Information processing device, program, and method for generating a knowledge graph

Abstract

This invention provides an information processing device that generates knowledge graphs that are easy for people to understand and that can lead to the discovery of new knowledge. [Solution] The information processing device 1 comprises a first control unit 151 that outputs a first knowledge graph generated by analyzing text, a second control unit 152 that outputs information indicating an episode associated with an episode node included in the first knowledge graph, and a third control unit 153 that outputs a second knowledge graph generated based on the first knowledge graph, in which the two nodes are connected by an edge, and which includes at least two nodes from among a location node, a person node, a non-person object node, and an event node, and an edge connecting the two nodes. At least two nodes are generated from episode nodes included in the first knowledge graph, and the edge connecting the two nodes is generated from a history edge.

Description

Technical Field

[0001] This disclosure relates to the technology of generating knowledge graphs.

Background Art

[0002] A knowledge graph refers to a network of knowledge that systematically connects various knowledge (=knowledge) and is represented in a graph structure composed of nodes and edges. A knowledge graph is also referred to as a knowledge graph.

[0003] For example, in Patent Document 1, in order to solve the problem that in a conventional knowledge graph, a difficult-to-understand knowledge graph is generated due to insufficient quality control, a knowledge graph support system that supports the understanding of a knowledge graph based on natural language generation technology is disclosed. According to Patent Document 1, it is possible to assist an expert in the relevant field to understand the knowledge graph and perform quality control of the knowledge graph.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, in the knowledge graph support system disclosed in Patent Document 1, it only supports the understanding of the generated knowledge graph by converting the generated knowledge graph into natural language text using natural language generation technology, and the problem of generating a difficult-to-understand knowledge graph has not been solved.

[0006] Furthermore, while generated knowledge graphs can generally show how information (knowledge) relates to and combines with other information (knowledge), there is a problem that as the amount of information (knowledge) increases, the knowledge graph becomes more complex, making it difficult for people to read and use that information (knowledge).

[0007] In one aspect, this disclosure aims to provide an information processing device, etc., that can generate knowledge graphs that are easy for people to view and understand, and that can lead to the discovery of new knowledge. [Means for solving the problem]

[0008] (1) The information processing device of the present disclosure is an information processing device that generates a knowledge graph composed of a plurality of nodes and edges indicating the relationships between each of the plurality of nodes. The information processing device includes a storage unit that stores a first knowledge graph generated by analyzing a document, the first knowledge graph including episode nodes indicating each of a plurality of episodes contained in the document and history edges indicating the history contained in the document; a first control unit that controls the output of the first knowledge graph from the storage unit in a manner in which the episode nodes contained in the first knowledge graph are connected by the history edges; a reception unit that receives instructions from a user; and a second control unit that outputs information indicating the episodes associated with the episode nodes contained in the output first knowledge graph based on the instructions received by the reception unit. The system comprises a control unit and a third control unit that controls the output of a second knowledge graph generated based on the first knowledge graph, in which the two nodes are connected by the edge, and which includes at least two nodes from among a field node which is a node relating to a location, a person node which is a node relating to a person, an object node which is a node relating to an object other than a person, and an event node which is a node relating to an event, and an edge which connects the two nodes, wherein the at least two nodes are generated from the episode node included in the first knowledge graph, and the edge which connects the two nodes is generated from the history edge.

[0009] The "first knowledge graph" is a knowledge graph (episode graph) generated by treating each of the multiple episodes contained in a text as a node (episode node) and each of the events in the sequence of episodes contained in the text as an edge (sequence edge).

[0010] The "second knowledge graph" is a knowledge graph (human-object-word graph) composed of location nodes (location nodes), person nodes (people nodes), non-person object nodes (non-people objects), event nodes (event nodes), and edges connecting two pairs of nodes. The second knowledge graph is generated based on the first knowledge graph.

[0011] • "Text" here refers to any text that contains multiple episodes, such as a history. However, the term "text" here is not limited to historical texts; it could also include, for example, a cooking recipe or the contents of a meeting transcript.

[0012] (2) In such an information processing device, the history edge may include at least one of a physical history edge that shows history over time or in a procedure, and a logical history edge that shows history in terms of causal relationships.

[0013] "Background" refers to the concept of physical or logical background, among the four types of relationships between episodes (identical or similar relationships, relationships based on physical background, relationships based on logical background, and relationships that indicate completely different events).

[0014] (3) The system may further include a generation unit that performs a 5W1H analysis on the text, generates episode nodes and history edges by treating sentences containing the 4W (When, Where, Who, What) elements as episodes and sentences containing the 1W1H (Why, How) elements as history.

[0015] "5W1H analysis" refers to organizing a text using the five W1H elements: When, Where, Who, What, Why, and How. 5W1H analysis yields a text composed of at least one of these five elements.

[0016] (4) The text may also represent a story relating to history or legend, the person node may represent a being that is the subject of the episode, the place node may represent a location related to the episode, the object node may represent an object related to the episode, and the event node may represent an event related to the episode. "Place" refers to a location, such as a historical or legendary spot or area. "Things" refers to items related to an episode, such as historical or legendary cultural artifacts or works of art. "Person" refers to the subject of the episode, such as a human, a monster, an animal, or a legendary being. "Event" refers to an event related to an episode, such as a historical or legendary occurrence or incident.

[0017] (5) Such an information processing device may further include a fourth control unit that, when the receiving unit receives a designation of a predetermined node among the nodes included in the second knowledge graph, outputs an episode description associated with the predetermined node and an episode description showing the relationship between the predetermined node and other nodes connected by the edge.

[0018] (6) The episode also includes time information of the event that occurred, and the fourth control unit may, if multiple episodes are associated with the predetermined node, arrange and display the multiple episodes in chronological order based on the time information.

[0019] (7) The program of the present disclosure is an information processing device that generates a knowledge graph comprising a plurality of nodes and edges indicating the relationships between each of the plurality of nodes, and includes a first control step that controls outputting the first knowledge graph from a storage unit that stores a first knowledge graph generated by analyzing a document, the first knowledge graph comprising episode nodes indicating each of a plurality of episodes contained in the document and history edges indicating the history contained in the document, in a manner in which the episode nodes contained in the first knowledge graph are connected by the history edges; a reception step that receives instructions from a user; and based on the instructions received in the reception step, controls outputting the episode nodes contained in the output first knowledge graph. The method involves causing a computer to perform a second control step of outputting information indicating an episode associated with a node, and a third control step of controlling the output of a second knowledge graph generated based on the first knowledge graph, which includes at least two nodes from among a location node, a person node, a non-person object node, and an event node, and an edge connecting the two nodes, wherein the two nodes are connected by the edge, and the at least two nodes are generated from the episode node included in the first knowledge graph, and the edge connecting the two nodes is generated from the history edge.

[0020] (8) The method for generating a knowledge graph of the present disclosure is a method for generating a knowledge graph composed of a plurality of nodes and edges indicating the relevance between each of the plurality of nodes. This generation method is a first knowledge graph generated by analyzing a text, and includes an episode node indicating each of a plurality of episodes included in the text, and a context edge indicating the context included in the text. A first control step of performing control to output the first knowledge graph in a manner of connecting the episode nodes included in the first knowledge graph with the context edges; a reception step of receiving an instruction from a user; and based on the instruction received in the reception step, a second control step of outputting information indicating an episode associated with the episode node included in the output first knowledge graph; at least two nodes among a place node that is a node related to a location, a person node that is a node related to a person, an object node that is a node related to an object other than a person, and an event node that is a node related to an event; and an edge connecting the two nodes, and a third control step of performing control to output a second knowledge graph generated based on the first knowledge graph in a manner of connecting the two nodes with the edge, wherein the at least two nodes are generated from the episode nodes included in the first knowledge graph, and the edge connecting the two nodes is generated from the context edge.

Advantages of the Invention

[0021] An information processing apparatus of the present disclosure can generate a knowledge graph that is easy for people to view and understand and leads to the discovery of new knowledge.

Brief Description of the Drawings

[0022] [Figure 1] It is a functional block diagram of the information processing apparatus of the present disclosure. [Figure 2] It is a conceptual diagram showing an example of an episode graph. [Figure 3] It is a conceptual diagram showing an example of a human-object-event graph. [Figure 4]It is a functional block diagram of a user terminal of the present disclosure. [Figure 5] It is a schematic diagram showing an example of the hardware configuration of an information processing apparatus. [Figure 6] It is a schematic diagram showing an example of the hardware configuration of a user terminal. [Figure 7] It is a schematic flow showing an example of the operation of an information processing apparatus. [Figure 8] It is a schematic flow showing an example of the operation of generating an episode graph. [Figure 9] It is a conceptual diagram for explaining a method of generating an episode graph. [Figure 10] It is a conceptual diagram for explaining a method of generating an episode graph.

Mode for Carrying Out the Invention

[0023] Hereinafter, an information processing apparatus and the like according to an aspect of the present disclosure will be specifically described with reference to the drawings. Note that each of the embodiments described below shows a specific example of the present disclosure. The numerical values, shapes, components, arrangement positions of the components, etc. shown in the following embodiments are merely examples and are not intended to limit the present disclosure. In addition, among the components in the following embodiments, components not described in the independent claims indicating the highest-level concept are described as optional components. Also, in all embodiments, the respective contents can be combined.

[0024] Also, the following drawings are schematic, and some configurations or the like may be omitted for the sake of explanation. Also, the same reference numerals are given to common parts in one or more embodiments, and the description may be omitted.

[0025] [1. Schematic Explanation]

[0026] Figure 1 is a functional block diagram of the information processing device 1 of this disclosure. Figure 2 is a conceptual diagram showing an example of an episode graph. Figure 3 is a conceptual diagram showing an example of a human-object-word graph. Figure 4 is a functional block diagram of the user terminal 2 of this disclosure. First, I will explain the general layout of the information processing device 1 and the user terminal 2.

[0027] (Information processing device) The information processing device 1 shown in Figure 1 generates a knowledge graph composed of multiple nodes and edges that indicate the relationships between each of the multiple nodes.

[0028] The information processing device 1 generates a knowledge graph called an episode graph based on, for example, multiple historical episodes, and then generates another knowledge graph called a people-things-word graph based on the generated episode graph. The information processing device 1 then uses the generated people-things-word graph to represent multiple historical episodes. The information processing device 1 can also transmit a portion of the generated people-things-word graph to the user terminal 2.

[0029] (episode) An episode can be a story that specifically describes an event, or it can be a more coherent story, i.e., a short anecdote. The episodes generated by the information processing device 1 are sentences that describe who took what action (or what state occurred), and are not limited to historical events. The episodes generated by the information processing device 1 may also be, for example, a single step in a cooking recipe, or the content of a statement made by a single person in meeting minutes.

[0030] (Episode graph) An episode graph is a knowledge graph composed of nodes (episode nodes) that represent each of the multiple episodes contained in a text, and edges (chronology edges) that represent each of the episodes in the text. An example of an episode graph is shown in Figure 2. Specifically, the episode graph consists of episodes 1 to 7, with episode 1 and episode 2 connected by chronology edge 401, episode 2 and episode 5 connected by chronology edge 402, and episode 2 and episode 3 connected by chronology edge 403. Furthermore, episode 2 and episode 4 are connected by chronology edge 404, and episode 3 and episode 5 are connected by chronology edge 405. Note that chronology edges 401 to 405 are represented as directed line segments.

[0031] (Human-Thing-Word Graph) The Human-Object-Word Graph is a knowledge graph composed of four types of nodes: field nodes (nodes related to locations), person nodes (nodes related to people), object nodes (nodes related to things other than people), and event nodes (nodes related to events), along with edges connecting these nodes. It is generated based on an episode graph. An example of a Human-Object-Word Graph is shown in Figure 3. Specifically, the Human-Object-Word Graph consists of person nodes 501, 502, and 503, field nodes 504 and 505, object nodes 506 and 507, and event node 508. Person node 501 and field node 504 are connected by edge 601. Field node 504 and object node 506 are connected by edge 602, object node 506 and field node 505 are connected by edge 603, field node 505 and person node 502 are connected by edge 604, and person node 502 and person node 503 are connected by edge 605. Person node 502 and event node 508 are connected by edge 606, event node 508 and object node 507 are connected by edge 607, and event node 508 and field node 504 are connected by edge 608. Edges 601 to 608 are represented by undirected line segments.

[0032] (User terminal 2) User terminal 2 displays, for example, a knowledge graph (part of the human-object-word graph) transmitted from information processing device 1. For example, when a node in the knowledge graph displayed on user terminal 2 is specified (selected) by the user of user terminal 2, user terminal 2 displays the episode associated with the specified node. The user of user terminal 2 can read the displayed episode and further read through the episodes of the human-object-word graph by specifying nodes connected by edges to the node corresponding to the displayed episode.

[0033] (Communication network) The information processing device 1 and the user terminal 2 are connected to each other via the Internet or an intranet (communication network). The communication network is, for example, the Internet. Any wired or wireless communication network is acceptable. Examples of wireless communication include Bluetooth®, Wi-Fi®, and LPWA (Low Power Wide Area). For LPWA, LTE® or frequency bands that do not require a license may be used.

[0034] [2. Detailed explanation] (Information Processing Device 1) Information processing device 1 is a computer, such as a personal computer (PC).

[0035] In this embodiment, the information processing device 1, as shown in Figure 1, includes an acquisition unit 11, an analysis unit 12, a generation unit 13, a storage unit 14, a control unit 15, and a reception unit 16. However, the information processing device 1 does not necessarily have to include the analysis unit 12 or the storage unit 14. The storage unit 14 can be located outside the information processing device 1 and can be used by the information processing device 1.

[0036] (Information processing device 1, acquisition unit 11) The acquisition unit 11 acquires texts containing multiple episodes, either at the instruction of the user of the information processing device 1 or automatically. For example, the acquisition unit 11 acquires texts such as historical documents via a communication network such as the Internet. Note that the texts are not limited to historical documents; they can be any texts containing multiple episodes, such as cooking recipes or the contents of statements included in meeting minutes.

[0037] (Information processing device 1, analysis unit 12) The analysis unit 12 analyzes the text acquired by the acquisition unit 11. More specifically, the analysis unit 12 performs a 5W1H analysis on the text acquired by the acquisition unit 11 and summarizes it into one or more sentences composed of 5W1H elements. Here, 5W1H analysis means organizing the text using the 5W1H elements: When, Where, Who, What, Why, and How. Through 5W1H analysis, a sentence is obtained that consists of at least one of the 5W1H elements.

[0038] The analysis unit 12 extracts sentences containing the 4W elements as episodes and sentences containing the 1W1H element as sequences from the aggregated 1 or more sentences. Here, the 1W1H elements are "Why" and "How." Sentences containing the 4W elements, such as "When," "Where," "Who," and "What," reveal who took what action (or what state they were in), and can be considered episodes. Sentences containing the 1W1H element reveal the sequence of events between episodes, allowing the episodes to be connected. In this embodiment, the sequence of events shown in sentences containing the 1W1H element is a sequence of events based on the passage of time or procedure (physical sequence of events), or a sequence of events based on causal relationships (logical sequence of events).

[0039] Furthermore, the relationships between episodes extracted by the analysis unit 12 are as follows: 1) Identical or similar relationship, 2) Relationships that indicate the progression of events over time or through procedures, i.e., relationships of physical progression. 3) A relationship that shows the sequence of events through cause and effect, that is, a relationship of logical sequence of events. 4) These relationships can be classified into four categories: relationships that indicate completely different things, or relationships with different histories. The analysis unit 12 can extract sentences containing the 1W1H elements, thereby extracting either the physical or logical relationship among the four relationships between episodes as the history between episodes.

[0040] The analysis unit 12 may perform a 5W1H analysis on a text by prompting it using an AI such as a generation AI, and then consolidate it into one or more sentences composed of the 5W1H elements. Alternatively, the analysis unit 12 may, by prompting it using an AI such as a generation AI, extract sentences containing four W elements as episodes and sentences containing one W1H element as continuations from the consolidated sentences. In this case, such an AI may be located outside the information processing device 1 and used by the analysis unit 12 via a communication network. Alternatively, such an AI may be located within the information processing device 1 and used by the analysis unit 12.

[0041] (Information processing device 1, generation unit 13) The generation unit 13 generates an episode graph using episode nodes, which are nodes associated with episodes extracted by the analysis unit 12, and history edges, which are edges associated with history extracted by the analysis unit 12. The generation unit 13 stores the generated episode graph in the storage unit 14.

[0042] The generation unit 13 extracts episodes from the generated episode graph that are associated with episode nodes connected by time-travel edges. From the extracted episodes, the generation unit 13 extracts the entity that is the subject of the episode, the location related to the episode, the object related to the episode, or the event related to the episode. The generation unit 13 generates a human-object-word graph using human nodes, which are nodes representing the entity that is the subject of the episode; location nodes, which are nodes representing the location related to the episode; object nodes, which are nodes representing the object related to the episode; and event nodes, which are nodes representing the event related to the episode. The generation unit 13 stores the generated human-object-word graph in the storage unit 14. In this way, the generation unit 13 generates the nodes that constitute the human-object-word graph from the episode nodes included in the episode graph, and generates the edges that constitute the human-object-word graph from the time-travel edges included in the episode graph.

[0043] (Information processing device 1, storage unit 14) The storage unit 14 is composed of an HDD (Hard Disk Drive), an SSD (Solid State Drive), or memory. In this embodiment, for example, the storage unit 14 stores the analysis results from the analysis unit 12, as well as the episode graph and human-object-word graph generated by the generation unit 13.

[0044] (Information processing device 1, control unit 15) As shown in Figure 1, the control unit 15 comprises a first control unit 151, a second control unit 152, a third control unit 153, and a fourth control unit 154.

[0045] The first control unit 151 controls the output of the episode graph from the storage unit 14 in a manner in which the episode nodes included in the episode graph are connected by longitude and meridian edges. Specifically, the first control unit 151 outputs an episode graph as shown in Figure 2, for example. The first control unit 151 may also output and display the episode graph on a display connected to the information processing device 1. In the episode graph, as shown in Figure 2, each episode node is displayed in a manner that includes the episodes associated with each episode node (episodes 1 to 7 in the figure), and longitude and meridian edges are displayed as directed line segments. In this way, the episode graph allows a person to easily read through the episodes. Therefore, the episode graph can serve as a user interface that allows a person to easily read through the episodes.

[0046] The second control unit 152 outputs information indicating the episode associated with the episode node included in the outputted episode graph, based on the instructions received by the reception unit 16. Specifically, the second control unit 152 outputs the content of the episode associated with the episode node instructed by the user of the information processing device 1, as received by the reception unit 16.

[0047] The third control unit 153 controls the output of a human-object-word graph generated based on the episode graph, which includes at least two nodes from among the field node, person node, object node, and event node, and edges connecting the two nodes. The third control unit 153 outputs the human-object-word graph in a manner in which at least two nodes are connected by edges.

[0048] Specifically, the third control unit 153 outputs a human-object-word graph, for example, as shown in Figure 3. The third control unit 153 may also output and display the human-object-word graph on a display connected to the information processing device 1. In the human-object-word graph, as shown in Figure 3, the four types of nodes—person nodes, object nodes, event nodes, and place nodes—are displayed with icons that allow for easy identification of their type, and their edges are displayed as undirected line segments. In addition, episodes associated with the person nodes, object nodes, event nodes, and place nodes are hidden. In this way, the human-object-word graph classifies episodes into four types of nodes and displays the content of each episode, making it easy for people to read through the episodes. Therefore, the human-object-word graph can serve as a user interface that allows people to easily read through episodes.

[0049] The fourth control unit 154 outputs an episode description associated with a specified predetermined node from among the nodes included in the human-object-word graph output by the third control unit 153, and an episode showing the relationship between the predetermined node and other nodes connected by edges. In other words, by specifying any of the person nodes, object nodes, event nodes, or place nodes that make up the human-object-word graph, the episode description (content) associated with the specified node and the content of the episode related to the specified node are displayed.

[0050] Furthermore, if an episode contains time information about when an event occurred, and multiple episodes are associated with a specified node, the fourth control unit 154 may arrange and display the multiple episodes in chronological order or yearly order based on the time information. In other words, if there are multiple episodes associated with a specified node, and those episodes contain time information, the multiple episodes will be displayed in chronological order.

[0051] (Information processing device 1, reception unit 16) The reception unit 16 receives instructions from the user of the information processing device 1 or the user of the user terminal 2. In this embodiment, the reception unit 16 can receive instructions from the user of the information processing device 1. The reception unit 16 can also receive instructions from the user of the user terminal 2 transmitted from the user terminal 2. For example, the reception unit 16 may receive an instruction from the user of the information processing device 1 to specify an episode node from among the episode nodes included in the episode graph output by the first control unit 151. Alternatively, for example, the reception unit 16 may receive an instruction from the user of the information processing device 1 or the user terminal 2 to specify a predetermined node from among the nodes included in the human-object-word graph output by the third control unit 153.

[0052] (User terminal 2) User terminal 2 is a device capable of communicating with information processing device 1, such as a smartphone. Alternatively, a tablet PC, personal computer, or other device may be used as user terminal 2. In this embodiment, the user terminal 2 includes a display unit 21, a terminal acquisition unit 22, and an instruction transmission unit 23, as shown in Figure 2.

[0053] (User terminal 2, display unit 21) The display unit 21 displays information acquired by the user terminal 2 from the information processing device 1. For example, the display unit 21 may be an image display device such as a liquid crystal display (LCD), a plasma display panel (PDP), or an organic electroluminescent (EL) display. In this embodiment, the display unit 21 displays at least a portion of the human-object-word graph acquired by the user terminal 2 from the information processing device 1. Furthermore, the display unit 21 can display an episode description associated with a predetermined node designated by the user of the user terminal 2, as well as an episode showing the relationship between the predetermined node and other nodes connected by edges.

[0054] (User terminal 2, terminal acquisition unit 22) The terminal acquisition unit 22 acquires information transmitted by the information processing device 1. In this embodiment, the terminal acquisition unit 22 acquires a human-object-word graph and the like from the information processing device 1. The terminal acquisition unit 22 can acquire, from among the nodes included in the human-object-word graph displayed on the display unit 21, an episode description associated with a predetermined node designated by the user of the user terminal 2, and an episode showing the relationship between the predetermined node and other nodes connected by edges.

[0055] (User terminal 2, instruction transmission unit 23) The instruction transmission unit 23 transmits user instructions from the user terminal 2 to the information processing device 1. In this embodiment, the instruction transmission unit 23 can transmit an instruction to the information processing device 1 to specify a particular node among the nodes included in the human-object-word graph displayed on the display unit 21.

[0056] [3. Hardware Configuration] Figure 5 is a schematic diagram showing an example of the hardware configuration of the information processing device 1. Figure 6 is a schematic diagram showing an example of the hardware configuration of the user terminal 2.

[0057] Next, the hardware configurations of the information processing device 1 and the user terminal 2 will be described using Figures 5 and 6.

[0058] (Hardware configuration of information processing device 1) The hardware configuration of the information processing device 1 is described below. As shown in Figure 5, the information processing device 1 of this embodiment uses, for example, a computer. The information processing device 1 is equipped with a CPU 170. The CPU 170 is connected to a memory 171, a connection port 173 for connecting / reading storage devices 172, etc., and a communication circuit 174 for communicating with the outside via a network, all via a bus line 175. The memory 171 stores an information processing device program 1713 for processing the information processing device 1. It may also store a browser program 1712 and an OS 1711 (operating system). Furthermore, it may store an episode graph management database (hereinafter referred to as episode graph DB) 1715 and / or a human-object-word graph management database (hereinafter referred to as human-object-word graph DB) 1716. Note that the memory 171 may be a storage unit 14, or it may be something different from the storage unit 14. If memory 171 and storage unit 14 are different, the episode graph DB 1715 and the human-object-word graph DB 1716 are stored in storage unit 14.

[0059] (Hardware configuration of user terminal 2) The hardware configuration of user terminal 2 is almost the same as that of the information processing device 1 described above, so the same parts are denoted by the same reference numerals, and their explanation is omitted. In this embodiment, user terminal 2 uses, for example, a smartphone.

[0060] The computer's memory 171 stores a user terminal program 1717 for processing user terminal 2. It may also store a browser program 1712 and an OS 1711 (operating system).

[0061] In this embodiment, the information processing device program 1713 and the user terminal program 1717 work together using the functions of the OS 1711 and the browser program 1712, respectively. Alternatively, the information processing device program 1713 and the user terminal program 1717 may operate independently without using the browser program 1712 and the OS 1711, respectively.

[0062] In the hardware configuration of the information processing device 1 and user terminal 2 described above, the functions shown in Figures 3 and 4 are implemented, for example, using the CPU 170, the user terminal program 1717, and the information processing device program 1713. However, some or all of these functions may be sequence-controlled using logic circuits such as a microcontroller or a PLC (programmable logic controller).

[0063] (Episode Graph DB1715) Episode graph DB1715 stores the episode graph generated by the information processing device 1. The episode graph does not need to be in the form of a knowledge graph as shown in Figure 2, for example; it is sufficient to store elements that constitute an episode graph, such as multiple episode nodes, multiple timeline edges, and the connection relationships between those timeline edges.

[0064] (Human-Object-Word Graph DB1716) The Human-Object-Language Graph DB1716 stores the Human-Object-Language Graph generated by the Information Processing Device 1. The Human-Object-Language Graph does not need to be in the form of a knowledge graph as shown in Figure 3, for example; it is sufficient to store elements that constitute a Human-Object-Language Graph, such as multiple nodes including field nodes, person nodes, object nodes, and event nodes, multiple edges, and the connection relationships between those edges.

[0065] [4. Operation of Information Processing Device 1] (Operation of Information Processing Device 1) Figure 7 is a schematic flow chart showing an example of the operation of the information processing device 1.

[0066] The operation of the information processing device 1 will be explained below using Figure 7. Figure 7 shows a flowchart illustrating one embodiment of the processing of the information processing device program 1713 used in the information processing device 1. The symbol "S" shown in the figure represents a step.

[0067] (S11) The CPU 170 of the information processing device 1 outputs an episode graph that has been generated by analyzing the text and stored in the memory unit 14. In this embodiment, the information processing device 1 outputs the episode graph stored in the memory unit 14 to a display connected to the information processing device 1 in a manner in which the episode nodes are connected by longitude and longitude edges. For example, an episode graph like the one shown in Figure 2 is displayed on the display.

[0068] (S12) The CPU 170 of the information processing device 1 receives instructions from the user. In this embodiment, the information processing device 1 receives instructions from the user to specify an episode node from among the episode nodes included in the episode graph output in S11.

[0069] (S13) The CPU 170 of the information processing device 1 outputs information indicating the episode associated with the episode node based on the instruction received in S12. In this embodiment, the information processing device 1 outputs the contents of the episode associated with the specified episode node. The information processing device 1 may also output and display an episode graph on a display connected to the information processing device 1.

[0070] (S14) The CPU 170 of the information processing device 1 outputs a human-object-word graph generated based on the episode graph. In this embodiment, the information processing device 1 extracts an episode from the episode graph output in step S11 and extracts the entity that is the main subject of the episode, the location related to the episode, the object related to the episode, or the event related to the episode as nodes. The information processing device 1 outputs a human-object-word graph consisting of a person node representing the entity that is the main subject of the episode, a location node representing the location related to the episode, an object node representing the object related to the episode, and an event node representing the event related to the episode. The information processing device 1 may also output and display the human-object-word graph, for example, as shown in Figure 3, on a display connected to the information processing device 1.

[0071] Figure 8 is a schematic flow chart showing an example of the episode graph generation operation of the information processing device 1. Figures 9 and 10 are conceptual diagrams to explain the method of generating the episode graph. Note that Figure 10 is the same as Figure 2.

[0072] The following describes the generation operation of the episode graph of the information processing device 1 using Figures 8 to 10. Figure 8 shows a flowchart illustrating one embodiment of the processing of the information processing device program 1713 used in the information processing device 1.

[0073] (S21) The CPU 170 of the information processing device 1 retrieves a document containing multiple episodes, either at the instruction of the user of the information processing device 1 or automatically. For example, the information processing device 1 may retrieve a document that tells a story about history or legend.

[0074] (S22) The CPU 170 of the information processing device 1 performs a 5W1H analysis on the text acquired in S21. For example, the information processing device 1 performs a 5W1H analysis on the text acquired in S21 that describes a story about history or legend, and summarizes it into one or more sentences composed of the 5W1H elements. The information processing device 1 may display the summarized one or more sentences as a bulleted list on a display connected to the information processing device 1.

[0075] (S23) The CPU 170 of the information processing device 1 extracts sentences containing the 4W elements as episodes based on the sentences analyzed using the 5W1H method in S22, and generates episode nodes. For example, the information processing device 1 may extract sentences containing the 4W elements, such as where the events took place, who was active, what cultural properties or works of art were involved, and what events or incidents occurred, from one or more sentences aggregated through the 5W1H analysis in S22. The information processing device 1 may then generate episode nodes corresponding to the extracted episodes.

[0076] The information processing device 1 may also display episode nodes 301 to 307, for example, as shown in Figure 9, on a display connected to the information processing device 1. The example shown in Figure 9 shows a case where episodes 1 to 7 are extracted based on the 5W1H analysis of the text, and episode nodes 301 to 307 are generated.

[0077] (S24) The CPU 170 of the information processing device 1 extracts sentences containing 1W1H elements based on the sentences analyzed in S22 using the 5W1H method, and generates a timeline edge. For example, the information processing device 1 may extract sentences containing 1W1H elements that indicate the circumstances (physical or logical) between the episodes extracted in S23, and generate a timeline edge.

[0078] Furthermore, the information processing device 1 generates meridian edges as directed line segments, and multiple meridian edges connecting the same two episode nodes can be combined into a single meridian edge. Also, from the perspective of making the episode graph easier to read, the information processing device 1 does not display autoregressive meridian edges (connecting to only one episode node) on the display.

[0079] Furthermore, the information processing device 1 may display, for example, the timeline edges 401-405 shown in Figure 10 on a display connected to the information processing device 1. In the example shown in Figure 10, timeline edges 401-405, represented by directed line segments, are generated based on the 5W1H analysis of the text, and an episode graph is shown with these edges added to the episode nodes 301-307. In this way, in the episode graph, the episode nodes 301-307, which are associated with episodes 1-4, are linked by the timeline edges 401-405. On the other hand, in the episode graph, the episode nodes 306 and 307, which are associated with episodes 6 and 7, are nodes with unclear timelines and exist as standalone nodes without connected timeline edges. Standalone nodes are not used in the human-object-word graph. Note that in the episode graph, multiple timeline edges are connected to a single episode node, such as episode node 305, which is associated with episode 5; this indicates that there are various alternative theories in history.

[0080] The information processing device 1 generates the human-object-word graph shown in Figure 3 based on the episode graph shown in Figure 10. In this case, the person nodes 501, 502, and 503 represent the subject of the episode, such as a person, a monster, an animal, or a legendary being. The place nodes 504 and 505 represent locations, such as historical or legendary spots or areas. The object nodes 506 and 507 represent objects related to the episode, such as historical or legendary cultural properties or works of art. The event node 508 represents an event related to the episode, such as historical or legendary events or incidents.

[0081] [5. Others] The above embodiments can be used in combination as appropriate. For example, the analysis unit 12 and the generation unit 13 of the information processing device 1 may perform processing using AI provided inside or outside the information processing device 1. Furthermore, the episode graph DB1715 and the human-object-word graph DB1716 may be provided in the storage unit 14. The storage unit 14 of the information processing device 1 may be provided in a server or the like that is connected via communication.

[0082] Furthermore, although the above embodiment was described using the example of the information processing device 1 generating a two-dimensional episode graph and a two-dimensional human-object-word graph, it may also generate a three-dimensional episode graph and a three-dimensional human-object-word graph.

[0083] Furthermore, while the above embodiment described a case where the information processing device 1 generates an episode graph using text such as history, it is not limited to this. The information processing device 1 may also generate an episode graph using sentences that show how content was used in the market as episodes. In such a case, the episodes should be sentences that show who used what content at what event or place. In this way, the episode graph can serve as a tool or interface for organizing various causal relationships.

[0084] [6. Summary] (1, 7, 8) The information processing device 1 is an information processing device that generates a knowledge graph composed of multiple nodes and edges that indicate the relationships between each of the multiple nodes. The information processing device 1 includes a storage unit 14 that stores a first knowledge graph generated by analyzing a document, which includes episode nodes that indicate each of the multiple episodes contained in the document and history edges that indicate the history contained in the document; a first control unit 151 that controls the output of the first knowledge graph from the storage unit 14 in a manner in which the episode nodes contained in the first knowledge graph are connected by history edges; a reception unit 16 that receives instructions from the user; and the reception unit 16 that outputs information indicating the episodes associated with the episode nodes contained in the output first knowledge graph based on the instructions received by the reception unit 16. The system comprises a second control unit 152 and a third control unit 153 that controls the output of a second knowledge graph generated based on a first knowledge graph, in which the two nodes are connected by an edge, and which includes at least two nodes from among a field node which is a node relating to a location, a person node which is a node relating to a person, an object node which is a node relating to an object other than a person, and an event node which is a node relating to an event, and an edge connecting the two nodes, wherein at least two nodes are generated from episode nodes included in the first knowledge graph, and the edge connecting the two nodes is generated from a history edge.

[0085] Generally, people like and understand stories. This is because nerve cells called "mirror neurons" in the human brain process information from stories, causing people to imagine the scenes in the story and creating a sense of being present in those scenes.

[0086] Furthermore, while general knowledge graphs show how information relates to and combines with other information, they become complex as the amount of information increases, making it difficult for people to read and understand the data.

[0087] Information processing device 1 generates a first knowledge graph (episode graph) using nodes that represent episodes (episode nodes) for multiple episodes contained in a text, such as history. Based on the first knowledge graph (episode graph), information processing device 1 generates a second knowledge graph (human-object-word graph) using four types of nodes: person nodes, object nodes, event nodes, and place nodes. In this way, information processing device 1 represents multiple episodes contained in a text using a dual-structured knowledge graph.

[0088] In the first knowledge graph (episode graph), multiple episodes contained in a text are represented as episode nodes, and the sequence of events between episodes is connected by sequence edges. This gives the information (episode nodes) in the first knowledge graph (episode graph) a narrative quality, making it easier for people to read through the information (episode nodes) and understand the information in the first knowledge graph (episode graph). Furthermore, by reading through the information organized in the first knowledge graph (episode graph), people can discover new knowledge.

[0089] In the second knowledge graph (Human-Object-Word Graph), information about people, things, events, or places extracted from the episode nodes of the first knowledge graph is represented by four types of nodes: person nodes, object nodes, event nodes, and place nodes. The connections between these nodes are made up of edges. This classifies the information in the second knowledge graph (Human-Object-Word Graph) into four types of nodes, allowing users to predict the type of content each node (information) contains. This makes it easier for users to read through the nodes (information) and understand the information in the second knowledge graph (Human-Object-Word Graph). Furthermore, by reading through the information organized in the second knowledge graph (Human-Object-Word Graph), users can discover new knowledge. In addition, the classification of information in the second knowledge graph (Human-Object-Word Graph) into four types of nodes allows for a gamified element where users can select any node they are interested in from among the person, object, event, or place nodes. As a result, users can enjoy reading through episodes that become new knowledge and acquire knowledge in an enjoyable way. Thus, the information processing device 1 can generate knowledge graphs that are easy for people to see and understand, and that can lead to the discovery of new knowledge.

[0090] (2) In such an information processing device 1, the timeline edge may include at least one of a physical timeline edge that shows the timeline or procedure, and a logical timeline edge that shows the timeline based on causal relationships. This allows a person to see and understand that there are physical timeline relationships and logical timeline relationships between the episode nodes connected by the timeline edge.

[0091] (3) The system can also be further equipped with a generation unit 13 that analyzes text using the 5W1H framework, generates episode nodes and sequence edges by treating sentences containing 4W elements as episodes and sentences containing 1W1H elements as sequences. This allows the episodes, which become the information (episode nodes) in the first knowledge graph (episode graph), to have a 5W1H narrative structure, making it easier for people to read through the episode nodes and understand the information in the first knowledge graph (episode graph).

[0092] (4) Here, the text may represent a story relating to history or legend, the person node may represent the subject of the episode, the place node may represent a location related to the episode, the object node may represent an object related to the episode, and the event node may represent an event related to the episode.

[0093] For example, stories about history or legends are a collection of anecdotes containing the 5W1H elements related to numerous events that occurred in the past, and can be considered knowledge that is enjoyed and understood by many people. However, history taught in school is sometimes ridiculed as merely a "subject to be memorized," and for some people, it appears as nothing more than a massive block of information without the appeal of history as a story, leading to what is known as "historical aversion."

[0094] The information processing device 1 can classify episodes such as history, represented in the second knowledge graph (human-object-word graph), into the aforementioned person nodes, object nodes, event nodes, and place nodes, and display them in an organized manner, showing what kind of content each episode contains. This makes it easier for people to read through episodes such as history, and allows them to appreciate the interest of history.

[0095] (5) The system may also be further equipped with a fourth control unit 154 that, when the reception unit receives a designation of a predetermined node from among the nodes included in the second knowledge graph (human-object-word graph), outputs an episode description associated with the predetermined node and an episode description showing the relationship between the predetermined node and other nodes connected by edges. As a result, the episode description (content) associated with the node specified by the person and the content of episodes related to the specified node are output, so that by reading through related episodes, new knowledge can be discovered and knowledge can be acquired in an enjoyable way.

[0096] (6) Furthermore, each episode includes time information about when an event occurred, and the fourth control unit 154 can also display multiple episodes in chronological order based on the time information if multiple episodes are associated with a given node. In this way, if there are multiple episodes associated with a specified node and those episodes include time information, the multiple episodes are displayed in chronological order. This makes it easier for people to read through the episodes and acquire knowledge in an enjoyable way. [Explanation of symbols]

[0097] 1. Information Processing Device 2 User terminals 11 Acquisition Department 12 Analysis Department 13 Generation part 14 Storage section 15 Control Unit 16 Reception Department 21 Display section 22 Terminal acquisition section 23 Instruction transmission unit 151 First Control Unit 152 Second Control Unit 153 Third Control Unit 154 Fourth Control Unit 170 CPU 171 memory 172 Storage Devices 173 connection ports 174 Communication Circuit 175 Bus Line Episode nodes 301, 302, 303, 304, 305, 306, and 307 401, 402, 403, 404, 405 Latitude / Longitude Edge 501, 502, 503 person nodes 504, 505 Field Nodes 506, 507 Object Nodes 508 Node 601, 602, 603, 604, 605, 606, 607, 608 Edge 1711 OS 1712 Browser Program 1713 Information Processing Device Program 1715 Episode Graph DB 1716 Human-Thing-Word-Graph DB 1717 User Terminal Program

Claims

1. An information processing device that generates a knowledge graph composed of multiple nodes and edges indicating the relationships between each of the multiple nodes, A storage unit that stores a first knowledge graph generated by analyzing a text, the first knowledge graph including episode nodes representing each of a plurality of episodes contained in the text, and history edges representing the history contained in the text. A first control unit performs control to output the first knowledge graph from the storage unit in a manner in which the episode nodes included in the first knowledge graph are connected by the longitude edges, A reception desk that receives instructions from users, A second control unit outputs information indicating an episode associated with the episode node included in the output first knowledge graph, based on the instruction received by the reception unit, The system comprises: a field node which is a node relating to a location; a person node which is a node relating to a person; an object node which is a node relating to an object other than a person; an event node which is a node relating to an event; and an edge connecting the two nodes, and a third control unit which controls the output of a second knowledge graph generated based on the first knowledge graph, in a manner in which the at least two nodes are connected by the edge, An information processing device wherein the at least two nodes are generated from the episode nodes included in the first knowledge graph, and the edges connecting the two nodes are generated from the history edges.

2. The information processing apparatus according to claim 1, wherein the history edge includes at least one of a physical history edge indicating history over time or in a procedure, and a logical history edge indicating history based on causal relationships.

3. The information processing apparatus according to claim 1, further comprising a generation unit that performs a 5W1H analysis on the aforementioned text, generates episode nodes and history edges, with sentences containing each of the 4W (When, Where, Who, What) elements as episodes and sentences containing each of the 1W1H (Why, How) elements as history.

4. The aforementioned text presents a story related to history or legend. The aforementioned person node is a node representing the entity that is the subject of the aforementioned episode, The field node is a node representing a location related to the episode, The aforementioned object node is a node that represents an object related to the episode, The information processing apparatus according to claim 1, wherein the article node is a node representing an event related to the episode.

5. The information processing apparatus according to claim 4, further comprising a fourth control unit that, when the receiving unit receives a designation of a predetermined node among the nodes included in the second knowledge graph, outputs an episode description associated with the predetermined node and an episode description showing the relationship between the predetermined node and other nodes connected by the edge.

6. The aforementioned episode includes time information of when the event occurred, The information processing apparatus according to claim 5, wherein the fourth control unit further displays the multiple episodes in chronological order based on the time information when multiple episodes are associated with the predetermined node.

7. In an information processing device that generates a knowledge graph composed of multiple nodes and edges indicating the relationships between each of the multiple nodes, A first control step involves controlling a storage unit that stores a first knowledge graph generated by analyzing a text, the first knowledge graph including episode nodes representing each of a plurality of episodes contained in the text and history edges representing the history contained in the text, to output the first knowledge graph in a manner in which the episode nodes contained in the first knowledge graph are connected by the history edges. A reception step for receiving instructions from the user, A second control step outputs information indicating the episode associated with the episode node included in the outputted first knowledge graph, based on the instruction received in the reception step. A third control step is performed to cause a computer to output a second knowledge graph, which is generated based on the first knowledge graph, in a manner in which the two nodes are connected by the edge, and which includes at least two of the following nodes: a field node which is a node relating to a location, a person node which is a node relating to a person, an object node which is a node relating to an object other than a person, and an event node which is a node relating to an event, and an edge connecting the two nodes. A program in which at least two of the aforementioned nodes are generated from the episode nodes included in the first knowledge graph, and the edges connecting the two nodes are generated from the history edges.

8. A method for generating a knowledge graph comprising multiple nodes and edges indicating the relationships between each of the multiple nodes, A first control step involves controlling a storage unit that stores a first knowledge graph generated by analyzing a text, the first knowledge graph including episode nodes representing each of a plurality of episodes contained in the text and history edges representing the history contained in the text, to output the first knowledge graph in a manner in which the episode nodes contained in the first knowledge graph are connected by the history edges. A reception step for receiving instructions from the user, A second control step outputs information indicating the episode associated with the episode node included in the outputted first knowledge graph, based on the instruction received in the reception step. The third control step includes controlling the output of a second knowledge graph generated based on the first knowledge graph, in which the two nodes are connected by the edge, and which includes at least two of the following nodes: a field node which is a node relating to a location, a person node which is a node relating to a person, an object node which is a node relating to an object other than a person, and an event node which is a node relating to an event, and an edge connecting the two nodes, A method for generating a knowledge graph, wherein the at least two nodes are generated from the episode nodes included in the first knowledge graph, and the edges connecting the two nodes are generated from the history edges.

Images