, 2,... according to the number of child nodes. The root node RN is represented as (0,). For example, the child nodes of the root node (0,) are assigned as (0,0), (0,1), (0,2),.... The child nodes of node (0,1) are assigned as (0,1,0), (0,1,1),.... The degree of coordinates is the value obtained by adding 1 to the value of the hierarchical depth D. For example, for a node whose hierarchical depth D value is N, the degree is N+1.

[0040] (Search section) The search unit 81 is a means for receiving a title specified by a user, searching for and outputting a category name to which the title belongs in a tree structure graph. The title specified by the user is input from an input line 130. The search unit 81 references a genre tree RTR to which the title in the genre tree storage unit 122 is assigned, and outputs a category name to which the specified title belongs from an output line 140. If the title belongs to multiple categories, then those categories are output. If the title being searched for is not assigned to a node in the genre tree, then no category is output. The search unit 81 can check whether the genre tree RTR includes a person of interest, and can immediately identify the genre to which the person belongs.

[0041] (Similarity calculation part) The similarity calculation unit 82 is a means for calculating the similarity between category names in a tree structure graph using coordinates in the tuple format. The category names for which similarity is to be calculated are input from an input line 130. The similarity calculation unit 82 refers to the genre tree RTR to which the coordinates in the tuple format are assigned in the genre tree storage unit 122, calculates the similarity between the designated category names, and outputs the calculated similarity from an output line 140. To the node to which a classification name is assigned, coordinates in a tuple format are assigned by the coordinate assignment unit 70. The similarity calculation unit 82 calculates and outputs a similarity from two coordinates in a tuple format.

[0042] With reference to FIG. 9, the case of calculating the similarity between node NA and node NB will be described. First, the paths from the root node RN to node NA and node NB are traced. The two paths pass through the same node NC at a layer with a layer depth of D=1. Then, they split into different nodes at a layer with a layer depth of D=2. In this case, the similarity is calculated by dividing the layer depth D=2 at which they split into different nodes by the maximum depth. In this example, the maximum depth is 5, so the similarity is 2 divided by 5, or 0.4. The similarity calculated by the similarity calculation unit 82 can be used to check whether the similarity between category names in the generated genre tree RTR is within an expected range.

[0043] The genre generation device 1 with the above configuration can adjust the size of the genre tree while retaining information by performing a merge process after a growth process. In addition, after performing a growth process that incorporates all data that meets certain conditions to generate a big genre tree, the height can be adjusted by a merge process, thereby reducing the size of the genre tree while comprehensively utilizing the information in the repository. By keeping the genre tree at a moderate size, data management is also made easier.

[0044] The genre generation device 1 can set the starting point and range of classification in the initial state tree, and can also set how detailed the classification should be by specifying the height of the genre tree. This makes it possible to generate a genre tree that includes all information about the genre field desired by the user, even if the genre tree is made small. The genre generation device 1 can efficiently and effectively generate a genre tree that covers a wide range of information with little effort, even if the repository has a large amount of information. In addition, it can automatically assign people and things to the genre tree, thereby reducing costs and effort.

[0045] The data extracted from the repository may be one type, not two types of title data and hierarchical data, but a set of three items: a category name, a title belonging to the category name, and a category name of a lower hierarchy. The genre generation device 1 may also have an internal repository.

[0046] [Genre generator in action] Next, the operation of the genre generation device 1 will be described with reference to Fig. 10. Note that here, a case will be described in which Wikipedia is used to generate a genre tree of sports players (people). In step S10, the title extraction unit 10 extracts title data (FIG. 2B) from Wikipedia, which is a pair of a person's title and a category to which the person belongs. In step S20, the hierarchy extraction unit 20 extracts hierarchical data (FIG. 3B) from Wikipedia, which is a pair of a category and a category in a lower hierarchy of the category. Note that steps S10 and S20 may be performed in either order, or may be performed in parallel.

[0047] In step S30, the initial setting unit 30 sets an initial state tree ITR including a root node RN. Here, the initial state tree ITR is set with "athlete" as the root node RN and athlete categories for five sports as child nodes (FIG. 4). The initial setting unit 30 refers to the hierarchical data to present options for categories to be child nodes of athletes, and can assign "baseball player" and the like by accepting the selected result. In step S40, the growth processor 40 starts from the leaf nodes of the initial state tree ITR, and while moving from the nodes in the shallowest layers, repeats the generation work of generating child nodes and assigning classification names of lower layers to generate the big genre tree BTR. The growth processor 40 excludes categories that have already been assigned to other nodes and categories that include predetermined excluded words such as words and phrases that indicate functions, and assigns them to the child nodes. Then, the process continues until there are no more nodes to move.

[0048] The process (step S40) in the growth processing unit 40 can be performed according to the flowchart illustrated in FIG. 11. First, a leaf node of the initial state tree ITR is set as a working node (step S41). Then, it is determined whether or not the classification name assigned to the working node has a classification name of a lower hierarchy that satisfies the condition (step S42). If there is a classification name of a lower hierarchy that satisfies the condition, a child node is generated and the classification name is assigned (step S43). Next, it is determined whether or not there is an unprocessed node in the processing hierarchy (step S44). If there is an unprocessed node, it moves to that node (step S45) and performs step S42. If there is no unprocessed node, it is determined whether or not there is an unprocessed hierarchy below the current processing hierarchy (step S46). If there is an unprocessed hierarchy, the processing hierarchy is lowered by one (step S47), and step S42 is performed for each working node. If there is no unprocessed hierarchy, the big genre tree is output (step S48), and the process ends.

[0049] Next, in step S50, the title assignment unit 50 assigns to each node of the big genre tree BTR a player who belongs to the category of that node (FIG. 6). In step S55, the group setting unit 55 groups new players who do not yet have articles created on Wikipedia with players who have already been assigned, and assigns them to the same category as those players. In step S60, the merge processor 60 assigns the title in the deepest hierarchy to a parent node and repeatedly deletes the node in the deepest hierarchy to generate a genre tree RTR that is lowered to a predetermined height. The predetermined height can be specified by the user, and may be specified, for example, when setting the initial state tree ITR.

[0050] The process (step S60) in the merge processing unit 60 can be carried out according to the flowchart illustrated in FIG. 12. First, the deepest layer is set as the processing layer (step S61). Then, it is determined whether the height of the tree structure graph is higher than a predetermined height (step S62). If it is not higher than the predetermined height, the genre tree RTR is output (step S66) and the process ends. If it is higher than the predetermined height, the category name of each node in the processing layer and the title belonging to it are assigned to each parent node (step S63). Next, the processing layer is raised by one (step S64), and the node of the previous processing layer, which is the layer one level lower, is deleted (step S65), and the process returns to step S62.

[0051] In step S70, the coordinate assignment unit 70 outputs the genre tree RTR to which coordinates in the tuple format have been assigned. In step S81, the search unit 81 searches for the specified player and outputs the category to which the player belongs in the genre tree RTR. In step S82, the similarity calculation unit 82 calculates and outputs the similarity between categories. Note that steps S81 and S82 can be performed when necessary.

[0052] The genre generation device 1 operates as described above, and can generate genres on a scale desired by the user while comprehensively utilizing the information in the repository. In addition, the title may be assigned to a node in the step (S43) of generating a child node and assigning a lower hierarchical classification name in the process (S40) of the growth processing unit 40, in which the title belonging to the classification name is also assigned.

[0053] The genre generation device 1 can also be realized by operating a computer with a genre generation program. In other words, a computer into which the genre generation program is loaded becomes the genre generation device 1. The genre generation program is a program for causing a computer to function as the genre generation device 1. The genre generation program can be obtained through an electric communication line and can be recorded on a computer-readable recording medium. [Explanation of symbols]

[0054] 1. Genre Generator 10 Title extraction part 20 Hierarchy extraction part 30 Initial setting section 40 Growth Processing Section 50 Title Allocation Section 55 Group Settings 60 Merge Processing Section 70 Coordinate assignment section 81 Search Department 82 Similarity calculation part 110 Control section 120 Storage section 121 Extracted data storage unit 122 Genre Tree Memory 130 Input Line 140 Output Line 200 lines 300 repositories

Claims

1. A genre generation device that generates genre data for classifying people and things in the form of a tree structure graph from a repository having information on people and things and hierarchical classifications to which they belong, comprising: a title extraction unit that extracts title data including titles, which are names of people or things, and names of categories to which the titles belong, from the repository; a hierarchy extraction unit that extracts, from the repository, hierarchy data including the classification name and classification names at a lower level than the classification name; an initial setting unit that sets an initial state tree including a root node and in which the classification name is assigned to each node; a growth processing unit that repeats a generation operation of generating child nodes and assigning classification names of the lower hierarchical levels to the child nodes while moving from a leaf node of the initial state tree to a node in a shallow hierarchical level in order; a title assignment unit that assigns a title belonging to a classification name to each node, the growing processing unit performs the generating operation by excluding from the lower hierarchical classification names a classification name that has already been assigned to another node and a classification name that includes a predetermined excluded word; Genre generator.

2. a merge processing unit that performs a merge process of assigning a classification name of a node in the deepest hierarchy and a title belonging thereto to a parent node and deleting the node in the deepest hierarchy, The growth processing unit repeats the generation operation until there are no more nodes to be moved; the merging processing unit repeatedly performs the merging process so that the tree structure graph has a predetermined height. The genre generating device according to claim 1 .

3. the initial setting unit presents options for classification names based on the hierarchical data to a user, accepts a result of a selection from the options, and sets the initial state tree from the root node to a hierarchy desired by the user according to the user's selection; the merge processing unit receives a height of the tree structure graph designated by a user, and sets the height to the predetermined height; The genre generating device according to claim 2 .

4. and a search unit that receives a title specified by a user, searches for and outputs a category name to which the title belongs. The genre generating device according to claim 1 .

5. a group setting unit for grouping a title not assigned to the tree structure graph with a designated title designated by a user from among the titles assigned to the tree structure graph, and assigning the designated title to the same node as the designated title; The genre generating device according to claim 1 .

6. a coordinate assignment unit that assigns coordinates in a tuple format to nodes of the tree structure graph; a similarity calculation unit that calculates a similarity between classification names in the tree structure graph using the coordinates, The genre generating device according to claim 1 .

7. A genre generation program for causing a computer to function as the genre generation device according to any one of claims 1 to 6.

Images