JP6622236B2 - Idea support device and idea support method - Google Patents

Description

  The present invention relates to an idea support device and an idea support method.

  As a background art in this technical field, there is JP-A-2007-241901 (Patent Document 1). This gazette includes: “a discussion point extracting means for extracting a plurality of issues related to the theme from a group of opinion texts related to the input theme, and the opinion including the issues for each of the plurality of issues. Specificity calculation means for calculating the degree of uniqueness that expresses the proportion of opinions that have a single position in the issue from the text, and importance for calculating the importance of opinions that include the single position for each of the multiple issues Degree calculation means, related word extraction means for extracting related words related to the plurality of issues, representative opinion selection means for selecting a plurality of representative opinions in affirmative or negative opinions for each of the plurality of issues, And interface means for outputting the uniqueness and the importance for each of the plurality of issues output from the means ”(refer to the summary).

JP 2007-241901 A

  For example, it is assumed that a user performs a composition or utterance on a given theme. The technique described in Patent Document 1 supports writing and utterance by a user by extracting in advance issues related to a given theme.

  However, when the user proceeds with composition and utterance, each of the issues extracted by the technique described in Patent Document 1 is not necessarily an issue that matches the composition or context of the conversation. In view of the above, an object of one embodiment of the present invention is to support a user's idea along the context of the composition or utterance by the user while the user is composing or uttering.

  In order to solve the above problems, one embodiment of the present invention employs the following configuration. An idea support device for supporting a user's idea, comprising: a processor and a memory, wherein the memory is a node between elements indicating a proposition included in a document defining knowledge, and between the elements indicated by the proposition The first type edge indicating the relation, the graph information indicating the structure of the first graph including the context, and the context information indicating the appearance order of the relation in the document, the processor accepts the input of the proposition, An element included in the proposition that has received the input and a relation between the elements indicated by the proposition that has received the input are extracted, and includes a node indicating the extracted element, and a relation between the extracted elements A second graph including the first type edge is generated, and the relation indicated by the second graph is determined by referring to the graph information. If it is determined to be included in full, by referring to said context information to identify a relationship that appeared in the document after the relations, and outputs information indicating the relation with the identified, creativity support device.

  According to one embodiment of the present invention, a user's idea can be supported along the context of the composition or utterance by the user while the user is composing or uttering.

  Problems, configurations, and effects other than those described above will be clarified by the following description of embodiments.

It is a block diagram which shows the structural example of the idea assistance apparatus in Example 1. FIG. It is a flowchart which shows an example of the process which the idea assistance apparatus in a present Example implements. 3 is an example of a knowledge graph indicating knowledge in a document according to the first exemplary embodiment. 3 is an example of a knowledge graph indicating knowledge in a document according to the first exemplary embodiment. 3 is an example of a knowledge graph indicating knowledge in a document according to the first exemplary embodiment. 4 is an example of a node table in the first embodiment. 3 is an example of an edge table in the first embodiment. 3 is an example of a knowledge graph indicating knowledge in a document according to the first exemplary embodiment. 3 is an example of a knowledge graph indicating knowledge in a document according to the first exemplary embodiment. 4 is an example of a node table in the first embodiment. 3 is an example of an edge table in the first embodiment. 3 is an example of a knowledge graph indicating knowledge in a document according to the first exemplary embodiment. 3 is an example of a knowledge graph indicating knowledge in a document according to the first exemplary embodiment. 3 is an example of a knowledge graph indicating knowledge in a document according to the first exemplary embodiment. 3 is an example of a knowledge graph indicating knowledge in a document according to the first exemplary embodiment. 3 is an example of a knowledge graph indicating knowledge in a document according to the first exemplary embodiment. 3 is an example of a knowledge graph indicating knowledge in a document according to the first exemplary embodiment. 4 is an example of a node table in the first embodiment. 3 is an example of an edge table in the first embodiment. 3 is an example of a knowledge graph indicating knowledge in a document according to the first exemplary embodiment. 4 is an example of a subsequent knowledge table in the first embodiment. 3 is an example of a knowledge graph indicating knowledge in a document according to the first exemplary embodiment. 4 is an example of a subsequent knowledge table in the first embodiment. 3 is an example of a knowledge graph generated by an utterance understanding process in the first embodiment. 3 is an example of a knowledge graph generated by an utterance understanding process in the first embodiment. 6 is an example of a transition of a screen displayed on the display device in the first embodiment.

  Embodiments of the present invention will be described below with reference to the accompanying drawings. It should be noted that this embodiment is merely an example for realizing the present invention, and does not limit the technical scope of the present invention.

  In order to cultivate expressive power, thinking power, etc., composition training (for example, essay writing) is carried out at schools. Unlike the knowledge that can be acquired through a collection of questions, writing is difficult to learn independently, and in many cases, individual instruction from the teacher is required to improve students' writing skills. In addition, when students are not used to writing, there is a problem that they do not know what to write first, and in many cases, the teacher needs careful guidance to solve the problem. Similarly, when conducting conversation training in a foreign language, there are many cases in which it is difficult to know what to speak first.

  Therefore, the present embodiment supports the idea for writing a composition by presenting hints and comments to the user's composition. Similarly, in the present embodiment, in the foreign language conversation, the conversation is assisted by giving a hint for the next speech. Thereby, the user can get used to the foreign language conversation, and as a result, it becomes easy to practice the conversation with the person. The idea support device according to the present embodiment supports the user's idea by giving the user a hint for the next statement or giving a comment in the composition or conversation, and thus supports the user's writing or conversation.

  FIG. 1 is a block diagram illustrating a configuration example of the idea support apparatus according to the present embodiment. The idea support apparatus of the present embodiment is configured by a computer including, for example, an input device 202, a display device 203, a communication device 204, a computing device (CPU) 205, a memory 206, and an auxiliary storage device 207.

  The input device 202 accepts input of commands and the like from the user, and is, for example, a keyboard and a mouse. The input device 202 receives an input of a command or the like executed for control of a program executed by the arithmetic unit (CPU) 205 and control of a device connected to the idea support device 201. For example, according to a predetermined protocol, the communication device 204 transmits the processing content of the idea support device 201 to an external device or receives information from the external device.

  An arithmetic unit (CPU) 205 includes a processor and executes a program stored in the memory 206. The memory 206 includes a ROM that is a nonvolatile storage element and a RAM that is a volatile storage element. The ROM stores an immutable program (for example, BIOS). The RAM is a high-speed and volatile storage element such as a DRAM (Dynamic Random Access Memory), and temporarily stores a program executed by the processor 101 and data used when the program is executed.

  The auxiliary storage device 207 is a large-capacity non-volatile storage device such as a magnetic storage device (HDD) or a flash memory (SSD), for example, and is used when executing a program executed by the arithmetic device (CPU) 205 and the program. Data to be stored. That is, the program is read from the auxiliary storage device 207, loaded into the memory 206, and executed by the arithmetic device (CPU) 205.

  A program executed by the arithmetic unit (CPU) 205 is provided to the idea support device 201 via a removable medium (CD-ROM, flash memory, etc.) or a network, and is a non-volatile auxiliary storage device 207 that is a non-temporary storage medium. Stored in For this reason, the idea support apparatus 201 may have an interface for reading data from a removable medium.

  The idea support device 201 is a computer system that is configured on a single computer or a plurality of computers that are logically or physically configured, and operates on separate threads on the same computer. It may be possible to operate on a virtual machine constructed on a plurality of physical computer resources.

  The arithmetic unit (CPU) 205 includes a knowledge graph generation unit 211, a context calculation unit 212, an utterance understanding unit 213, a context recognition unit 214, and a phase recognition unit 215. For example, the arithmetic unit (CPU) 205 operates as the knowledge graph generation unit 211 by operating according to the knowledge graph generation program loaded in the memory 206, and operates according to the context calculation program loaded into the memory 206. It functions as the context calculator 212. The same applies to other units included in the arithmetic unit (CPU) 205.

  The knowledge graph generation unit 211 generates a graph (knowledge graph) representing knowledge indicated by a document in the document DB 103 described later. The context calculation unit 212 calculates the appearance order of the knowledge indicated by the knowledge graph held in the knowledge graph DB 105 described later in the document in the document DB 103. The utterance understanding unit 213 receives input of a composition or conversation from the user and creates text data of the composition or conversation.

  The context recognition unit 214 calculates the appearance order of knowledge in the text data created by the utterance understanding unit 213, and compares the appearance order with the appearance order of knowledge calculated by the context calculation unit 212. The phase recognizing unit 215 recognizes the phase of the user's composition or conversation by determining whether or not the predetermined content required for the composition or conversation is insufficient in the text input by the user.

  The auxiliary storage device 207 holds a document DB 103, a knowledge graph DB 105, and a context DB 107. The document DB 103 holds a document written in a natural language, particularly a statement sentence. Since the document held in the document DB 103 is a document for defining knowledge in the idea support apparatus 201, that is, a document serving as a knowledge source, the document DB 103 describes matters to be discussed and contents related to the discussion. It is desirable to hold a document that has been

  Web documents, news articles, in-house reports, academic papers, official publications of public offices and international organizations, and survey documents such as think tanks are examples of documents held in the document DB 103. The past utterance history by the user is an example of a document held in the document DB 103. The document DB 103 may further hold bibliographic information and the like of each document.

  The knowledge graph DB 105 stores information indicating the structure of the knowledge graph generated by the knowledge graph generation unit 211. The context DB 107 stores information indicating the appearance order of knowledge calculated by the context calculation unit 212.

  In the present embodiment, the information used by the idea support apparatus 201 may be expressed in any data structure without depending on the data structure. For example, a data structure appropriately selected from a table, list, database or queue can store the information. In the present embodiment, an example is shown in which each data held in the auxiliary storage device 207 is expressed in a table structure.

  The idea support device 201 may include at least one of the input device 202, the display device 203, and the communication device 204. When the idea support device 201 does not include the input device 202, for example, the communication device 204 receives an input of a command or the like from an external device. When the idea support device 201 does not include the display device 203, for example, the communication device 204 transmits a processing result generated by the idea support device 201 to an external device.

  Each processing unit may execute input / output to other processing units via the memory 206 or the auxiliary storage device 207. For example, when the knowledge graph generation unit 211 outputs the processing result to the context calculation unit 212, the knowledge graph generation unit 211 actually stores the processing result in the memory 206 or the auxiliary storage device 207, and the context calculation unit 212 The output result stored in the memory 206 or the auxiliary storage device 207 may be acquired as an input.

  FIG. 2 is a flowchart illustrating an example of processing performed by the idea support apparatus 201 according to this embodiment. The process executed by the idea support apparatus 201 is roughly divided into two parts, that is, a learning process (S110) and an operation process (S120). The learning process includes the processes of steps S111 to S112, and the operation phase includes the processes of steps S121 to S125. First, the learning process (S110) will be described. First, the knowledge graph generation unit 211 generates a knowledge graph to be stored in the knowledge graph DB 105 by converting the information of the document stored in the document DB 103 into graph-type knowledge (S111). Details of step S111 will be described below.

  3A to 3C are examples of knowledge graphs showing knowledge in a document. In this embodiment, an example of a document written in Japanese is used, but the same processing is performed even in another language such as English. Hereinafter, nodes indicated by single-line ellipses in the graphs in the figure are nodes indicating entities such as persons, things, organizations, or entities representing abstract concepts. Moreover, the character described along the arrow between entities represents the relationship between entities.

  The knowledge graph 301 is a knowledge graph generated by the knowledge graph generation unit 211 from a sentence “Experts adopt discipline at school increases discipline” in the document DB 103.

  In generating the knowledge graph, the knowledge graph generating unit 211 first extracts an entity from the sentence. The knowledge graph generation unit 211 extracts an entity using, for example, machine learning. That is, a learning document set in which an entity to be extracted is labeled is prepared in advance, and the knowledge graph generation unit 211 creates a classifier for extracting the entity, and uses the created classifier to Extract entities. The entities in the sentence of FIG. 3A are “expert”, “uniform”, and “discipline”.

  The knowledge graph generation unit 211 may extract an entity with reference to a word dictionary or a word ontology prepared in advance. In addition, since most of the entities are represented by phrases corresponding to nouns and adjectives, the knowledge graph generation unit 211 may perform syntax analysis and extract the nouns and adjectives as entities.

  Next, the knowledge graph generation unit 211 calculates a relationship between entities. In the sentence, “Professionals use discipline at school to raise discipline.” “Uniform” and “discipline” are predicate arguments for “enhance” (“uniform” is the subject, “discipline” is the subject. Object). For example, the knowledge graph generation unit 211 extracts a predicate having two entities as arguments as a relation between the entities by syntax analysis.

  The knowledge graph generation unit 211 generates the knowledge graph 301 by connecting the two entities and the relationship between them with an edge. In generating the knowledge graph, the knowledge graph generation unit 211 generates an edge starting from a node indicating the entity that is the subject and ending with a node indicating the entity that is the predicate. Note that the knowledge graph may be an undirected graph.

  Here, an example is shown in which the knowledge graph generation unit 211 extracts a predicate having an entity (subject and object) as arguments, but other relations between entities may be extracted. For example, the knowledge graph generation unit 211 may extract a dependency between entities as a relation, or may extract a distance in a sentence between entities as a relation. In addition, the knowledge graph generation unit 211 may extract relations using a technique such as Semantic Role Labeling, for example.

  The knowledge graph 302 is a knowledge graph generated by the knowledge graph generation unit 211 from the sentence “School uniforms impede freedom” in the document DB 103. The knowledge graph generation unit 211 extracts the two entities “uniform” and “freedom” and the relationship “inhibit” between the entities from the sentence in the same manner as in the example of FIG. To do.

  As described above, the knowledge graph generation unit 211 extracts an entity from each sentence in a given document, and describes a relation between the entities in the sentence using the knowledge graph.

  FIG. 3C is an example of a knowledge graph in which two knowledge graphs are integrated. For example, when the same node exists in a plurality of knowledge graphs, the knowledge graph generation unit 211 may integrate the plurality of knowledge graphs by equating the nodes. The knowledge graph 303 is a knowledge graph generated when the knowledge graph generation unit 211 identifies the “uniform” node in the knowledge graph 301 and the “uniform” node in the knowledge graph 302.

  For example, the knowledge graph generation unit 211 identifies nodes having the same surface layer representation. In addition, when a node is generated using a word dictionary, an ontology, or the like, the knowledge graph generation unit 211 may treat the same expression as the same node on the word dictionary or the ontology, for example. Specifically, for example, the knowledge graph generation unit 211 may identify synonymous nodes. Further, the knowledge graph generation unit 211 may identify the edges representing the same relation among the edges connecting two nodes on the graph.

  The knowledge graph generation unit 211 represents the information of the generated knowledge graph by, for example, tables shown in FIGS. 3D and 3E, and stores the table in the knowledge graph DB 105. FIG. 3D is an example of a node table stored in the knowledge graph DB 105. The node table 304 is a table that holds node information of the knowledge graph. The node table 304 in FIG. 3D holds information on nodes in the knowledge graph 303.

  The knowledge graph generation unit 211 assigns an ID for identifying the node on the knowledge graph to the node of the generated knowledge graph, and stores the assigned ID in the node table 304. In addition, the knowledge graph generation unit 211 may include other information regarding the node in the node table 304. In the example of FIG. 3D, the node table 304 includes information on the surface layer representation of each node. Further, for example, the node table 304 may include bibliographic information of a document from which a surface layer expression corresponding to each node is extracted.

  FIG. 3E is an example of an edge table stored in the knowledge graph DB 105. The edge table 305 is a table including edge information of the knowledge graph. In the example of FIG. 3E, the edge table 305 holds edge information of the knowledge graph 303.

  The knowledge graph generation unit 211 assigns an ID for identifying an edge on the knowledge graph to the edge of the generated knowledge graph, and associates the ID of each edge, the ID of the start node, and the ID of the end node. In addition, the knowledge graph generation unit 211 may include other information related to the edge in the edge table 305. In the example of FIG. 3E, the edge table 305 includes information on the surface layer representation of each edge. In addition, for example, the edge table 305 may include bibliographic information of a document from which a surface layer expression corresponding to each edge is extracted. The knowledge table structure is defined by the node table 304 and the edge table 305.

  Hereinafter, another example of the knowledge graph generation process will be described. 4A to 4C are examples of knowledge graphs. The sentence of FIG. 4A is the same as the sentence of FIG. 3A. The knowledge graph 301 expresses information indicating that “the uniform increases discipline” indicated by the sentence. However, the knowledge graph 301 does not represent information indicating that “the uniform increases discipline” or “the expert has stated” (that is, information regarding the relation “described”) indicated by the sentence. .

  For example, the knowledge graph generation unit 211 can generate the knowledge graph 402 in addition to the knowledge graph 301 if the statement “Uniform increases discipline” is regarded as an entity. The statement is a proposition indicating a plurality of elements constituting the sentence and a relation between the plurality of elements. Each element may be an entity, a clause, or a sentence, for example. Each element may be a statement.

  Since the knowledge graph 301 and the knowledge graph 402 are independent knowledge graphs, the relationship between the two knowledge graphs indicated by the sentence in FIG. 4A cannot be expressed. In addition, when each of a plurality of related knowledge is expressed by such an independent knowledge graph, the number of entities becomes enormous.

  Therefore, the knowledge graph generation unit 211 generates a knowledge graph 403 shown in FIG. 4B in order to express such information. An example of the generation process of the knowledge graph 403 will be described. As described above, when at least one of the objects connected by the relation in the sentence of the document is a statement, the knowledge graph generation unit 211 generates the knowledge graph by regarding the statement as an entity as in the knowledge graph 402. .

  When the knowledge graph generation unit 211 determines that there is a node of the knowledge graph that includes all the statement information indicated by other knowledge graphs generated from the same document (that is, has a nested structure) If there is a node), a node representing the statement (hereinafter also referred to as a representative node) is generated from the other knowledge graph.

  In the example of FIG. 4A, since the “uniforms increase discipline” node of the knowledge graph 402 includes all the statement information indicated by the knowledge graph 301 generated from the same document as the knowledge graph 402, the knowledge graph generation unit 211 Generate a representative node. For example, the knowledge graph generation unit 211 changes the “enhance” edge indicating the relation of the knowledge graph 301 to the “enhance” node and determines it as a representative node. Then, the knowledge graph generation unit 211 starts from the “uniform” node that is the starting point of the “enhancement” edge, starts from the representative node, and ends from the representative node as the end point of the “enhancement” edge. Create an edge with the "Discipline" node as its end point. Thereby, a portion surrounded by a rectangular dotted line of the knowledge graph 403 is generated.

  As described above, the “enhance” node indicated by a double line in the knowledge graph 403 is a representative node representing one statement including three nodes (“uniform”, “enhance”, and “discipline”). . A representative node representing N statements (N is an integer equal to or greater than 1) is referred to as an (N + 1) th order node. That is, the “enhance” node indicated by the double line in the knowledge graph 403 is a secondary node. A node indicating the entity itself is called a primary node. That is, when a relation between Nth-order nodes is converted into a representative node, the representative node is an N + 1-order node.

  In addition, the edge indicated by a double line in the knowledge graph 403 does not represent a relation between nodes, but a statement composed of three nodes included in a part surrounded by a rectangular dotted line in the knowledge graph 403. This is an edge for representing that the representative node represents. Hereinafter, such an edge is called a statement edge.

  Further, as described above, in the knowledge graph 402, the “expert” node that is the starting point and the node that “the uniform raises discipline” that is the end point are connected by the edge indicating the relation “described”. Therefore, the knowledge graph generation unit 211 indicates a relation that “describes” two nodes starting from the “expert” node, which is representative of the statement “uniforms increase discipline”. A knowledge graph 403 is generated by connecting with an edge.

  As a result, the knowledge graph generation unit 211 can express, in one knowledge graph 403, that “uniforms increase discipline” and that “experts” state that “uniforms increase discipline”. Even after the graph integration processing as described above is performed, if a knowledge graph including the statement itself as an entity remains, the knowledge graph generation unit 211 may delete the knowledge graph.

  FIG. 4C is an example of the node table 304. The node table 304 may further hold the order of nodes. That is, the knowledge graph generation unit 211 stores the node ID and the degree of the node in the node table 304 in association with each other.

  FIG. 4D is an example of the edge table 305. The edge table 305 may hold a flag indicating whether or not the edge is a statement edge. That is, the knowledge graph generation unit 211 stores the edge ID and the flag indicating whether the edge is a statement edge in the edge table 305 in association with each other. Since the statement edge does not indicate the surface layer expression, in the edge table 305, a null value is stored in the surface layer expression corresponding to the statement edge.

  In this way, the knowledge graph generation unit 211 creates not only a relationship between simple entities but also more complicated information by creating a knowledge graph having a graph structure including a secondary node indicating a statement. Can be expressed as

  5A to 5C are examples of knowledge graphs. The knowledge graph 501 is a knowledge graph created from the sentence “Uniforms increase discipline but impede freedom”. As described with reference to FIG. 3C, the knowledge graph generation unit 211 first generates a knowledge graph 303 indicating that “a uniform increases discipline” and “a uniform impedes freedom” from the sentence. create.

  When the knowledge graph generation unit 211 determines that there is a surface expression indicating a paradox in the sentence by referring to, for example, a dictionary prepared in advance, the statement “Uniform increases discipline” that is the main clause, and subordinates A representative node is generated from each of the clauses “Uniforms impede freedom”. For example, the knowledge graph generation unit 211 generates the representative node “enhance” from the main clause and the representative node “inhibit” from the subordinate clause in the same manner as described above. These representative nodes are all secondary nodes.

  Then, the knowledge graph generation unit 211 starts from the “uniform” node that is the starting point of the “enhancement” edge and starts from the representative node as the end point, and the “discipline” that is the end point of the “enhancement” edge from the representative node. And an edge with the node as an end point. Similarly, the knowledge graph generation unit 211 starts with the “uniform” node that was the starting point of the “inhibition” edge, starts with the representative node as the end point, and starts with the representative node as the starting point, with the end point of the “inhibiting” edge. An edge having the “free” node as an end point is generated. All of these generated edges are statement edges.

  Furthermore, the knowledge graph generation unit 211 generates a “paradox” edge indicating a paradoxical relation starting from the representative node “enhance” of the main clause and starting from the representative node “inhibit” of the subordinate clause. Through this process, the knowledge graph generation unit 211 can generate the knowledge graph 501.

  Hereinafter, an example in which the knowledge graph generation unit 211 generates the knowledge graph 503 illustrated in FIG. 5C from the sentence “Uniform increases discipline according to the survey” illustrated in FIG. 5B will be described. The knowledge graph generation unit 211 first generates a knowledge graph 301 as in the example of FIG. 4A. Further, the knowledge graph generation unit 211 generates a knowledge graph 502. An example of a method for generating the knowledge graph 502 will be described.

  For example, when the knowledge graph generation unit 211 refers to a dictionary prepared in advance and determines that a word indicating an information source (“investigation” in FIG. 5B) exists in the sentence, the knowledge graph generation unit 211 performs parsing and the like, Information indicated by the information source (“Uniform increases discipline” in FIG. 5B) is extracted.

  The knowledge graph generation unit 211 regards a statement that “uniforms increase discipline” as an entity. The knowledge graph generation unit 211 generates an “information source” edge as a relation between the two entities, starting from the “investigation” node and ending with the “uniform uniform increases discipline” node. Thereby, the knowledge graph generation unit 211 can generate the knowledge graph 502 shown in FIG. 5C.

  Further, the knowledge graph generation unit 211 can generate the knowledge graph 503 by integrating the knowledge graph 301 and the knowledge graph 502 by a method similar to the method described with reference to FIGS. 4A and 4B.

  Hereinafter, an example in which the knowledge graph generation unit 211 generates the knowledge graph 505 illustrated in FIG. 5E from the sentence “The expert stated that the uniform increases discipline but inhibits freedom” illustrated in FIG. 5D. Will be explained. The knowledge graph generation unit 211 generates the knowledge graph 501 by the method described with reference to FIG. 5A from the portion “the uniform increases discipline but inhibits freedom” of the sentence. In addition, the knowledge graph generation unit 211 generates a knowledge graph 504 by regarding a statement that “uniforms increase discipline but inhibit freedom” as an entity.

  Since the “Uniform increases discipline but inhibits freedom” node of the knowledge graph 504 includes all the statement information indicated by the knowledge graph 501 generated from the same sentence, the “paradox” indicating the relation of the knowledge graph 501 An edge is converted into a “paradox” node and determined as a representative node. The degree of the representative node “paradox” is 3.

  Then, the knowledge graph generation unit 211 starts the “upgrade” node that is the starting point of the “paradox” edge, the statement edge whose end point is the “paradox” node, and the “paradox” edge from the “paradox” node. A statement edge whose end point is the “inhibit” node that was the end point of is generated.

  Furthermore, the knowledge graph generation unit 211 uses the “paradox” node representing the statement “Uniforms increase discipline but inhibits freedom” as the end point, and the “expert” node as the start point, The knowledge graph 505 is generated by connecting with an edge indicating the relation "."

  As described above, the knowledge graph generation unit 211 according to the present exemplary embodiment expresses nested information with graph-type knowledge by using a high-order node and a special edge (statement edge) connecting the high-order node. As a result, the knowledge graph generation unit 211 can represent not only simple information but also a complex context with a small number of knowledge graphs. That is, the idea support apparatus 201 according to the present embodiment suppresses the data amount of the knowledge graph DB 105 even though it can handle a complex context, and further performs a knowledge graph matching process in the utterance understanding process described later at high speed. be able to.

  Returning to the description of FIG. Subsequently, the context calculation unit 212 uses the knowledge graph held by the knowledge graph DB 105 to execute a context calculation process for calculating the order in which information appears in the sentence (S112).

  As described above, the knowledge graph generation unit 211 graphs the individual information extracted from the sentence, and identifies the nodes according to predetermined criteria (surface expression, word dictionary, ontology positioning, etc.), for example. By doing so, the knowledge graph shown in FIG. 3C, FIG. 4B, FIG. 5A, FIG. 5C, FIG.

  FIG. 6A is an example of a knowledge graph. A sentence 601 is an example of a statement that opposes the introduction of school uniforms. Sentence 601 expresses the opinion that it is against the introduction of uniforms, and then describes the merits of increasing discipline, while also describing the disadvantages of hindering students' freedom. Furthermore, in the sentence 601, an opinion from the viewpoint of hindering the fairness of the market by adopting a specific uniform is described.

  As described above, there is a flow of information presentation (context) in composition, conversation, and statement. The context calculation unit 212 of this embodiment calculates the context of a sentence in the document DB 103 and stores the calculation result in the context DB 107. The context calculation unit 212 performs context calculation by calculating the order of knowledge appearing in the sentence.

  The idea support apparatus 201 according to the present embodiment recognizes the context of the composition or conversation by the user using the knowledge graph DB 105 and the context DB 107 and presents the user with appropriate hints, comments, questions, and the like considering the context. . Thereby, the idea support apparatus 201 can deepen a user's thought and can support an idea.

  Details of the context calculation process (S112) will be described below. Note that the knowledge in the context calculation process is knowledge indicated by a relation included in a statement, that is, knowledge indicated by an edge that is not a secondary or higher order node or a statement edge. A knowledge graph 602 in FIG. 6A is an example of a knowledge graph generated from the sentence 601. FIG. 6B is an example of the node table 304 corresponding to the knowledge graph 602. FIG. 6C is an example of the edge table 305 corresponding to the knowledge graph 602.

  For example, for each document included in the document DB 103, the context calculation unit 212 calculates the order of knowledge appearing in the document on the knowledge graph. For example, for the knowledge “Uniform increases discipline” in the knowledge graph 602, the context calculation unit 212 arranges the IDs of nodes and edges related to the knowledge in the order of appearance in the sentence (N1, E1, N2 , E2, N3), the knowledge may be expressed. According to this method, for example, the knowledge that “uniforms increase discipline but inhibit freedom” is, for example, (N1, E1, N2, E2, N3, E3, N4, E5, N1, E6, N6, E7, N7).

  Note that the context calculation unit 212 preferably expresses knowledge more simply by using the method described below. Specifically, the context calculation unit 212 expresses knowledge by specifying an edge that is not a statement edge and a secondary or higher order node. As a result, the context calculation unit 212 can concisely handle complex knowledge including complex sentences, heavy sentences, conjunctions, and the like, such as “Uniforms increase discipline but inhibit freedom” in the knowledge graph. . Hereinafter, a specific example will be described.

  First, the knowledge graph 602 includes secondary nodes N2 and N6 and a tertiary node N4 as secondary and higher nodes. The secondary node N2 “enhance” is connected to the node N1 “uniform” and node N3 “discipline” at the statement edge, and represents the knowledge that “the uniform enhances the discipline”. Similarly, the secondary node N6 “inhibits” represents the knowledge that “uniforms inhibit freedom”. The third-order node N4 “paradox” is connected by a statement edge between N2 representing “Uniforms enhance discipline” and N6 representing “Uniforms impedes freedom”, so “Uniforms enhance discipline” Expresses the knowledge that "does disturb freedom."

  Next, the knowledge graph 602 includes E4 and E8 as edges that are not statement edges. The edge E4 connects the node N5 “expert” and the node N4 “paradox”, and represents the knowledge that “the expert stated that the uniform increases discipline but inhibits freedom”. Similarly, the edge E8 represents knowledge that “a uniform impairs market fairness”.

  By using the above method, the context calculation unit 212 uses, for example, an ID of an edge that is not a statement edge, such as N4 to express complicated knowledge such as “uniforms increase discipline but inhibit freedom”, or Knowledge can be expressed concisely using the IDs of the secondary and higher nodes.

  The context calculation unit 212 calculates the order in which knowledge appears in a document in context calculation processing. An example of the calculation method will be described with reference to FIGS. 7A and 7B. FIG. 7A is an example of a knowledge graph. The knowledge graph 703 is a generated knowledge graph from the sentences 701 and 702 included in the document DB 103, and when knowledge is expressed using an edge that is not a statement edge and a secondary or higher node as described above, In the sentence 701, knowledge appears in the order of N6, E8, N2, and E9. Similarly, knowledge appears in the sentence 702 in the order of N6 and E9.

  For each piece of knowledge that appears in the knowledge graph 703, the context calculation unit 212 calculates the number of appearances of knowledge that appears after the knowledge (hereinafter also referred to as subsequent knowledge) in the document in the document DB 103, and the context DB 107 Store in the subsequent knowledge table.

  That is, the context calculation unit 212 specifies knowledge corresponding to secondary or higher order nodes appearing in the knowledge graph 703 and edges that are not statement edges. For each identified knowledge, the context calculation unit 212 is knowledge that appears after the knowledge in the document in the document DB 103 and that is indicated by an edge that is not a secondary node or statement edge that appears in the knowledge graph 703. The number of appearances of is calculated.

  FIG. 7B is an example of a subsequent knowledge table. The subsequent knowledge table 704 holds the node ID or edge ID corresponding to the knowledge, the subsequent knowledge of the knowledge, and the number of times the subsequent knowledge appears after the knowledge in association with each other. For example, the second line of the subsequent knowledge table 704 in FIG. 7B indicates that knowledge E8 appears once, knowledge N2 once, and knowledge E9 appears twice after knowledge N6. Since the knowledge E9 appears once after N6 in each of the sentences 701 and 702, the number of appearances of E9 as the subsequent knowledge of N6 is two in the subsequent knowledge table 704 of FIG. 7B. ing.

  In this way, the context calculation unit 212 associates the subsequent knowledge with the number of appearances of the subsequent knowledge for each knowledge and stores them in the subsequent knowledge table 704, so that the context DB 107 can determine the order of the knowledge in the document and the context. Etc. can be stored.

  Note that the context calculation unit 212 may calculate the distance of each knowledge with the subsequent knowledge on the document and store the distance in the subsequent knowledge table 704. For subsequent knowledge in which the number of appearances is two or more, the context calculation unit 212 stores the average value or minimum value of the distance between the knowledge and the subsequent knowledge on the document in the subsequent knowledge table 704. The learning process has been described above.

  Next, the operation process (S120) will be described. In the operation process, the idea support apparatus 201 uses the knowledge graph and subsequent knowledge created by the learning process to understand the context in the composition and conversation by the user, and presents an appropriate hint in the middle of the composition and conversation. This encourages the user to think.

  First, the utterance understanding unit 213 receives an input of a composition or conversation including a statement from the user, and converts it into a text (S121). Specifically, the utterance understanding unit 213 may accept text input from an input interface such as a keyboard, for example, or may convert uttered speech input from a microphone or the like into text using a speech recognition technique. For example, the utterance understanding unit 213 may convert text written on paper by OCR or the like. Further, the utterance understanding unit 213 may convert the user's movement into text using a technique such as sign language recognition.

  Subsequently, the utterance understanding unit 213 performs an utterance understanding process for understanding the text acquired in step S121 using the knowledge graph 105 (S122). Here, the utterance understanding unit 213 understands the user's utterance by matching the input text with the knowledge graph 105. Details of the speech understanding process will be described below with reference to FIGS. 8A, 8B, 9A, 9B, and 9C.

  A knowledge graph 801 in FIG. 8A is a part of the knowledge graph indicated by the node table 304 and the edge table 305 stored in the knowledge graph DB 105. 8B is a part of the subsequent knowledge table 704 stored in the context DB 107.

  For example, suppose that the user has created a statement in the position of “in favor of introducing a uniform into school” with respect to the theme “whether a uniform should be introduced into school”. Then, it is assumed that the utterance understanding unit 213 receives an input of a sentence “School uniform should be introduced” on the display screen 903 illustrated in FIG. 9C. The utterance understanding unit 213 generates the knowledge graph 901 in FIG. 9A from the sentence using the same method as the knowledge graph generation processing in step S111.

  Subsequently, the utterance understanding unit 213 collates the knowledge graph 901 and the knowledge graph 801 and searches the knowledge graph 801 for knowledge included in the knowledge indicated by the knowledge graph 901. Using the knowledge representation method described in the context calculation process, the knowledge E10 indicated by the knowledge graph 801 (that is, knowledge obtained by connecting the “school” node and the “uniform” node by the “introducing” relation) is the knowledge graph 901. Therefore, the utterance understanding unit 213 extracts E10. As described above, the utterance understanding unit 213 extracts the knowledge included in the user's input and understands the user's input by collating the extracted knowledge with the knowledge graph learned in advance.

  Further, the utterance understanding unit 213 uses, for example, the same expression method as the context calculation process, and the node ID and edge ID on the knowledge graph 801 of the knowledge indicated by the text input by the user, and the knowledge appearance order in the text Are output to the context recognition unit 214. When the sentence shown on the display screen 903 is input, the utterance understanding unit 213 outputs E10 to the context recognition unit 214, for example. When the sentence 702 in FIG. 7A is input, the utterance understanding unit 213 outputs the edge IDs to the context recognition unit 214 in the order of N6 and E9, for example.

  By the way, the knowledge graph 901 completely coincides with a part of the knowledge graph 801 (the portion consisting of N10, E10, and N1). Even if one relation or entity does not completely match the other relation or entity in the matching of the knowledge graph, for example, if the similarity is equal to or higher than a predetermined value, the utterance understanding unit 213 You may treat the other as the same thing.

  For example, it is assumed that the utterance understanding unit 213 generates the knowledge graph 902 of FIG. 9B from text input from the user. The edge of the knowledge graph 902 is “take in”, but the utterance understanding unit 213 assumes that “take in” and E10 “introduction” of the knowledge graph 801 are synonymous, and the knowledge indicated by the knowledge graph 902 matches E10. You may judge.

  The utterance understanding unit 213 calculates the similarity between entities and relations based on, for example, a synonym relationship in a word dictionary or ontology. The utterance understanding unit 213 can increase the knowledge in the knowledge graph DB 105 that matches the knowledge extracted from the user input by treating relations and entities having a similarity equal to or higher than a predetermined value as synonyms. The possibility of being able to give appropriate tips is increased.

  Returning to the description of FIG. Subsequently, the context recognition unit 214 executes a context recognition process for collating the appearance order of knowledge indicated by the user input text acquired in the utterance understanding process with the subsequent knowledge table 704 (S123). For example, when a text corresponding to the knowledge graph 901 is input from the user, E10 is output in step S122 as described above. At this time, the context recognizing unit 214 outputs the information indicated by the second line of the subsequent knowledge table 704 in FIG. 8B (the line whose knowledge value is E10) to the phase recognizing unit 215.

  Next, the phase recognition unit 215 executes phase recognition processing for recognizing the user's composition or conversation phase (S124). For example, when a user discusses a theme (in the examples of FIG. 9A to FIG. 9C, “Should a school uniform be introduced?”) In a pros or cons, the content required for the statement There is.

  A statement of approval or disapproval, the points of approval or disapproval, the reason for approval or disapproval, and the presentation of reasons, etc. are examples of the content required for the statement. The phase recognizing unit 215 recognizes the phase of the user's composition or conversation by determining whether or not the predetermined content required for the composition or conversation is insufficient in the text input by the user.

  FIG. 9C shows an example of the transition of the screen displayed on the display device 203. For example, the phase of the sentence displayed on the display screen 903 is a phase in which a position in favor of the introduction of a uniform has been expressed, but no points have been raised, and no reason or basis has been presented. The phase recognizing unit 215 specifies contents that are not described in the text input by the user among the predetermined contents.

  Note that the content required for composition and conversation is defined in advance. For example, the idea support apparatus 201 holds a learning document in advance. For example, the idea support apparatus 201 creates a label of a location corresponding to the content before starting the operation process, and creates a discriminator that can recognize the content by machine learning using the label. The phase recognition unit 215 uses this classifier to determine whether or not the content is included in the text input by the user. In the example of FIG. 9C, it is assumed that the composition requires a statement of approval or disapproval for the introduction of the uniform to the school, and an argument for the approval or disapproval expressed.

  For example, the idea support apparatus 201 adds an annotation indicating that the location corresponds to the content of the document DB 103 (for example, a document including the content or a text including the content). May be.

  Then, when the knowledge graph generation unit 211 generates a knowledge graph including knowledge using the annotation part in the knowledge graph generation process, the knowledge graph generation unit 211 indicates that the knowledge indicates the content indicated by the annotation part on the additional information on the knowledge graph. May be given as That is, for example, the knowledge graph generation unit 211 adds a column to the node table 304 and the edge table 305, and stores the additional information in the added column.

  Returning to the description of FIG. Subsequently, the phase recognition unit 215 performs a hint presentation process for presenting a hint to the user based on the results of the context recognition process and the phase recognition process (S125). The phase recognition unit 215 generates a hint to be presented from the subsequent knowledge indicated by the subsequent knowledge table 704.

  As described above, when the sentence displayed on the display screen 903 is input, in the context recognition process, the information of the record whose knowledge in the subsequent knowledge table 704 is E10 is output. That is, the phase recognition unit 215 has already acquired information indicating that the subsequent knowledge of the knowledge E10 is N2 and E9.

  Accordingly, the phase recognizing unit 215 determines the statement indicated by the relationship between the subsequent knowledge N2 and the subsequent knowledge E9 as a hint candidate to be presented to the user. The phase recognition unit 215 may output all the candidates as hints to the display device 203.

  In addition, for example, when there is a candidate corresponding to knowledge indicating content determined to be insufficient in the sentence displayed on the display screen 903 in the phase recognition process, the phase recognition unit 215 prioritizes the candidate. May be displayed on the display device 203 as a hint.

  Note that “a candidate is displayed with priority” means, for example, that only the candidate is displayed on the display device 203, and when the candidate is displayed on the display device 203, the candidate is highlighted. It means being done.

  In addition, the phase recognition unit 215 has a high number of appearances in a record whose knowledge in the subsequent knowledge table 704 is E10 (in the example of FIG. 8B, N2 is 10 times and E9 is 5 times). May be presented as hints (for example, a predetermined number of candidates are presented in descending order of the number of appearances, or all candidates whose number of appearances is a predetermined number or more are presented). Thereby, the phase recognition part 215 can output the knowledge strongly related to the content input by the user as a hint.

  When the subsequent knowledge table 704 stores the distance between the knowledge and the subsequent knowledge on the document, the phase recognition unit 215 has the knowledge that the distance in the record whose knowledge in the subsequent knowledge table 704 is E10 is small. May be presented as a hint on the display device 203 (for example, a predetermined number of candidates are presented in ascending order of distance, or all candidates whose distance is equal to or smaller than a predetermined value are presented). Thereby, the phase recognition part 215 can output the knowledge strongly related to the content input by the user as a hint.

  For example, suppose that this is an argument for or against knowledge N2 and knowledge E9. At this time, for example, the phase recognizing unit 215 gives priority to the number of appearances and presents the content “Uniform enhances discipline” as a hint as displayed in the display screen 904 of FIG. 9C. Note that the phase recognition unit 215 displays the hint on the display device 203, for example, when it is estimated that the user needs a hint, or when a hint presentation request is received from the user. Specifically, for example, the phase recognizing unit 215 may present a hint when there is no next input even after a certain time has elapsed from the user's input, or a request from the user such as a button press by the user. You may give a hint if there is

  Here, although the phase recognition unit 215 presents only one hint on the display screen 904, a plurality of hints may be presented. Note that the phase recognition process is a process for narrowing the hints to be presented, and thus may not be executed.

  The user continues to input text as displayed on the display screen 905 with reference to the hint presented on the display screen 904, and returns to the speech understanding process. Note that the user may input text that is not related to the presented hint. Then, the speech understanding process and the context recognition process are executed again on the input text. Outputs in the context recognition process when the sentence displayed on the display screen 905 is input are E10 and N2.

  At this time, in the subsequent knowledge table 704 in FIG. 8B, the subsequent knowledge common to E10 and N2 is E9. Therefore, the phase recognition unit 215 performs “ “Improve learning motivation” is suggested as a hint.

  The idea support apparatus 201 according to the present embodiment can give an appropriate hint to the user when the user is stuck in text input or the user requests a hint by the above-described processing. It is possible to support the next utterance by the user.

  The phase recognition unit 215 may perform the following process using a knowledge graph in the hint presentation process. As described above, when the user inputs a sentence displayed on the display screen 905, knowledge E10 and knowledge N2 are output in the context recognition process.

  In the knowledge graph 801, there is knowledge N6 that is inversely connected to the knowledge N2 output in the context recognition processing. As described above, when the knowledge of the paradox of the knowledge input from the context recognition unit 214 is included in the knowledge graph and the knowledge of the paradox is not input from the context recognition unit 214, the phase recognition unit 215 A hint corresponding to the knowledge may be presented. Specifically, for example, the phase recognizing unit 215 asks a question such as “What do you think, although there is an opinion that uniforms increase discipline, but uniforms impede freedom”, as shown on display screen 907. Comments may be presented with hints.

  Through this process, the user can know his / her opinions from the opposite standpoint, and as a result, can assist in deepening the user's ideas in discussion, writing, and conversation. In the above-described example, the example in which the phase recognizing unit 215 presents the knowledge of the paradox of knowledge input from the context recognizing unit 214 as a hint has been described. However, the phase recognizing unit 215 is input from the context recognizing unit 214. Knowledge that is in a relation other than the paradox of knowledge and that is not input from the context recognition unit 214 may be presented as a hint.

  In addition, this invention is not limited to an above-described Example, Various modifications are included. For example, the above-described embodiments have been described in detail for easy understanding of the present invention, and are not necessarily limited to those having all the configurations described. Further, a part of the configuration of a certain embodiment can be replaced with the configuration of another embodiment, and the configuration of another embodiment can be added to the configuration of a certain embodiment. Further, it is possible to add, delete, and replace other configurations for a part of the configuration of each embodiment.

  Each of the above-described configurations, functions, processing units, processing means, and the like may be realized by hardware by designing a part or all of them with, for example, an integrated circuit. Each of the above-described configurations, functions, and the like may be realized by software by interpreting and executing a program that realizes each function by the processor. Information such as programs, tables, and files for realizing each function can be stored in a memory, a hard disk, a recording device such as an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.

  Further, the control lines and information lines are those that are considered necessary for the explanation, and not all the control lines and information lines on the product are necessarily shown. Actually, it may be considered that almost all the components are connected to each other.

  103 Document DB, 105 Knowledge Graph DB, 107 Context DB, 201 Idea Support Device, 202 Input Device, 203 Display Device, 204 Communication Device, 205 Computing Device (CPU), 206 Memory, 207 Auxiliary Storage Device, 211 Knowledge Graph Generation Unit 212 context calculation unit, 213 utterance understanding unit, 214 context recognition unit, 215 phase recognition unit

Claims (10)

An idea support device that supports a user's idea,
Including a processor and memory,
The memory is
Graph information indicating a structure of a first graph including a node indicating an element constituting a proposition included in a document defining knowledge and a first type edge indicating a relation between the elements indicated by the proposition;
And context information indicating the order of appearance of relations in the document,
The processor is
Accepts the input of a proposition,
Extracting an element included in the proposition that has received the input and a relation between the elements indicated by the proposition that has received the input;
Generating a second graph including a node indicating the extracted element and including a relation between the extracted elements as the first type edge;
When it is determined that the relation indicated by the second graph is included in the first graph with reference to the graph information, the relation that has appeared in the document after the relation is specified with reference to the context information;
An idea support apparatus that outputs information indicating the specified relation. The idea support device according to claim 1,
The document includes a plurality of propositions;
The nodes of the first graph are
A first type node indicating entities included in the plurality of propositions;
A second kind node indicating a relation of a first kind proposition included in the plurality of propositions, wherein all elements and relations are propositions contained in one element of another proposition in the plurality of propositions; Including,
The first graph includes a second type edge that connects a node indicating an element constituting the first type proposition and the second type node;
The processor is
Accepts multiple proposition inputs,
In generating the second graph,
An entity among the extracted elements is determined as the first type node,
When it is determined that the plurality of propositions that have received the input include the first type proposition, the relation of the first type proposition is determined as the second type node,
An idea support apparatus that connects a node indicating an element constituting the first type proposition and the determined second type node with the second type edge. The idea support device according to claim 1,
The context information indicates, for each relation indicated by the proposition of the document, the number of occurrences of the relation that appears after the relation in the document,
The processor is
Based on the number of occurrences of the specified relation indicated by the context information, the relation is selected from the specified relation;
An idea support device that outputs information indicating the selected relation. The idea support device according to claim 1,
The context information indicates, for each relation indicated by the proposition of the document, a distance on the document with a relation that appears after the relation in the document,
The processor is
Based on the distance between the relation indicated by the second graph included in the first graph indicated by the context information and the specified relation, a relation is selected from the specified relation;
An idea support device that outputs information indicating the selected relation. The idea support device according to claim 1,
The idea support device, wherein the processor outputs information indicating the predetermined relation when the predetermined relation is included in the specified relation. An idea support device is a method for supporting a user's idea,
The idea support device
Graph information indicating a structure of a first graph including a node indicating an element constituting a proposition included in a document defining knowledge and a first type edge indicating a relation between the elements indicated by the proposition;
And context information indicating the order of appearance of relations in the document,
The method
The idea support device
Accepts the input of a proposition,
Extracting an element included in the proposition that has received the input and a relation between the elements indicated by the proposition that has received the input;
Generating a second graph including a node indicating the extracted element and including a relation between the extracted elements as the first type edge;
When it is determined that the relation indicated by the second graph is included in the first graph with reference to the graph information, the relation that has appeared in the document after the relation is specified with reference to the context information;
A method of outputting information indicating the identified relation. The method of claim 6, comprising:
The document includes a plurality of propositions;
The nodes of the first graph are
A first type node indicating entities included in the plurality of propositions;
A second-type node indicating a relation of a first-type proposition that is a proposition included in the plurality of propositions, wherein all elements and relations are propositions included in one element of another proposition in the plurality of propositions; Including,
The first graph includes a second type edge that connects a node indicating an element constituting the first type proposition and the second type node;
The method
The idea support device
Accepts multiple proposition inputs,
In generating the second graph,
An entity among the extracted elements is determined as the first type node,
When it is determined that the plurality of propositions that have received the input include the first type proposition, the relation of the first type proposition is determined as the second type node;
A method of connecting a node indicating an element constituting the first type proposition and the determined second type node by the second type edge. The method of claim 6, comprising:
The context information indicates, for each relation indicated by the proposition of the document, the number of occurrences of the relation that appears after the relation in the document,
The method
The idea support device
Based on the number of occurrences of the specified relation indicated by the context information, a relation is selected from the specified relation;
A method of outputting information indicating the selected relation. The method of claim 6, comprising:
The context information indicates, for each relation indicated by the proposition of the document, a distance on the document with a relation that appears after the relation in the document,
The method
The idea support device
Based on the distance between the relation indicated by the second graph included in the first graph indicated by the context information and the specified relation, a relation is selected from the specified relation;
A method of outputting information indicating the selected relation. The method of claim 6, comprising:
A method in which the idea support device outputs information indicating the predetermined relation when the specified relation includes the predetermined relation.

Images