Apparatus and method for generating structured data for documents representing multiple procedures

Abstract

To generate highly accurate structured data for a document which represents multiple procedures.SOLUTION: A data processing device performs structuralization update processing for generating updated structured data of structured data relating to a document representing a plurality of procedures. The structured data are graph data representing a graph including a plurality of entity nodes and one or a plurality of edges. Each of the plurality of entity nodes is a node representing an entity in the document. The structuralization update processing involves updating a structured graph, which is a graph represented by the structured data or a copy thereof, on the basis of update definition data, which are data defining at least one update of the nodes and the edges using an expression employing a taxonomy, and the taxonomy of at least one entity node in the structured graph. The updated structured data are data representing a graph after update of the structured graph.SELECTED DRAWING: Figure 25

Description

[Technical Field]

[0001] The present invention generally relates to generating structured data for documents representing multiple procedures. [Background technology]

[0002] In recent years, there has been an apparent need in various fields to use AI to support, streamline, and optimize business processes consisting of multiple steps. For example, in the industrial field, AI that recommends operating procedures for equipment and processes for dealing with equipment failures has been put to practical use, in the medical field, AI that supports diagnosis, treatment, and medication has been put to practical use, and in the materials field, AI that recommends synthesis processes for new materials has been put to practical use.

[0003] To realize business process support using AI, it is generally necessary to prepare data that can process the business process. However, information related to business processes is often stored as documents written in natural language (such as equipment maintenance reports, medical charts, and experiment reports), making it difficult to process the information as is. Therefore, the information written in the documents must be converted into structured data that can be processed.

[0004] 24A and 24B are diagrams showing an image of structuring business processes. Fig. 24A shows an image of structuring business processes related to maintenance, and Fig. 24B shows an image of structuring business processes related to material manufacturing.

[0005] Manually generating structured data from documents requires a huge amount of time and specialized knowledge. Therefore, a technology that automatically generates structured data from documents is desired. In response to this, the technologies described in Patent Document 1 and Non-Patent Document 1 are known.

[0006] Patent Document 1 describes a document understanding support device "including a word extraction condition learning unit, a word extraction unit, a word relationship extraction condition learning unit, a word relationship extraction unit, and an output unit." It also describes that "the word extraction condition learning unit generates word extraction conditions for extracting words from the supporting electronic document by learning based on the feature amounts assigned to each word," "the word extraction unit extracts words that satisfy the word extraction conditions," "the word relationship extraction condition learning unit generates word relationship extraction conditions for extracting related words from the supporting electronic document by learning based on the feature amounts for the word relationships to be extracted," and "the word relationship extraction unit extracts word relationships that satisfy the word relationship extraction conditions."

[0007] Non-patent document 1 describes a method for recognizing order using rules that feature dependencies between ingredients and operations, step numbers, etc., to perform state transitions and information complementation in structured cooking recipes. [Prior art documents] [Patent documents]

[0008] [Patent Document 1] Japanese Patent Application Publication No. 2019-79321 [Non-patent literature]

[0009] [Non-Patent Document 1] Cooking Scenario - Recipe Scenario and its Applications, IPSJ Research Report Database System (DBS), 2003(71(2003-DBS-131)), pp.25-31 Summary of the Invention [Problem to be solved by the invention]

[0010] The technology in Patent Document 1 requires a large amount of training data to ensure accuracy, making it difficult to apply to fields with little training data.

[0011] In addition, in the technology of Non-Patent Document 1, the rules are written based on Japanese grammar, so rules must be created for each language. Furthermore, when the description of consecutive procedures is scattered in a document (for example, when multiple processes are described across sentences), it is difficult to create rules.

[0012] The structured data generated for a document can be used for any given or any purpose. For example, the structured data can be used as data for training a machine learning model. If the structured data is inaccurate (for example, if information is missing due to omissions in the document), it may be difficult to achieve the given or any purpose.

[0013] However, for the reasons mentioned above, it is difficult for the techniques disclosed in Patent Document 1 and Non-Patent Document 1 to generate accurate structured data for a document.

[0014] The present invention has been made in view of the above-mentioned problems, and has as its object to generate accurate structured data for a document that represents a plurality of procedures. [Means for solving the problem]

[0015] A representative example of the invention disclosed in the present application is as follows. That is, a data processing device performs a structured update process, which is a process of generating updated structured data for structured data on a document representing multiple procedures. The structured data is graph data representing a graph including multiple entity nodes and one or multiple edges. Each of the multiple entity nodes is a node representing an entity in the document. The structured update process includes updating a structured graph, which is a graph represented by the structured data or a copy thereof, based on update definition data, which is data that defines an update of at least one node and edge using a taxonomy, and the taxonomy of at least one entity node in the structured graph. The updated structured data is data that represents the graph after the structured graph has been updated. [Effects of the Invention]

[0016] According to the present invention, it is possible to generate highly accurate structured data for a document that describes a plurality of procedures. Problems, configurations, and effects other than those described above will become apparent from the following description of the embodiments. [Brief explanation of the drawings]

[0017] [Figure 1] FIG. 1 illustrates an example of a system according to a first embodiment. [Figure 2] FIG. 2 is a diagram illustrating an example of a hardware configuration of a computer according to a first embodiment. [Figure 3] FIG. 2 is a diagram illustrating an example of a document database according to the first embodiment. [Figure 4] FIG. 2 is a diagram illustrating an example of an entity / category dictionary stored in a structure rule database according to the first embodiment. [Figure 5] FIG. 10 is a diagram illustrating an example of procedure category determination rule information stored in the structuring rule database according to the first embodiment. [Figure 6] FIG. 10 is a diagram illustrating an example of main entity determination rule information stored in the structure rule database according to the first embodiment. [Figure 7] FIG. 10 is a diagram illustrating an example of parallelism determination rule information stored in the structure rule database according to the first embodiment. [Figure 8] FIG. 10 is a diagram illustrating an example of business process order determination rule information stored in the structuring rule database according to the first embodiment. [Figure 9] FIG. 10 is a diagram illustrating an example of procedure order determination rule information stored in the structuring rule database according to the first embodiment. [Figure 10] FIG. 10 is a diagram illustrating an example of relationship definition information stored in a structure rule database according to the first embodiment. [Figure 11] 1 is a flowchart illustrating an outline of a structured data generation process executed by a structuring processing device according to a first embodiment. [Figure 12] FIG. 2 is a diagram showing an example of information generated by the structuring processing device according to the first embodiment. [Figure 13]FIG. 2 is a diagram showing an example of information generated by the structuring processing device according to the first embodiment. [Figure 14] FIG. 2 is a diagram showing an example of information generated by the structuring processing device according to the first embodiment. [Figure 15] FIG. 2 is a diagram showing an example of information generated by the structuring processing device according to the first embodiment. [Figure 16] FIG. 2 is a diagram showing an example of information generated by the structuring processing device according to the first embodiment. [Figure 17] FIG. 2 is a diagram showing an example of information generated by the structuring processing device according to the first embodiment. [Figure 18] FIG. 2 is a diagram showing an example of structured data generated by the structuring processing device according to the first embodiment. [Figure 19A] FIG. 10 is a diagram illustrating an example of structured data displayed on a user terminal according to the first embodiment. [Figure 19B] FIG. 10 is a diagram illustrating an example of structured data displayed on a user terminal according to the first embodiment. [Figure 20] 10 is a flowchart illustrating an example of a procedure category determination process executed by the structuring processing device according to the first embodiment. [Figure 21] 10 is a flowchart illustrating an example of a main entity determination process executed by the structuring processing device according to the first embodiment. [Figure 22] 10 is a flowchart illustrating an example of a parallelism determination process executed by the structuring processing device according to the first embodiment. [Figure 23] 10 is a flowchart illustrating an example of a procedure order determination process executed by the structuring processing device according to the first embodiment. [Figure 24A] This is a diagram showing an image of structuring a business process. [Figure 24B] This is a diagram showing an image of structuring a business process. [Figure 25] FIG. 10 is a diagram showing an example of an overall flow of the second embodiment. [Figure 26] FIG. 10 is a diagram showing an example of taxonomy data in Example 2. [Figure 27] FIG. 11 is a diagram illustrating an example of an integrated graph indicated by integrated graph data in the second embodiment. [Figure 28]FIG. 10 is a diagram illustrating an example of processing list data according to the second embodiment. [Figure 29] FIG. 11 is a diagram illustrating an example of a generalized graph indicated by generalized graph data in the second embodiment. [Figure 30] FIG. 11 is a diagram illustrating an example of a graph pattern indicated in a graph pattern database according to the second embodiment. [Figure 31] 10 is a flowchart illustrating an example of graph integration processing executed by the structuring processing device according to the second embodiment. [Figure 32] 10 is a flowchart illustrating an example of graph generalization processing executed by the structuring processing device according to the second embodiment. [Figure 33] 10 is a flowchart illustrating an example of a graph update process executed by the structuring processing device according to the second embodiment. [Figure 34A] FIG. 10 is a diagram showing an example of a separation pattern in the second embodiment. [Figure 34B] FIG. 10 is a diagram illustrating an example of a replication pattern 1 in the second embodiment. [Figure 34C] FIG. 10 is a diagram illustrating an example of a replication pattern 2 in the second embodiment. [Figure 34D] FIG. 10 is a diagram illustrating an example of an additional pattern 1 in the second embodiment. [Figure 34E] FIG. 10 is a diagram illustrating an example of an additional pattern 2 in the second embodiment. [Figure 34F] FIG. 10 is a diagram illustrating an example of an integrated pattern 1 in the second embodiment. [Figure 34G] FIG. 10 is a diagram illustrating an example of an integrated pattern 2 in the second embodiment. [Figure 34H] FIG. 10 is a diagram showing an example of an edge replacement pattern in the second embodiment. [Figure 35] FIG. 11 is a diagram showing an example of the overall flow of the third embodiment. [Figure 36] FIG. 11 is a diagram illustrating an example of a master-slave relationship table according to the third embodiment. [Figure 37] FIG. 11 is a diagram showing an example of replaced structured data in the third embodiment. [Figure 38]11 is a flowchart illustrating an example of graph generalization processing executed by a structuring processing device according to a third embodiment. [Figure 39] FIG. 11 is a diagram showing a specific example of graph generalization processing executed by the structuring processing device according to the third embodiment. [Figure 40] FIG. 10 is a diagram showing an example of the overall flow of the fourth embodiment. [Figure 41] FIG. 13 is a diagram illustrating an example of a name matching relationship table according to the fourth embodiment. [Figure 42] FIG. 13 is a diagram showing an example of merged taxonomy data in the fourth embodiment. [Figure 43] 13 is a flowchart illustrating an example of graph generalization processing executed by the structuring processing device according to the fourth embodiment. [Figure 44] FIG. 13 is a diagram showing a specific example of graph generalization processing executed by the structuring processing device according to the fourth embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0018] Hereinafter, embodiments of the present invention will be described with reference to the drawings. The following description and drawings are examples for explaining the present invention, and some omissions and simplifications have been made as appropriate for clarity of explanation. The present invention can be implemented in various other forms. Unless otherwise specified, each component may be singular or plural.

[0019] In the following explanation, the same or similar components may be assigned the same reference numerals, and redundant explanations may be omitted. Furthermore, in the following explanation, the letter "S" before a reference numeral indicates a processing step. Furthermore, in the following explanation, various types of information may be described using expressions such as "table" and "information," but the various types of information may also be expressed using data structures other than these.

[0020] Furthermore, in the following explanation, an example of structuring information about the synthesis process of a material described in an experimental report will be described, but the structuring object can be applied to the various fields, objects, and use cases described in the background art. [Example]

[0021] Fig. 1 is a diagram illustrating an example of a system according to the first embodiment. Fig. 2 is a diagram illustrating an example of a hardware configuration of a computer 200 according to the first embodiment.

[0022] The system 10 shown in Fig. 1 is composed of a structuring processing device 100 and a user terminal 101. The structuring processing device 100 and the user terminal 101 are connected via a communication network 102 in a state where two-way communication is possible. The communication network 102 is, for example, a LAN (Local Area Network), a WAN (Wide Area Network), the Internet, a public communication network, a dedicated line, etc. The number of user terminals 101 may be two or more. In the following description, the system 10 is also referred to as a structuring system 10.

[0023] The structuring processing device 100 and the user terminal 101 are configured, for example, by a computer 200 as shown in Fig. 2. The computer 200 includes an arithmetic unit 201, a main memory unit 202, an auxiliary memory unit 203, an input unit 204, an output unit 205, and a communication unit 206.

[0024] The arithmetic device 201 executes a program stored in the main memory device 202. The arithmetic device 201 is, for example, a central processing unit (CPU), a micro processing unit (MPU), a graphics processing unit (GPU), a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), an artificial intelligence (AI) chip, etc. The arithmetic device 201 operates as a functional unit (module) that realizes a specific function by executing processing in accordance with the program. In the following description, when processing is described using a functional unit as the subject, it indicates that the arithmetic device 201 is executing a program that realizes the functional unit.

[0025] The main memory device 202 stores programs and data executed by the arithmetic device 201. The main memory device 202 is, for example, a non-volatile memory such as a read-only memory (ROM), a random access memory (RAM), or a non-volatile RAM (NVRAM). The main memory device 202 is also used as a work area.

[0026] The auxiliary storage device 203 permanently stores data. The auxiliary storage device 203 is, for example, a solid state drive (SSD) or a hard disk drive. The computer 200 does not necessarily have to have the auxiliary storage device 203. In this case, the programs and data may be acquired from an optical storage device such as a compact disc (CD) or a digital versatile disc (DVD), an IC card, an SD card, or the like, or may be acquired from a storage area on an externally connected storage system or cloud system. The programs and data stored in the auxiliary storage device 203 are read by the arithmetic device 201 and loaded into the main storage device 202.

[0027] The input device 204 is an interface that accepts input from the outside, and is, for example, a keyboard, a mouse, a touch panel, a card reader, a pen-input tablet, a voice input device, or the like.

[0028] The output device 205 is an interface that outputs various information such as processing progress and processing results, etc. The output device 205 is, for example, a display device such as a liquid crystal monitor or LCD (Liquid Crystal Display), an audio output device, a printer, etc.

[0029] The computer 200 does not necessarily have to have the input device 204 and the output device 205. In this case, the computer 200 inputs and outputs information via the communication device 206.

[0030] The communication device 206 communicates with other devices, and is, for example, a network interface card (NIC), a wireless communication module, a USB module, or the like.

[0031] The structuring processing device 100 generates structured data from document data including text in which a business process is described in a natural language.

[0032] Here, it is assumed that a business process is composed of multiple procedures. The structured data is data for understanding the structure of the multiple procedures, and may be, for example, data in JSON format, data in XML format, data in RDF format, or data in GraphML format. The present invention is not limited to the data format of the structured data. It is assumed that the structured data in Example 1 is data in GraphML format.

[0033] In the following, one or more sentences or a group of one or more sentences that describe a business process will be referred to as a document. In addition, in the following, we will explain that processing is performed on a document-by-document basis, but the processing unit is not necessarily limited to this.

[0034] The structuring processing device 100 has an information management unit 110 and a structuring processing unit 120 , and also holds a document database 130 , a structuring rule database 140 , a processing database 150 , and a structured data database 160 .

[0035] The document database 130 is a database that stores documents to be processed. The structuring rule database 140 is a database that stores rules used in structuring processing. The processing database 150 is a database that stores the results of structuring processing. The structured data database 160 is a database that stores structured data generated by structuring processing.

[0036] The information management unit 110 manages documents, rules, structured data, etc. The structuring processing unit 120 executes structuring processing. Note that the information management unit 110 and the structuring processing unit 120 may be realized as a function of middleware that manages an operating system, a file system, a relational database, and NoSQL such as KVS (Key-Value Store).

[0037] The structuring processing unit 120 executes the following processes in the structuring process.

[0038] (1) The structuring processor 120 extracts expressions such as words related to the procedures of a business process from the text included in the document as entities, and classifies the extracted entities into categories (entity categories).

[0039] (2) The structuring processor 120 generates an entity group by grouping together entities related to one procedure.

[0040] (3) The structuring processor 120 classifies the category of the procedure (procedure category) corresponding to the entity group based on the entity category of the entity included in the entity group.

[0041] (4) The structuring processing unit 120 identifies, from among the entities included in the entity group, an entity (main entity) that represents the characteristics of the procedure corresponding to the entity group.

[0042] (5) The structuring processor 120 determines which procedures are performed in parallel among the procedures included in the business process based on the relationships between the main entities.

[0043] (6) The structuring processing unit 120 determines the order of the procedures based on the relationships between the main entities and the relationships between the order of the procedures and the procedure categories.

[0044] (7) The structuring processing unit 120 checks the consistency of the determination results of (5) and (6) and records the check result.

[0045] (8) The structuring processing unit 120 generates structured data based on the results of the determinations in (5) and (6) and the results of the consistency check.

[0046] (9) The structuring processing unit 120 generates display information for displaying the structured data and transmits the display information to the user terminal 101.

[0047] The user terminal 101 has a registration unit 170 that displays a screen for registering documents and various rules, and a display unit 180 that displays a screen for presenting and correcting structured data.

[0048] The functions of the structuring processing device 100 may be realized using a computer system configured from a plurality of computers 200. All or part of the functions of the structuring processing device 100 may also be realized using virtualization technology. For example, a method of realizing all or part of the functions of the structuring processing device 100 using cloud services such as SaaS (Software as a Service), PaaS (Platform as a Service), and IaaS (Infrastructure as a Service) is conceivable.

[0049] The structuring processing device 100 and the user terminal 101 may be integrated into one device.

[0050] FIG. 3 is a diagram illustrating an example of the document database 130 according to the first embodiment.

[0051] The document database 130 stores entries each including a document ID 301 and text 302. One entry exists for one document. Note that the fields included in an entry are merely an example and are not limiting.

[0052] Document ID 301 is a field for storing identification information of a document. Text 302 is a field for storing text included in a document. Note that the data format of the text stored in text 302 is not limited.

[0053] FIG. 4 is a diagram showing an example of the entity / category dictionary 400 stored in the structure rule database 140 according to the first embodiment.

[0054] The entity / category dictionary 400 is information for managing expressions such as words extracted as entities and the categories (types) of the entities. The entity / category dictionary 400 stores entries including entities 401 and categories 402. One entry exists for one expression (entity). Note that the fields included in the entry are merely examples and are not limiting.

[0055] Entity 401 is a field for storing an expression to be extracted. Category 402 is a field for storing the entity category of the expression.

[0056] FIG. 5 is a diagram showing an example of procedure category determination rule information 500 stored in the structuring rule database 140 according to the first embodiment.

[0057] The procedure category determination rule information 500 is information for managing determination rules for procedure categories of procedures corresponding to an entity group. The procedure category determination rule information 500 stores entries including a rule ID 501, a category ID 502, a category 503, and a rule 504. One entry exists for one rule. Note that the fields included in the entry are merely examples and are not limited to these.

[0058] Rule ID 501 is a field that stores identification information of a rule. Category ID 502 is a field that stores identification information of a procedure category of a procedure that matches the rule. Category 503 is a field that stores the procedure category of a procedure that matches the rule. Rule 504 is a field that stores a determination rule for the procedure category.

[0059] Here, the procedure category is a type of procedure. In a business process related to material production, procedure categories such as "preparation," "operation," and "measurement" are possible, while in a business process related to maintenance, procedure categories such as "reporting," "cause identification," and "treatment" are possible.

[0060] Possible rules for determining procedure categories include rules that use the entity categories of entities included in an entity group. For example, there is a rule that determines the procedure category of an entity group that includes an entity whose entity category is "substance" as "substance." There are also possible rules that determine procedure categories based on a combination of entity categories included in an entity group. For example, in the business process related to maintenance in FIG. 24A, there is a rule that determines the procedure category of an entity group that includes entities whose entity categories are "alarm" and "phenomenon" as "report." Note that the above-mentioned rules are merely examples and are not limiting.

[0061] The first entry in Figure 5 defines a rule that determines the procedure category to be "operation" if the variable "entity_categories" that represents the entity category of each entry in the entity group contains "operation." The second entry in Figure 5 defines a rule that determines the procedure category to be "substance" if the variable "entity_categories" contains "substance."

[0062] FIG. 6 is a diagram showing an example of main entity determination rule information 600 stored in the structure rule database 140 according to the first embodiment.

[0063] The main entity determination rule information 600 is information for managing rules (main entity determination rules) for identifying a main entity from among the entities included in an entity group. The main entity determination rule information 600 stores entries including a rule ID 601 and a rule 602. One entry exists for one rule. Note that the fields included in the entry are merely examples and are not limited to these.

[0064] The rule ID 601 is a field for storing identification information of the rule. The rule 602 is a field for storing the main entity determination rule.

[0065] The primary entity determination rule may be a rule that uses an entity category. For example, it may be a rule that identifies an entity whose entity category is "substance" as the primary entity. Note that the above rule is merely an example and is not limiting.

[0066] The first entry in FIG. 6 defines a rule that specifies an entity whose variable "entity_category" representing the entity category is "operation" as the primary entity.

[0067] Additionally, the structured rule database 140 may include information for managing rules for identifying secondary entities that are complementary to primary entities.

[0068] FIG. 7 is a diagram showing an example of parallelism determination rule information 700 stored in the structure rule database 140 according to the first embodiment.

[0069] The parallelism determination rule information 700 is information for managing rules (parallelism determination rules) for determining whether two procedures are performed in parallel. The parallelism determination rule information 700 stores entries including a rule ID 701, parallelism 702, and rule 703. One entry exists for one rule. Note that the fields included in the entry are merely an example and are not limited to this.

[0070] Rule ID 701 is a field that stores identification information of a rule. Parallelism 702 is a field that stores a value that indicates whether two procedures are performed in parallel. Rule 703 is a field that stores a parallelism determination rule.

[0071] The parallelism determination rule may be a rule that uses a phrase contained in a sentence that connects the main entities of two entity groups. Note that the above-mentioned rule is an example and is not limited to this.

[0072] The first entry in Figure 7 defines a rule that determines that if the variable "word_between main_entityA_and_main_entityB," which represents the words included in the sentence connecting main entity A and main entity B, contains "however," then the procedure corresponding to the entity group including main entity A and the procedure corresponding to the entity group including main entity B are performed in parallel. The second entry in Figure 7 defines a rule that determines that if the variable "word_between main_entityA_and_main_entityB" contains "after," then the procedure corresponding to the entity group including main entity A and the procedure corresponding to the entity group including main entity B are not performed in parallel.

[0073] FIG. 8 is a diagram illustrating an example of business process order determination rule information 800 stored in the structuring rule database 140 according to the first embodiment.

[0074] The business process order determination rule information 800 is information for managing rules (business process order determination rules) for determining the order of each procedure based on the procedure category. The business process order determination rule information 800 stores entries including a rule ID 801, an order 802, and a rule 803. There is one entry for one rule. Note that the fields included in the entry are merely examples and are not limited to these.

[0075] Rule ID 801 is a field that stores identification information for a rule. Order 802 is a field that stores information that indicates the general order of the procedures. "Start point" indicates the first procedure of the entire business process, "middle" indicates an intermediate procedure of the entire business process, and "end point" indicates the final procedure of the entire business process. Rule 803 is a field that stores the business process order determination rule.

[0076] A possible business process sequence determination rule is a rule that uses only the procedure category. Note that the above-mentioned method of defining a procedure pattern is an example and is not intended to be limiting. For example, a rule that uses the procedure category and the position of the main entity may also be used.

[0077] Depending on the business process, it may be common to generate structured data in which procedures are arranged in a predetermined order. For example, in the business process related to maintenance shown in Figure 24A, the procedures are generally arranged in the order of "report," "cause confirmation," and "action." Therefore, the order of the procedures in the structured data is defined in advance.

[0078] The first entry in Figure 8 defines a rule that determines that a procedure is the first procedure of the entire business process when the procedure category is "substance" and the primary entity is in the first half of the text. The second entry in Figure 8 defines a rule that determines that a procedure is an intermediate procedure of the entire business process when the procedure category is "operation". The third entry in Figure 8 defines a rule that determines that a procedure is a latter procedure of the entire business process when the procedure category is "substance" and the primary entity is in the second half of the text.

[0079] FIG. 9 is a diagram showing an example of procedure order determination rule information 900 stored in the structure rule database 140 according to the first embodiment.

[0080] The procedure order determination rule information 900 is information for managing rules (procedure order determination rules) for determining the order between two procedures based on the relationship between the main entities. The procedure order determination rule information 900 stores entries including a rule ID 901, an order 902, and a rule 903. One entry exists for one rule. Note that the fields included in the entry are merely an example and are not limited to this.

[0081] Rule ID 901 is a field that stores identification information of a rule. Order 902 is a field that stores an order relationship between entities. Rule 903 is a field that stores a procedure order determination rule.

[0082] The procedure order determination rule may be a rule using words contained in sentences connecting main entities. It may also be a rule based on entities that have a synonymous relationship. For example, if "Disk No. 3" and "Disk 3" are synonymously related, a rule may be used to arrange an entity group containing "Disk No. 3" and an entity group containing "Disk 3" in order of appearance. In addition to synonymous relationships, relationships between device configuration states (within modules of the same device), relationships between substances, etc. may also be used. The above-mentioned rules are merely examples and are not limiting.

[0083] The first entry in Figure 9 defines a rule that arranges an entity group containing main entity A before an entity group containing main entity B if the variable "word_between main_entityA_and_main_entityB," which represents the words included in the sentence connecting main entity A and main entity B, contains "after." The second entry in Figure 9 defines a rule that arranges an entity group containing main entity B before an entity group containing main entity A if the variable "word_between main_entityA_and_main_entityB" contains "before." The third entry in Figure 9 defines a rule that arranges an entity group containing main entity A at the beginning of the business process if the variable "main_before main_entityA," which represents the word immediately before main entity A, contains "beginning." The fourth entry in Figure 9 defines a rule that arranges an entity group containing main entity A before an entity group containing main entity B if the variable "main_entityA," which represents the main entity, and the variable "main_entityB," which represents main entity B, contain terms indicating a specific relationship. The specific relationship is defined in relationship definition information 1000 (see FIG. 10) which will be described later.

[0084] FIG. 10 is a diagram illustrating an example of the relationship definition information 1000 stored in the structuring rule database 140 of the first embodiment.

[0085] The relationship definition information 1000 is information for managing a specific relationship between entities (for example, a similarity relationship). The relationship definition information 1000 stores entries including a relationship ID 1001, a first entity 1002, a second entity 1003, and a relationship 1004. One entry exists for a relationship between one entity. Note that the fields included in the entry are merely examples and are not limited to these.

[0086] Relationship ID 1001 is a field that stores identification information of a relationship. First entity 1002 and second entity 1003 are fields that store entities. Relationship 1004 is a field that stores the relationship between the first entity and the second entity.

[0087] Fig. 11 is a flowchart outlining the structured data generation process executed by the structuring processing device 100 of the first embodiment. Figs. 12, 13, 14, 15, 16, and 17 are diagrams showing examples of information generated by the structuring processing device 100 of the first embodiment. Fig. 18 is a diagram showing an example of structured data generated by the structuring processing device 100 of the first embodiment. Figs. 19A and 19B are diagrams showing an example of structured data displayed on the user terminal 101 of the first embodiment.

[0088] The structuring processing device 100 starts the structured data generation process when it detects an execution trigger. The execution trigger is the reception of an execution instruction and the detection of execution timing, etc. In the following explanation, the process when an execution instruction including identification information of a document for which one piece of structured data is to be generated is explained as an example.

[0089] The structuring processing unit 120 acquires the text of the specified document from the document database 130, and executes entity extraction processing using the text and the entity / category dictionary 400 (step S1100). The structuring processing unit 120 stores information on the extracted entities in the processing database 150 as entity information 1200.

[0090] The entity information 1200 stores entries each including an entity ID 1201, an entity 1202, a location 1203, and a category 1204. One entry exists for one entity. Note that the fields included in the entry are merely an example and are not limited to these.

[0091] Entity ID 1201 is a field that stores identification information of an entity assigned by the structuring processor 120. Entity 1202 is a field that stores an expression extracted as an entity. Position 1203 is a field that stores the position of the entity in the text. Category 1204 is a field that stores the entity category.

[0092] In the entity extraction process, the structuring processor 120 extracts entities based on the entity / category dictionary 400 and generates entity information 1200 based on the extraction results. Note that the entity extraction method is not limited to the rule-based method. Existing named entity extraction techniques such as machine learning can also be used.

[0093] Next, the structuring processor 120 executes entity group generation processing using the extracted entities and texts (step S1200). Specifically, the following processing is executed.

[0094] (S1200-1) The structuring processing unit 120 executes document structure analysis processing on the text and acquires information about entity dependencies. The structuring processing unit 120 generates entity pairs having a corresponding relationship based on the information about entity dependencies. The entity pairs may be generated using a model that has learned the entity correspondences. The structuring processing unit 120 stores information about the generated pairs in the processing database 150 as entity pair information 1300.

[0095] The entity pair information 1300 stores entries including a pair ID 1301, an entity ID 1302, and an entity ID 1303. One entry exists for one entity pair. Note that the fields included in the entry are merely an example and are not limited to this.

[0096] Pair ID 1301 is a field for storing identification information of a pair of entities. Entity ID 1302 and entity ID 1303 are fields for storing identification information of the entities that make up the pair.

[0097] (S1200-2) The structuring processing unit 120 generates an entity group by grouping entities that are linked by a correspondence relationship with reference to the entity pair information 1300. The structuring processing unit 120 stores information on the generated entity group in the processing database 150 as entity group information 1400.

[0098] The entity group information 1400 stores entries including an entity group ID 1401, an entity list 1402, a category 1403, and a main entity ID 1404. There is one entry for one entity group. Note that the fields included in the entry are merely examples and are not limited to these.

[0099] The entity group ID 1401 is a field that stores the identification information of the entity group. The entity list 1402 is a field that stores a list of the identification information of the entities that make up the entity group. The category 1403 is a field that stores the procedure category. The main entity ID 1404 is a field that stores the identification information of the main entity of the entity group. At this point, the category 1403 and main entity ID 1404 of each entry are blank.

[0100] This concludes the description of the entity group generation process.

[0101] Next, the structuring processing unit 120 executes a procedure category determination process using the procedure category determination rule information 500 (step S1300). Details of the procedure category determination process will be explained using Fig. 20. The result of the procedure category determination process is reflected in the category 1403 of each entry in the entity group information 1400.

[0102] Next, the structuring processing unit 120 executes a main entity determination process using the main entity determination rule information 600 (step S1400). Details of the main entity determination process will be explained using Fig. 21. The result of the main entity determination process is reflected in the main entity ID 1404 of each entry in the entity group information 1400.

[0103] Next, structuring processor 120 executes parallelism determination processing using parallelism determination rule information 700 (step S1500). Details of the parallelism determination processing will be described with reference to Fig. 22. The results of the parallelism determination processing are stored in process database 150 as parallelism information 1500.

[0104] The parallelism information 1500 stores entries including an entity family ID 1501 and an entity group list 1502. There is one entry for a group of entity groups that are executed in parallel. In the following description, a group of entity groups that are executed in parallel will be referred to as an entity family. Note that the fields included in the entry are merely examples and are not limiting.

[0105] The entity family ID 1501 is a field that stores identification information of an entity family. The entity group list 1502 is a field that stores identification information of entity groups that make up an entity family.

[0106] Next, the structuring processing unit 120 executes a procedure order determination process (step S1600) using the business process order determination rule information 800, the procedure order determination rule information 900, and the relationship definition information 1000. Details of the procedure order determination process will be explained using Fig. 23. The result of the procedure order determination process is stored in the process database 150 as procedure order information 1600.

[0107] The procedure order information 1600 stores entries including an order pair ID 1601, an entity group ID (previous) 1602, and an entity group ID (next) 1603. There is one entry for a pair of entity groups corresponding to the procedure that defines the order relationship. Note that the fields included in the entry are merely examples and are not limited to these.

[0108] In the first embodiment, the order of steps is expressed as the direction of an edge connecting nodes (entity groups) in the GraphML format. However, there is no limitation to the method of expressing the order of steps.

[0109] The ordered pair ID 1601 is a field that stores identification information of a pair of entity groups that define an order relationship. The entity group ID (front) 1602 is a field that stores identification information of the front-end entity group. The entity group ID (back) 1603 is a field that stores identification information of the back-end entity group.

[0110] Next, the structuring processing unit 120 executes a consistency check process (step S1700) using the parallelism determination rule information 700, the business process order determination rule information 800, the procedure order determination rule information 900, and the relationship definition information 1000. Note that the consistency check process does not necessarily have to be executed.

[0111] Specifically, the structuring processing unit 120 determines whether the information registered in the entity information 1200, parallelism information 1500, and procedure order information 1600 is consistent with the rules defined using the parallelism determination rule information 700, business process order determination rule information 800, procedure order determination rule information 900, and relationship definition information 1000. If inconsistent information exists, the structuring processing unit 120 stores the information in the process database 150 as consistency check information 1700.

[0112] The consistency check information 1700 stores entries each including a check ID 1701, a target 1702, and a rule ID 1703. One entry exists for one violation. Note that the fields included in the entry are merely an example and are not limited to these.

[0113] The confirmation ID 1701 is a field that stores the identification information of the entry. The target 1702 is a field that stores the identification information that indicates the target of the violation. For example, the identification information of an ordered pair and an entity family is stored in the target 1702. The rule ID 1703 is a field that stores the identification information of the rule that the target violates.

[0114] Next, the structuring processing unit 120 executes structured data output processing using the entity information 1200, entity pair information 1300, entity group information 1400, parallelism information 1500, procedure order information 1600, and consistency check information 1700 (step S1800). Specifically, the structuring processing unit 120 generates data representing a graph in which entity groups are nodes, as structured data, and stores the generated structured data in the structured data database 160. The structured data is, for example, data in GraphML format as shown in FIG. 18. Note that entity groups corresponding to procedures executed in parallel may be grouped together into one node.

[0115] The structured data shown in FIG. 18 includes entries that define nodes (entity groups) of the graph, entries that define the main entities of the entity groups, and entries that define the connection relationships between the nodes.

[0116] The display unit 180 of the user terminal 101 uses the structured data to display screens such as those shown in Figures 19A and 19B. The dotted line boxes represent entity groups. Icons representing procedure categories are displayed in the entity groups. Icons representing entity categories and main entities are displayed in the boxes representing entities. Note that the dashed-dotted line boxes group together procedures (entity groups) that are executed in parallel.

[0117] The structuring processor 120 determines not only the simple order between entity groups but also the parallelism of the entity groups to generate structured data. This allows for accurate structuring of business processes that include procedures performed in parallel. Furthermore, the structuring processor 120 determines the order of procedures using rules based on the main entity and rules based on procedure categories. In this way, business processes can be structured accurately using a small number of rules. Note that rules based on procedure categories are not necessarily required.

[0118] FIG. 20 is a flowchart illustrating an example of the procedure category determination process executed by the structuring processing device 100 according to the first embodiment.

[0119] The structuring processor 120 selects an entity group (step S1301). Specifically, the structuring processor 120 selects one entry from the entity group information 1400.

[0120] The structuring processor 120 acquires information on each entity included in the entity group (step S1302). Specifically, the structuring processor 120 acquires an entity category from the entity information 1200 based on the identification information registered in the entity list 1402 of the entry.

[0121] The structuring processor 120 identifies the procedure category based on the entity category of each entity included in the entity group and the procedure category determination rule information 500 (step S1303). Specifically, the structuring processor 120 performs determination on the rule set in the rule 504 of each entry, and acquires the value of the category 503 of the entry corresponding to the matched rule.

[0122] The structuring processor 120 updates the entity group information 1400 (step S1304). Specifically, the structuring processor 120 sets the identified procedure category in the category 1403 of the entry selected in step S1301.

[0123] The structuring processing unit 120 determines whether or not processing has been completed for all entries in the entity group information 1400 (step S1305).

[0124] If processing has not been completed for all entries in the entity group information 1400, the structuring processing unit 120 returns to S1301. If processing has been completed for all entries in the entity group information 1400, the structuring processing unit 120 ends the procedure category determination processing.

[0125] FIG. 21 is a flowchart illustrating an example of the main entity determination process executed by the structuring processing device 100 according to the first embodiment.

[0126] The structuring processor 120 selects an entity group (step S1401). Specifically, the structuring processor 120 selects one entry from the entity group information 1400.

[0127] The structuring processor 120 acquires information on each entity included in the entity group (step S1402). Specifically, the structuring processor 120 acquires an entity category from the entity information 1200 based on the identification information registered in the entity list 1402 of the entry.

[0128] The structuring processor 120 identifies an entity that will be the main entity based on the entity category of each entity included in the entity group and the main entity determination rule information 600 (step S1403). Specifically, the structuring processor 120 makes a determination regarding the rules set in the rule 602 of each entry, and identifies entities that match the rules.

[0129] The structuring processing unit 120 updates the entity group information 1400 (step S1404). Specifically, the structuring processing unit 120 sets the identification information of the entity identified as the main entity to the main entity ID 1404 of the entry selected in step S1401.

[0130] The structuring processing unit 120 determines whether or not processing has been completed for all entries in the entity group information 1400 (step S1405).

[0131] If processing has not been completed for all entries in the entity group information 1400, the structuring processing unit 120 returns to step S1401. If processing has been completed for all entries in the entity group information 1400, the structuring processing unit 120 ends the main entity determination processing.

[0132] FIG. 22 is a flowchart illustrating an example of the parallelism determination process executed by the structuring processing device 100 according to the first embodiment.

[0133] The structuring processing unit 120 generates a pair of entity groups (step S1501). For example, a method of generating a pair of entity groups whose main entities are located close to each other is conceivable. The present invention is not limited to the method of generating a pair of entity groups.

[0134] The structuring processing unit 120 selects a pair of entity groups (step S1502).

[0135] The structuring processing unit 120 determines whether the two procedures corresponding to the paired entity groups are performed in parallel (step S1503) based on the text, the main entities of the paired entity groups, and the parallelism determination rule information 700. For example, the determination is made based on words included in a sentence connecting the main entity of one entity group with the main entity of the other entity group.

[0136] If the two procedures are not performed in parallel, the structuring processing unit 120 proceeds to step S1505.

[0137] If the two procedures are to be executed in parallel, the structuring processor 120 assigns a flag to the pair indicating that they will be executed in parallel (step S1504), and then proceeds to step S1505.

[0138] In step S1505, the structuring processing unit 120 determines whether or not the processing has been completed for all pairs of entity groups (step S1505).

[0139] If the processing has not been completed for all pairs of entity groups, the structuring processing unit 120 returns to step S1502.

[0140] When the processing has been completed for all entity group pairs, the structuring processor 120 generates entity families based on the information of the flagged pairs (step S1506). Specifically, the structuring processor 120 generates entity families by merging pairs that include the same entity group.

[0141] The structuring processor 120 generates information about the entity family as parallelism information 1500 (step S1507) and stores it in the process database 150.

[0142] FIG. 23 is a flowchart illustrating an example of the procedure order determination process executed by the structuring processing device 100 according to the first embodiment.

[0143] The structuring processor 120 determines the order of each procedure based on the business process order determination rule information 800 (step S1601), and generates procedure order information 1600 based on the processing results (step S1602). Specifically, the structuring processor 120 determines a rough order of the procedures based on the business process order determination rule information 800. The structuring processor 120 also determines the order of each procedure based on the position of the main entity included in the entity group, etc.

[0144] The structuring processing unit 120 generates a pair of entity groups (step S1603). For example, a method of generating a pair of entity groups whose main entities are located close to each other is conceivable. The present invention is not limited to the method of generating a pair of entity groups.

[0145] The structuring processing unit 120 selects a pair of entity groups (step S1604).

[0146] The structuring processing unit 120 refers to the procedure order determination rule information 900 and the relationship definition information 1000, and determines whether or not there is a rule that matches the pair of entity groups (step S1605).

[0147] If there is no rule that matches the pair of entity groups, the structuring processor 120 proceeds to step S1607.

[0148] If a rule matching the pair of entity groups exists, the structuring processing unit 120 determines the order between the procedures corresponding to the two entity groups that make up the pair based on the order 902 of the entry corresponding to the rule in question (step S1606), and then proceeds to step S1607.

[0149] In step S1607, it is determined whether or not the processing has been completed for all pairs of entity groups (step S1607).

[0150] If the processing has not been completed for all pairs of entity groups, the structuring processing unit 120 returns to step S1604.

[0151] When the processing has been completed for all entity group pairs, the structuring processing unit 120 determines the order of the procedures based on the determination results for the entity group pairs (step S1608).

[0152] The structuring processing unit 120 updates the procedure order information 1600 based on the processing result of step S1608 (step S1609).

[0153] The structuring processing device 100 does not need to store the business process order determination rule information 800. In this case, the procedure order is not determined using the business process order determination rule information 800, and therefore the procedure category determination process can be omitted. The structuring processing device 100 only needs to determine the procedure order based on the procedure order determination rule information 900 and the relationship definition information 1000.

[0154] As described above, the structuring processing device 100 of the first embodiment can generate structured data with high accuracy from a document describing a business process. The rules for determining the order of procedures are limited to rules based on the relationships between main entities and rules based on the order of procedures and the relationships between procedure categories, thereby reducing the cost required for setting rules.

[0155] The procedure category and the main entity may be determined without using rules. For example, a method of determining the procedure category and the main entity may be used that uses a model generated by a learning process.

[0156] The order of steps does not have to be determined using rules. For example, the order may be determined using a model generated by a learning process using words between main entities and a model generated by a learning process using data indicating the relationship between the order of steps and procedure categories. Alternatively, the order may be determined by combining rules and models.

[0157] It should be noted that rules using sub-entities may also be set. [Example]

[0158] In this example, a process for supplementing information omitted in a document using the structured data generated in Example 1 will be described. For example, in the example shown in Figures 19A and 19B, a node corresponds to an entity family, but in this example, a node may correspond to an entity.

[0159] In the description of this embodiment, differences from the first embodiment will be mainly described, and descriptions of commonalities with the first embodiment will be omitted or simplified.

[0160] FIG. 25 is a diagram illustrating an example of the overall flow of the second embodiment.

[0161] The structuring processing unit 2550 in the structuring processing device 2500 of the second embodiment executes the structured data generation process (step S2501) in the first embodiment to generate structured data (see FIG. 18) from the document data 2550. The structured data is stored in the structured data database 160.

[0162] Next, structuring processing unit 2550 executes taxonomy generation processing (step S2502) by referencing the structured data in structured data database 160, thereby generating taxonomy data 2520. Structuring processing unit 2550 generates integrated graph data by graph integration processing (step S2503) using the structured data in structured data database 160 and the generated taxonomy data 2520. The generated integrated graph data is stored in integrated graph database 2510.

[0163] The structuring processing unit 2550 converts the integrated graph data in the integrated graph database 2510 into generalized graph data 2530 by graph generalization processing (step S2504). In graph update processing (step S2505), the structuring processing unit 2550 matches the generalized graph represented by the generalized graph data 2530 with the graph pattern represented by the graph pattern database 2540 to identify information omitted in the document represented by the document data 2550 and its supplementary content, and updates the structured data.

[0164] The integrated graph database 2510 and the graph pattern database 2540 may be stored in a memory device (e.g., at least a part of the main memory device 202 and the auxiliary memory device 203) of one or more computers 200 based on the structure processing device 2500, similar to the above-mentioned databases 130, 140, 150, and 160. In addition, the generated taxonomy data 2520 and generalized graph data 2530 may be stored in the processing database 150.

[0165] Furthermore, a machine learning system 2590 may be provided inside or outside the structuring processing device 2500. The machine learning system 2590 may accept input of information (e.g., input of information representing a result) and output information (e.g., information representing a means for obtaining the result (e.g., a process, a recipe, etc.)) by inputting the information into a machine learning model. The machine learning model may be a neural network or other model. The training data for the machine learning model includes structured data. In this embodiment, since the training data includes structured data (updated structured data) that has been complemented with the structured data generated in Example 1, machine learning with higher accuracy of the machine learning model is expected. The machine learning system 2590 may be a functional unit or a computer system based on one or more computers.

[0166] FIG. 26 is a diagram illustrating an example of taxonomy data 2520 in the second embodiment.

[0167] The taxonomy data 2520 is composed of node taxonomy data 2600 that manages the correspondence between each node in the structured data and a taxonomy, and meta taxonomy data 2610 that manages the correspondence between taxonomies.

[0168] The node taxonomy data 2600 stores entries each including a structured data ID 2601, a node name 2602, a word 2603, and a label 2604. One entry exists for one expression (entity). Note that the fields included in the entry are merely examples and are not intended to be limiting.

[0169] Structured data ID 2601 is a field that stores identification information of structured data. Node name 2602 is a field that stores information that indicates an entity in the structured data. Word 2603 is a field that stores information that indicates the individual words that make up the entity in node name 2602. Label 2604 is a field that stores information that indicates the label taxonomy of the entity.

[0170] For example, according to the first entry of the node taxonomy data 2600, it can be seen that the “Substance 1” entity of the first structured data is composed of “substance” and “1” as a word taxonomy, and belongs to “substance name” as a label taxonomy.

[0171] Next, the meta taxonomy data 2610 stores entries including a node taxonomy 2611 and a relationship taxonomy 2612. One entry exists for one node taxonomy set. Note that the fields included in the entry are merely examples and are not limiting. A "node taxonomy set" is one or more node taxonomies. A "node taxonomy" is a word taxonomy or a label taxonomy.

[0172] The node taxonomy 211 is a field that stores information indicating one or more word taxonomies and / or one or more label taxonomies. The relationship taxonomy 2612 is a field that stores information indicating a relationship taxonomy assigned to a node taxonomy set. The information indicating the relationship taxonomy corresponds to meta information of the node taxonomy set.

[0173] For example, according to the first entry in the metataxonomy data 2610, the label taxonomy "Substance Name" as a node taxonomy is assigned the relationship taxonomy "Main." In other words, it can be seen that the entity belonging to the label taxonomy "Substance Name" is the main entity. Note that the relationship taxonomy "Sub" refers to a subordinate entity that is subordinate to the main entity. A subordinate entity may be synonymous with the aforementioned secondary entity.

[0174] The taxonomy data 2520 is data generated from structured data. For example, one or more pieces of structured data may exist for each piece of literature data 2550. The taxonomy data 2520 may be generated based on the structured data of one or more pieces of literature.

[0175] Furthermore, the structured data used in Example 2 may be structured data prepared by a method different from the method described in Example 1. Furthermore, although the document data 2550 is an example of document data, the document data may be data in a format other than text, for example, data in a table format.

[0176] FIG. 27 is a diagram illustrating an example of an integrated graph indicated by integrated graph data in the second embodiment.

[0177] Each of the node taxonomy and the relationship taxonomy is represented as a graph node. Hereinafter, a node representing a node taxonomy will be referred to as a "node taxonomy node," and a node representing a relationship taxonomy will be referred to as a "relationship taxonomy node."

[0178] The integrated graph connects entities (nodes) in the graph represented by the structured data with node taxonomy nodes to which the entities match, and also connects each node taxonomy node with an edge to a relationship taxonomy node corresponding to the node taxonomy node. A "node taxonomy node to which an entity matches" is a node representing a word obtained from the entity (word taxonomy node) or a node representing a label assigned to the entity (label taxonomy node). A "relationship taxonomy node corresponding to a node taxonomy node" is a relationship taxonomy node representing a relationship taxonomy identified from the metataxonomy data 2610 as a relationship taxonomy corresponding to the node taxonomy represented by the node taxonomy node.

[0179] For example, since "Substance 1" in the structured data belongs to "Substance Name", the "Substance 1 node" and the "Substance Name node" are connected.

[0180] 27, in the integrated graph, the edges connecting entities (nodes) in the graph represented by the structured data to node taxonomy nodes, and the edges connecting node taxonomy nodes to relationship taxonomy nodes, may be directed or undirected. Also, in the integrated graph illustrated in FIG. 27, some edges (for example, the edge connecting the "silicon-based composition" node to the "substance name" node taxonomy node) are omitted from the illustration.

[0181] FIG. 28 is a diagram illustrating an example of processing list data according to the second embodiment.

[0182] The process list data 2800 stores entries including a process 2801 and a generalization node 2802. One entry exists for one complement process. Note that the fields included in the entry are merely an example and are not limited to this.

[0183] The process 2801 is a field for storing information indicating the completion process of the structured data. The generalization node 2802 is a field for storing information indicating the taxonomy (the taxonomy represented by a node in the generalization graph) used when generalizing the structured data.

[0184] For example, the first entry shows that the generalized graph generated by applying the "node separation" completion process to the structured data includes a "label taxonomy" node. Also, the second entry shows that the generalized graph generated by applying the "node duplication" completion process to the structured data includes a "label taxonomy" node and a "process taxonomy" node.

[0185] FIG. 29 is a diagram illustrating an example of a generalized graph indicated by the generalized graph data 2530 in the second embodiment.

[0186] A generalization graph is a structured data with each node replaced by a taxonomy.

[0187] 29, the generalized graph is one in which each entity node (node ​​representing an entity) in the graph indicated by the structured data with structured data ID "1" has been replaced with a taxonomy node (node ​​representing a taxonomy). The taxonomy represented by the node after replacement is the taxonomy identified from the generalized node 2802 corresponding to the complementation process including the replacement (the taxonomy identified from the process list data 2800).

[0188] FIG. 30 is a diagram illustrating an example of a graph pattern indicated by the graph pattern database 2540 in the second embodiment.

[0189] A graph pattern is composed of a matching graph for identifying omissions and a complement graph for specifying a complement method. Specifically, in a graph pattern, a "matching graph" is all or part of a generalized graph before complementation, and is a graph pattern that is the target of complementation according to the graph pattern. Also, in a graph pattern, a "complementary graph" is a graph pattern after complementation according to the graph pattern is applied to the matching graph (i.e., the graph after complementation).

[0190] For example, the graph pattern following the example in Figure 30 shows that if the same label for the same substance name is used in multiple processes, as shown in the example of the generalized graph in Figure 29, in order to complement the structured data, it is sufficient to separate the graph containing the substance name node into graphs containing nodes for that substance name for each process.

[0191] FIG. 31 is a flowchart illustrating an example of the graph integration process (step S2503 in FIG. 25) executed by the structuring processing device 2500 according to the second embodiment.

[0192] The structuring processing unit 120 acquires taxonomy data 2520 (step S1701) and structured data (step S1702). The taxonomy data 2520 may be generated and acquired in this graph integration process, or may be generated by the structuring processing unit 120 based on the structured data of one or more documents including the document corresponding to the structured data when the structured data is generated, and stored in the structured data database 160.

[0193] Next, the structuring processing unit 120 extracts entity names and graph structure information included in the process information (at least a part of the information in the structured data) (step S1703), and generates a label taxonomy (step S1704), a word taxonomy (step S1705), and a relationship taxonomy (step S1706) based on the taxonomy data 2520. Here, the process information is information representing entity groups and connection relationships between the entity groups, as in the examples shown in Figures 19A and 19B.

[0194] The structuring processing unit 120 generates an integrated graph including the generated taxonomies as nodes, and outputs integrated graph data indicating the integrated graph (step S1707). The output data is stored in the integrated graph database 2510. In the integrated graph, a node indicated by the structured data is connected to a node taxonomy node that matches the node, and a relationship taxonomy node that matches the node is connected to the node taxonomy node.

[0195] FIG. 32 is a flowchart illustrating an example of the graph generalization process (step S2504 in FIG. 25) executed by the structuring processing device 2500 according to the second embodiment.

[0196] The structuring processing unit 120 acquires integrated graph data (step S1801) and process list data 2800 (step S1802).

[0197] Next, the structuring processing unit 120 selects one entry from the processing list data 2800 (step S1803). Steps S1804 to S1806 are performed for the selected entry.

[0198] That is, the structuring processing unit 120 identifies the taxonomy represented by the generalized node 2802 of the selected entry (step S1804), updates the node represented by the structured data to a node representing the taxonomy (step S1805), and registers the generalized graph data 2530 in which the node represented by the structured data has been replaced with the taxonomy node in the generalized graph list (step S1806). The generalized graph data 2530 may be registered in, for example, a generalized graph list (not shown).

[0199] Thereafter, the structuring processing unit 120 determines whether or not there is an unselected entry in the processing list data 2800 (step S1807). If there is an unselected entry, the process returns to step S1803. If there is no unselected entry, the structuring processing unit 120 outputs the generated generalized graph list (S1808). The generalized graph list may be stored in the processing database 150.

[0200] In this way, one integrated graph is generated for each graph represented by one or more structured data extracted from one document, and from this integrated graph, a generalized graph corresponding to each completion process is generated.

[0201] FIG. 33 is a flowchart illustrating an example of the graph update process (step S2505 in FIG. 25) executed by the structuring processing device 2500 according to the second embodiment.

[0202] The structuring processing unit 120 acquires a generalized graph list (step S1901) and refers to the graph pattern database 2540 (step S1902).

[0203] Next, the structuring processing unit 120 selects one generalized graph from the generalized graph list (step S1903). Steps S1904 to S1907 are performed on the selected generalized graph.

[0204] That is, the structuring processing unit 120 calculates the similarity between the graph structure of the matching graph in the graph pattern for completion processing corresponding to the selected generalized graph (the graph pattern represented by the graph pattern database 2540) and the graph structure of the selected generalized graph, and determines whether or not to select the graph pattern based on the calculated similarity (step S1904). This determination may be made by determining whether or not the similarity is equal to or greater than a similarity threshold. A GNN (Graph Neural Network) such as GraphSAGE may be used as a method for comparing the similarity of graph structures. If the result of this determination is false, the process proceeds to step S1908. Note that if there are multiple graph patterns for completion processing corresponding to the selected generalized graph, the determination in step S1904 may be made for each graph pattern.

[0205] If the result of this determination is true, the structuring processing unit 120 performs the "processing" represented by the selected graph pattern for each selected graph pattern. That is, the structuring processing unit 120 performs node update processing if the graph pattern is a node update pattern (step S1906), and performs edge update processing if the graph pattern is an edge update pattern (step S1907). Note that, depending on the selected graph pattern, both steps S1906 and S1907 may be performed on the selected generalized graph.

[0206] Thereafter, the structuring processing unit 120 determines whether or not there is an unselected generalized graph in the generalized graph list (step S1908). If there is an unselected generalized graph, the process returns to step S1903. If there is no unselected generalized graph, the structuring processing unit 120 applies, for each generalized graph for which at least one graph pattern has been selected, one or more updated generalized graphs obtained based on the generalized graph to the structured data (the structured data selected in the graph integration process). In other words, the structured data is updated. The structuring processing unit 120 outputs the updated structured data (step S1909). The output updated structured data is stored in the structured data database 160. Updated structured data may exist for each updated generalized graph (or may be structured data to which the updated generalized graph has been applied), or multiple updated generalized graphs may be applied to one piece of structured data.

[0207] 34A to 34H are diagrams showing examples of a plurality of graph patterns (a plurality of graph patterns represented by the graph pattern database 2540) in the second embodiment.

[0208] A graph pattern consists of two parts: a "condition" (matching graph) and a "process" (complementary graph). If the generalized graph matches the graph structure pattern of the "condition" (matching graph) (for example, if the similarity between all or part of the generalized graph and the matching graph satisfies the similarity condition), in step S1906 or step S1906, the corresponding graph in the generalized graph is replaced with the graph structure of the "process" (complementary graph). In this way, the generalized graph is updated, and the structured data can be updated by applying the updated generalized graph to the structured data. The "corresponding graph" referred to in this paragraph refers to all or part of the generalized graph.

[0209] Figure 34A shows an example of a separation pattern. The "condition" of the separation pattern is that the same label node (e.g., "state" node) of the same "substance" node is connected to the "operation" nodes of multiple processes. The "process" of the separation pattern is to separate a graph containing "substance" nodes into graphs containing "substance" nodes for each process (after separation, each "substance" node is connected to the "operation" node of one process).

[0210] FIG. 34B shows an example of replication pattern 1. The "condition" for replication pattern 1 is that there is a direct connection (a connection not via another node) from a "substance" node to another "substance" node, and an "operation" node is missing. For convenience, the process corresponding to the node set in which a node is missing is called the "missing process." The "process" for replication pattern 1 is to duplicate the "operation" node of the process preceding the missing process (the process immediately preceding the missing process) between the "substance" node and the "substance" node (i.e., to complement the "operation" node, which is an example of a missing node, in the node set corresponding to the missing process). In the example shown in FIG. 34B, a node with the node name "measurement" belonging to "operation" is complemented as the "operation" node.

[0211] Figure 34C shows an example of replication pattern 2. The "condition" of replication pattern 2 is that a "substance" node is missing before the "operation" node, i.e., the first node of the node set corresponding to the missing process is an "operation" node. The "processing" of replication pattern 2 is to duplicate the "substance" node of the process preceding the missing process before the "operation" node (first node) of the missing process, depending on the label of the "operation" node. A specific example of the "processing" is as follows. If the "operation" corresponding to the head node is "measurement", the "substance" node that is duplicated before the head node of the missing process is the "substance" node immediately before the "operation" node in the previous process. Note that the node immediately before node X means the parent node of node X, specifically, the node to which the base end of the directed edge connected to node X is connected. If the "operation" corresponding to the head node is "composite", the "substance" node that is duplicated before the head node of the missing process is the "substance" node (child node of the "operation" node) immediately after the "operation" node in the previous process. Note that the node immediately after node X means the child node of node X, specifically, the node connected to the tip of the directed edge connected to node X. Also, "composite" is the name of a node that belongs to "operation".

[0212] Figure 34D shows an example of addition pattern 1. The "condition" of addition pattern 1 is that a "substance" node (new "substance" node) different from the "substance" node immediately preceding the first "mixing" node exists as the immediately preceding node of the second "mixing" node immediately following the first "mixing" node. The "process" of addition pattern 1 is to add an "addition" node between the new "substance" node and the second "mixing" node. Both "mixing" and "addition" are node names that belong to "operation."

[0213] Figure 34E shows an example of addition pattern 2. The "condition" of addition pattern 2 is that the "substance" node immediately following the "operation" node does not have a node expressing a "state" corresponding to the "operation." The "processing" of addition pattern 2 is to add a node expressing the "state" as a nominalization of the operation as a "state" node immediately following the "substance" node. Here, nominalization means, for example, if the "operation" is "hardening," then the "state" is "hardened material." Because a "state" node is added, addition pattern 2 corresponds to the node update pattern. "Hardening" is the node name of the "operation."

[0214] Figure 34F shows an example of Integration Pattern 1. The "condition" for Integration Pattern 1 is that a combination of word taxonomies connected to the same label taxonomy node by a "distinction" related taxonomy is assigned. Here, "a combination of word taxonomies connected to a "distinction" related taxonomy" refers to multiple word taxonomies represented by the same entry's node taxonomy 2611 (see Figure 26) connected to multiple nodes of the same label taxonomy. If this "condition" is met, the "process" of Integration Pattern 1 is to integrate combinations that do not have a "distinction" related taxonomy assigned and duplicate the word taxonomy node in the same label node. For example, in the example of Figure 34F, there are three substances: a silicone gel composition, a silicone gel sheet, and a cured sheet. A "distinction" related taxonomy is assigned between the word taxonomies for "composition" and "sheet." Therefore, the structuring processing unit 2550 considers silicone gel sheets and cured sheets, each of which contains the word "sheet," to be the same thing, and considers silicone gel sheets and cured sheets to be separate silicone gel compositions. Therefore, according to integration pattern 1, the word taxonomy node for "silicone gel" is duplicated in the label taxonomy node to which "cured" and "sheet" belong. This indicates that "silicone gel" and "cured" can be used interchangeably.

[0215] Figure 34G shows an example of integration pattern 2. The "condition" of integration pattern 2 is that the same word taxonomy is assigned to the same label taxonomy node, and that the "integration" relationship taxonomy is assigned to it. The "same word taxonomy" here also means that multiple word taxonomies represented by the node taxonomy 2611 (see Figure 26) of the same entry are treated as the same word taxonomy. The "processing" of integration pattern 2 is to consider multiple label taxonomy nodes to which the same word taxonomy node assigned the "integration" relationship taxonomy belongs as the same, and to integrate them as a continuous operation. For example, in the example of Figure 34G, the "Substance" node after the operation of the first process is connected to the word taxonomy "silicone gel" and the state label "processed smoothly," and the "Substance" node before the operation of the second process is connected to the word taxonomy "silicone gel" and the state label "after processing," and furthermore, the state labels "processed smoothly" and "after processing" are assigned the relationship taxonomy of "integration," so by integrating these two "Substance" nodes, the two processes are integrated into one.

[0216] Figure 34H shows an example of an edge rewiring pattern. The "condition" of the edge rewiring pattern is that an edge originates from a "sub" relationship taxonomy node (the base end of the edge is connected to the "sub" relationship taxonomy node). The "process" of the edge rewiring pattern is to rewire the base end of the edge to the "master" relationship taxonomy node to which the "sub" relationship taxonomy node belongs.

[0217] As described above, the structuring processing device 2500 of the second embodiment can supplement information omitted in a document describing a business process from other documents without creating rules for linguistic knowledge. Furthermore, by expressing graph patterns as knowledge graphs, it is possible to use both automatic pattern generation by AI and manual rule description. This can reduce the amount of work required for extracting and structuring process information.

[0218] Note that instead of a matching graph, a graph pattern may employ a feature vector of a GNN that has previously matched. For example, the structuring processing unit 2550 may associate a feature vector of a GNN that has previously matched with each graph pattern. For each graph pattern, the structuring processing unit 2550 may determine whether the generalized graph matches the graph pattern based on the similarity (cosine distance of the vector) between the feature vector of the generalized graph (a feature vector generated from the generalized graph using the GNN) and the feature vector associated with the graph pattern. [Example]

[0219] In this example, a process is described in which, using information on the integrated graph generated in Example 2, information on Example Y (e.g., Y=2) in which only differences from Example X (e.g., X=1) are described in the document represented by the document data 2550 is complemented. Note that in this example, differences from Example 2 are mainly described, and descriptions of commonalities with Example 2 are omitted or simplified. Furthermore, in order to avoid confusion between the examples described in this specification and the examples described in the document represented by the document data 2550, in Example 3 (and Example 4), the examples described in the document may be referred to as "document examples."

[0220] FIG. 35 is a diagram showing the overall flow of the third embodiment.

[0221] The structuring processing unit 3550 in the structuring processing device 3500 of the third embodiment performs graph generalization processing (step S3500) shown in FIG. 35 instead of the graph generalization processing (step S2504) and graph update processing (step S2505) shown in FIG. 25. As the structured data in the second embodiment, replaced structured data 3510 is stored in the structured data database 160. The structuring processing unit 2550 uses the generated integrated graph data and the master-slave relationship table 3520 to identify information omitted in embodiment Y in the document and its supplementary content, and updates the structured data. As a result, updated structured data 3530 is generated. The structuring processing unit 3550 stores the updated structured data 3530 in the structured data database 160. The master-slave relationship table 3520 may be stored in a storage device of the computer 200.

[0222] FIG. 36 is a diagram illustrating an example of the master-slave relationship table 3520 according to the third embodiment.

[0223] The master-slave relationship table 3520 stores entries each including a master example 3601 and a slave example 3602. There is one entry for each pair of literature examples. Note that the fields included in the entry are merely examples and are not limiting.

[0224] Main example 3601 is a field that stores information indicating main example X. Subordinate example 3602 is a field that stores information indicating subordinate example Y. For example, the first entry indicates that literature example 2 is a subordinate example of literature example 1. The superior-subordinate relationship of literature examples may be a tree structure.

[0225] FIG. 37 is a diagram showing an example of replaced structured data 3510 in the third embodiment.

[0226] The replacement structured data 3510 is graph data. The replacement structured data 3510 may be data for each subordinate embodiment. In the graph represented by the replacement structured data 3510, nodes correspond to entities in a document, and the nodes are assigned label taxonomy such as "embodiment," "replacement target," or "replacement source." When there is a "replacement target" node taxonomy node and a "replacement source" node taxonomy node to which the "replacement target" node taxonomy node is directly connected, the entity represented by the "replacement source" node taxonomy node (an entity in the main embodiment of the subordinate embodiment) is replaced in the subordinate embodiment by the entity represented by the "replacement target" node taxonomy node.

[0227] 37, "triethoxychlorosilane" in main example 1 is replaced with "diethoxydichlorosilane" in subexample 2. Except for this, subexample 2 is similar to main example 1.

[0228] FIG. 38 is a flowchart illustrating an example of the graph generalization process (step S3500 in FIG. 35) executed by the structuring processing device 3500 according to the third embodiment.

[0229] The structuring processing unit 3550 acquires integrated graph data (step S2001) and acquires a master-slave relationship table (step S2002).

[0230] Next, the structuring processing unit 3550 selects one entry from the master-slave relationship table 3520 (step S2003). Steps S2004 to S2006 are performed for the selected entry.

[0231] That is, structuring processing unit 3550 searches structured data database 160 for structured data of the main embodiment represented by main embodiment 3601 of the entry, and duplicates the found structured data (step S2004). Structuration processing unit 3550 searches the duplicated structured data for a "source of replacement" node (a node to which "source of replacement" is attached) in replacement structured data 3510 corresponding to the subordinate embodiment represented by subordinate embodiment 3602 of the entry (step S2005). Structuration processing unit 3550 replaces the found node in the duplicated structured data with the same node as the "destination of replacement" node (a node to which "source of replacement" is attached) in replacement structured data 3510 (step S2006).

[0232] Thereafter, the structuring processing unit 3550 determines whether or not there are any unselected entries in the master-slave relationship table 3520 (step S2007). If there are any unselected entries, the process returns to step S2003. If there are no unselected records, the updated replicated structured data is output (S2007).

[0233] FIG. 39 is a diagram showing a specific example of the graph generalization process of FIG.

[0234] Structured data 3901 of main example 1 is duplicated, and the "source" node in the graph represented by the duplicated structured data is replaced with the "destination" node, thereby generating structured data 3902 of dependent example 2. Specifically, based on the substituted structured data 3510 illustrated in Figure 37, the "triethoxychlorosilane" node represented by the duplicated structured data is replaced with the "diethoxydichlorosilane" node, thereby obtaining structured data 3902 of example 2. [Example]

[0235] In this example, a process of name matching will be described using information on the integrated graph generated in Example 2 and information on the taxonomy. Note that in this example, differences from Example 2 will be mainly described, and descriptions of points in common with Example 2 will be omitted or simplified.

[0236] FIG. 40 is a diagram showing the overall flow of the fourth embodiment.

[0237] In addition to the taxonomy data 2520, name identification taxonomy data 4010 is prepared. The structuring processing unit 4550 in the structuring processing device 4500 of the fourth embodiment performs graph generalization processing (step S4000) shown in FIG. 40 instead of the graph generalization processing (step S2504) and graph update processing (step S2505) shown in FIG. 25. In the graph generalization processing (step S4000), the structuring processing unit 4550 updates the structured data using the generated integrated graph data and the name identification relationship table 4020. As a result, updated structured data 4030 is generated. The structuring processing unit 4550 stores the updated structured data 4030 in the structured data database 160. The name identification taxonomy data 4010 and the name identification relationship table 4020 may be stored in a storage device of the computer 200.

[0238] FIG. 41 is a diagram illustrating an example of the name identification relationship table 4020 according to the fourth embodiment.

[0239] The name identification relationship table 4020 stores entries including structured data 4101 and name identification taxonomy 4102. One entry exists for one piece of structured data. Note that the fields included in the entry are merely examples and are not limited to these.

[0240] Structured data 4101 is a field for storing information indicating an element (e.g., literature example) corresponding to the structured data. Matching taxonomy 4102 is a field for storing information indicating the taxonomy after matching. For example, the first entry indicates that the structured data of literature example 1 is matched using the matching taxonomy of "compound classification."

[0241] FIG. 42 is a diagram illustrating an example of the name merged taxonomy data 4010 according to the fourth embodiment.

[0242] The merged taxonomy data 4010 is graph data. The merged taxonomy data 4010 may exist for each merged taxonomy. The top node of the graph may be the node of the merged taxonomy, and the child nodes of the top node may be nodes representing the entities before merger. For example, the example in FIG. 42 is merged taxonomy data for "compound classification," and indicates that "silicon compounds" include "diethoxydichlorosilane" and "triethoxychlorosilane."

[0243] FIG. 43 is a flowchart illustrating an example of graph generalization processing executed by the structuring processing device 4500 according to the fourth embodiment.

[0244] The structuring processing unit 4550 acquires integrated graph data (step S2101) and acquires the name identification relationship table 4020 (step S2102).

[0245] Next, the structuring processing unit 4550 selects one entry from the name identification relationship table 4020 (step S2103). Steps S2104 to S2106 are performed for the selected entry.

[0246] That is, the structuring processing unit 4550 searches the structured data database 160 for structured data corresponding to the structured data 4101 of the entry, and duplicates the found structured data (step S2104). The structuring processing unit 4550 identifies the name identification taxonomy data 4010 corresponding to the name identification taxonomy 4102 of the entry, and searches the duplicated structured data for a node that matches a child node (a node representing an entity to be identified) represented by the identified name identification taxonomy data 4010 (step S2105). The structuring processing unit 4550 replaces the found node with a parent node (a node representing the taxonomy after identification) represented by the name identification taxonomy data 4010 (step S2106).

[0247] Thereafter, the structuring processing unit 4550 determines whether or not there are any unselected entries in the name identification relationship table 4020 (step S2107). If there are any unselected entries, the process returns to S2103. If there are no unselected records, the structuring processing unit 4550 outputs the updated duplicated structured data (step S2108).

[0248] FIG. 44 is a diagram showing a specific example of the graph generalization process of FIG.

[0249] Structured data 4401 of Example 1 is duplicated, and entity nodes (nodes representing entities to be merged) in the graph represented by the duplicated structured data are replaced with merged taxonomy nodes (nodes of the taxonomy after merged), thereby generating structured data 4402. Specifically, based on the merged taxonomy data 4010 illustrated in Figure 442, the node for "triethoxychlorosilane" represented by the duplicated structured data is replaced with the node for "silicon compound," thereby obtaining structured data 4402.

[0250] The present invention is not limited to the above-described embodiments, but includes various modifications. For example, the above-described embodiments are provided to explain the present invention in detail, and the present invention is not necessarily limited to those including all of the described configurations. Furthermore, some of the configurations of each embodiment can be added to, deleted from, or replaced with other configurations.

[0251] Furthermore, the above-described configurations, functions, processing units, processing means, etc. may be partially or entirely implemented in hardware, for example, by designing them as integrated circuits. The present invention can also be realized by software program code that implements the functions of the embodiments. In this case, a storage medium on which the program code is recorded is provided to a computer, and a processor included in the computer reads the program code stored in the storage medium. In this case, the program code itself read from the storage medium implements the functions of the above-described embodiments, and the program code itself and the storage medium on which it is stored constitute the present invention. Examples of storage media for providing such program code include flexible disks, CD-ROMs, DVD-ROMs, hard disks, solid-state drives (SSDs), optical disks, magneto-optical disks, CD-Rs, magnetic tapes, non-volatile memory cards, and ROMs.

[0252] Furthermore, the program code that realizes the functions described in this embodiment can be implemented in a wide range of program or script languages, such as assembler, C / C++, perl, Shell, PHP, Python, and Java (registered trademark).

[0253] Furthermore, the program code of the software that realizes the functions of the embodiments may be distributed via a network and stored in a storage means such as a computer's hard disk or memory, or in a storage medium such as a CD-RW or CD-R, and the processor of the computer may read and execute the program code stored in the storage means or storage medium.

[0254] In the above-described embodiment, the control lines and information lines are those that are considered necessary for the explanation, and not all control lines and information lines are necessarily shown in the product. All components may be interconnected.

[0255] The above description can be summarized, for example, as follows: The following summary may include supplementary explanations and explanations of variations of the above description.

[0256] A data processing device (e.g., a structuring processing device) includes a storage device (e.g., a main storage device 202 and an auxiliary storage device 203) in which structured data for a document representing a plurality of procedures is stored, and an arithmetic device (e.g., an arithmetic device 201) that performs a structuring update process, which is a process for generating updated structured data. The arithmetic device is connected to the storage device.

[0257] The structured data is graph data representing a graph including multiple entity nodes and one or multiple edges. Each of the multiple entity nodes is a node representing an entity in the document. The structured update process includes updating the structured graph based on update definition data, which is data that defines updates to at least one node and edge using a taxonomy, and the taxonomy of at least one entity node in the structured graph, which is a graph represented by the structured data or a copy thereof. The updated structured data is data representing the graph after the structured graph has been updated.

[0258] This makes it possible to generate accurate structured data for documents that represent multiple procedures. For example, by using structured data or update definition data for multiple documents, it is expected that a graph for one document can be complemented with a graph for another document. Furthermore, a large amount of training data is not required to generate accurate updated structured data. Furthermore, because updated structured data is generated based on update definition data that defines updates to at least one node and edge using a taxonomy expression, there is no need to prepare language-dependent rules. Therefore, it is expected that accurate structured data can be generated even if the descriptions of consecutive procedures are discrete in the document.

[0259] The structured data for a document may be data for a single document, data for a portion of a document, or data for a set of documents (multiple documents). In other words, the unit for which structured data is prepared is not limited to a single document. Furthermore, the format for expressing multiple steps in a document may be other than or in addition to a text format.

[0260] Furthermore, the updated structured data may be used as training data for machine learning. The graph represented by the updated structured data may include multiple nodes and multiple edges representing multiple procedures and the results of those multiple procedures, and such updated structured data may be used as training data. Because the updated structured data is accurate, improvement in the accuracy of machine learning (the learning accuracy of the machine learning model) is expected. Updated structured data modified by a user may also be used as training data. In other words, instead of the user presenting the "conditions" of a graph pattern, the user may present "processes" and learn patterns of "conditions" common to the presented "processes."

[0261] Furthermore, each example of the multiple procedures may be a "process" in the above-described embodiment. That is, the multiple procedures may be multiple processes, and the multiple processes may constitute a "business process." In the industrial field, a "business" may include a recommendation of an operating procedure for a device or a recommendation of a process for dealing with a device failure. In the medical field, a "business" may include diagnosis, treatment, and medication. In the materials field, a "business" may include a recommendation of a synthesis process for a new material.

[0262] The taxonomy may be either an entity taxonomy (e.g., a label taxonomy or a word taxonomy), which is a taxonomy of entities, or a relationship taxonomy, which is a taxonomy of entity taxonomies. The update definition data may include graph pattern data (e.g., a graph pattern database 2540) representing one or more graph patterns. Each of the one or more graph patterns may include a pattern condition (e.g., a "condition" of the graph pattern) that is a graph structure corresponding to the graph pattern or a summary thereof, and a completion process for the graph structure (e.g., a "process" of the graph pattern). The summary of the graph structure may be, for example, a feature vector of the graph structure or other types of features.

[0263] The structural update process may include a graph generalization process (for example, step S2504) and a graph update process (for example, step S2505).

[0264] The graph generalization process may include converting the graph represented by the structured data into a generalized graph including a plurality of generalized nodes and one or more edges, each of which may be a node representing an entity taxonomy or relationship taxonomy of the entity represented by the entity node, or a duplicate of the entity node.

[0265] The graph update process may generate updated structured data by updating the generalized graph by performing a complement process on the generalized graph in a graph pattern including a pattern condition to which the generalized graph conforms, and by changing the graph structure of the structured graph (the structure of a part or all of the structured graph) to match the graph structure of the updated generalized graph. For example, the connection relationships between entity nodes in the graph structure of the structured graph may be changed to match the connection relationships between the generalized graph nodes corresponding to those entity nodes.

[0266] Graph pattern data is data used to update a generalized graph. Therefore, the configuration of graph pattern data is expected to be simpler than graph pattern data (or similar data) for updating a structured graph without a generalized graph. Therefore, it is expected that updated structured data with high accuracy will be generated.

[0267] The structure update process may include a graph integration process (e.g., step S2503) that generates an integrated graph in which one or more taxonomy nodes are associated with the structured graph. For each of multiple entity nodes represented by the structured graph, if there are one or more taxonomies for the entity represented by the entity node, the one or more taxonomy nodes may be associated in the integrated graph. Each taxonomy node may be a node representing a taxonomy (e.g., a node taxonomy node or a relationship taxonomy node). After the graph integration process, the computing device may perform a graph generalization process. For each entity node, a generalization node in the generalization graph may correspond to the taxonomy node associated with the entity node in the integrated graph. This is expected to enable efficient generation of the generalization graph. For example, for an entity that has an associated entity taxonomy and a relationship taxonomy, in the integrated graph, a child node of the entity node for that entity may be a taxonomy node of the entity taxonomy, and a child node of the taxonomy node may be a taxonomy node of the relationship taxonomy for that entity. The graph structure of the generalization graph may be based on the graph structure of the integrated graph.

[0268] There may be multiple types of taxonomies for at least one of the entity taxonomy and the relationship taxonomy. For example, the entity taxonomy may include taxonomies such as "substance name," "state," and "operation." The relationship taxonomy may include taxonomies such as "master," "subordinate," "distinction," and "integration." The type of taxonomy to be used may be defined for each of multiple types of completion processes related to the generalization graph. For example, data representing such definitions (e.g., process list data 2800) may be prepared. The graph generalization process may include, for each of multiple types of completion processes, converting the structured graph into a generalization graph including a generalization node representing a taxonomy belonging to the type corresponding to the completion process. This allows an appropriate generalization graph to be prepared for each completion process according to the graph structure of the structured graph, thereby expected to generate accurate updated structured data. Furthermore, for each of the multiple graph patterns, the graph pattern may include any one of multiple types of complement processing, and the multiple types of complement processing may include two or more of the following types of complement processing. · Separation process to separate generalized graphs. A duplication process that uses a duplicate of a generalization node corresponding to a previous procedure of a certain procedure as a parent node or child node of the generalization node corresponding to the certain procedure. (The "generalization node corresponding to a certain procedure" may be, for example, one of the two "substance name" nodes corresponding to the missing process illustrated in FIG. 34B, or the "measurement" node corresponding to the missing process illustrated in FIG. 34C.) An addition process that adds a new generalized node as a parent node or a child node of a generalized node. (The "new generalized node" may be, for example, the "add" node illustrated in FIG. 34D.) An integration process that integrates different generalization nodes or their child generalization nodes that are associated with the same classification metataxonomy. (The "same classification metataxonomy" may be, for example, "Differentiation" as illustrated in FIG. 34F or "Integration" as illustrated in FIG. 34G. The "integration of different generalization nodes" may be, for example, merging the "Substance Name" node (whose parent node is an "Operation" node) illustrated in the upper half of FIG. 34G with the "Substance Name" node (whose child node is an "Operation" node) illustrated in the upper half of FIG. The "integration of child nodes of different generalization nodes" may be, for example, adding a child node (the "Silicon Gel" node) of another "Substance Name" node to the child node of the "Substance Name" node (which does not have a "Silicon Gel" node as a child node) in FIG. 34F.) Edge re-routing process in which the generalized node to which an edge is connected or the generalized node from which an edge is connected is changed to another generalized node.

[0269] A pattern condition that a generalized graph conforms to may be one in which the similarity between the generalized graph and the pattern condition is equal to or greater than a threshold value. This is expected to enable selection of an appropriate graph pattern for the graph structure of the generalized graph.

[0270] The update definition data may include master-slave relationship data (e.g., master-slave relationship table 3520) that represents a master-slave relationship between document parts or documents, and update method data (e.g., replacement structured data 3510) that represents, for a subordinate document part or document, a graph update method according to the difference between the master document part or document and the subordinate document part or document. An example of the "graph update method" here is the replacement in Example 3 (replacing the node of the entity (or taxonomy) from the source with the node of the entity (or taxonomy) to be replaced). Alternatively or additionally, it may include adding a node of the entity (or taxonomy) or deleting a node of the entity (or taxonomy). When the structured data corresponds to the master document part or document, the structured update process may include identifying the subordinate document part or document corresponding to the master document part or document from the master-slave relationship data and updating the structured graph represented by the copy of the structured data according to the graph update method represented by the update method data, thereby generating updated structured data. This is expected to generate accurate updated structured data for subordinate document portions or documents based on the structured data for the main document portion or document.

[0271] The update definition data may include name matching definition data (e.g., name matching taxonomy data 4010 and name matching relationship table 4020) that represents entities before name matching and taxonomies after name matching for each name matching type. The structure update process may include generating updated structured data by updating entity nodes corresponding to entities before name matching in a structured graph represented by the duplicated structured data to nodes that represent the taxonomy after name matching. This is expected to generate accurate updated structured data.

[0272] The structured data may be data prepared by any method. In the structured data, data of properties of elements such as entity nodes and edges may include data representing entities, taxonomies, etc.

[0273] The structured data may be generated as follows: A computing device may extract expressions related to multiple procedures as entities from a document. The computing device may classify the categories of the entities. The computing device may generate multiple entity groups, each consisting of one or more entities and corresponding to one procedure. For each entity group, the computing device may identify a primary entity that is an entity that characterizes the procedure corresponding to the entity group based on the categories of the one or more entities included in the entity group. The computing device may perform a first order determination process that determines the order of the multiple procedures based on the relationships between the primary entities. The computing device may determine the order of the multiple procedures based on the result of the first order determination process. The computing device may generate information about the ordered entity groups as structured data and output the structured data.

[0274] The computing device may execute a parallelism determination process to identify procedures to be executed in parallel based on the relationship between the main entities, and may determine the order of the multiple procedures based on the results of the first order determination process and the parallelism determination process. In the first order determination process, the order of two procedures may be determined based on at least one of a character string included in a sentence connecting the main entities and a similarity between the main entities. In the parallelism determination process, procedures to be executed in parallel may be identified based on a character string included in a sentence connecting the main entities. Information for managing rules for determining the order of two procedures based on at least one of a character string included in a sentence connecting the entities and a similarity between the entities, and information for managing rules for determining whether the procedures are to be executed in parallel based on a character string included in a sentence connecting the main entities may be stored in the data processing device.

[0275] The computing device may classify, for each entity group, a category of procedures corresponding to the entity group based on the categories of one or more entities included in the entity group. The computing device may execute a second order determination process to determine the order of multiple procedures based on the order of the procedures and the relationships between the procedure categories. The computing device may determine the order of the multiple procedures based on the first order determination process and the second order determination process. Information for managing rules defining the order of appearance of procedure categories in a business process may be held in the data processing device. [Explanation of symbols]

[0276] 100, 2500, 3500, 4000 Structured Processing Unit

Claims

1. A storage device storing graph data for a document representing multiple procedures; a computing device that is connected to the storage device and performs a structured update process that generates updated graph data; Equipped with the graph data is data representing a graph including a plurality of entity nodes and one or a plurality of edges; each of the plurality of entity nodes is a node representing an entity in a document; the structured update process includes updating the structured graph based on update definition data, which is data that defines an update of at least one of a node and an edge using an expression using a taxonomy, and a taxonomy of at least one entity node in the structured graph, which is a graph represented by the graph data or a copy thereof; the updated graph data is data representing a graph after the structured graph has been updated; A taxonomy is either an entity taxonomy, which is a taxonomy of entities, or a relationship taxonomy, which is a taxonomy of entity taxonomies; the update definition data includes graph pattern data representing one or more graph patterns, each of the one or more graph patterns includes a pattern condition that is a graph structure corresponding to the graph pattern or a summary thereof, and a completion process for the corresponding graph structure; the structural update process includes a graph generalization process and a graph update process; the graph generalization process includes converting the graph represented by the graph data into a generalized graph including a plurality of generalized nodes and one or a plurality of edges; each of the plurality of generalization nodes is a node representing an entity taxonomy or a relationship taxonomy of an entity represented by the entity node, or a copy of the entity node; The graph update process includes: updating the generalized graph by performing a complementation process on the generalized graph in a graph pattern including a pattern condition to which the generalized graph conforms; generating the updated graph data by making the graph structure of the structured graph conform to the graph structure of the updated generalized graph; Including, Data processing device.

2. the structured update process includes a graph integration process for generating an integrated graph in which one or more taxonomy nodes are associated with the structured graph; For each of a plurality of entity nodes represented by the structured graph, if there are one or more taxonomies for the entity represented by the entity node, the one or more taxonomy nodes are associated in the integrated graph; Each taxonomy node is a node that represents a taxonomy. the arithmetic unit performs the graph generalization process after the graph integration process, for each entity node, a generalization node in the generalization graph corresponds to the taxonomy node associated with that entity node in the integration graph; 2. The data processing device according to claim 1.

3. For entities that have associated entity taxonomies and relationship taxonomies, the integrated graph includes: The child node of the entity node of the entity is the taxonomy node of the entity taxonomy, A child node of the taxonomy node is a taxonomy node of a relationship taxonomy of the entity, the graph structure of the generalized graph is based on the graph structure of the integrated graph; 3. The data processing device according to claim 2.

4. There are multiple types of taxonomies for at least one of the entity taxonomy and the relationship taxonomy, The type of taxonomy to be used is defined for each of a plurality of types of completion processes for the generalized graph, the graph generalization process includes converting, for each of the plurality of types of completion processes, the structured graph into a generalization graph including a generalization node representing a taxonomy belonging to a type corresponding to the completion process; 2. The data processing device according to claim 1.

5. for each of the plurality of graph patterns, the graph pattern includes any one of the plurality of types of completion processes, The plurality of types of complementation processes include two or more of the following complementation processes: - Separation process to separate the generalized graph, A duplication process in which a duplicate of a generalization node corresponding to a previous procedure of a certain procedure is adopted as a parent node or a child node of the generalization node corresponding to the certain procedure. An addition process for adding a new generalized node as a parent node or a child node of a generalized node; A merger process that merges different generalization nodes or their child generalization nodes that are associated with the same classification metataxonomy; Edge rerouting process in which the generalized node to which an edge is connected or the generalized node from which an edge is connected is changed to another generalized node; 5. The data processing device according to claim 4.

6. The pattern condition to which the generalization graph conforms is a condition in which the similarity between the generalization graph and the pattern condition is equal to or greater than a threshold value of similarity.

2. The data processing device according to claim 1.

7. A storage device storing graph data for a document representing a plurality of procedures; a computing device that is connected to the storage device and performs a structured update process that generates updated graph data; Equipped with the graph data is data representing a graph including a plurality of entity nodes and one or a plurality of edges; each of the plurality of entity nodes is a node representing an entity in a document; the structured update process includes updating the structured graph based on update definition data, which is data that defines an update of at least one of a node and an edge using an expression using a taxonomy, and a taxonomy of at least one entity node in the structured graph, which is a graph represented by the graph data or a copy thereof; the updated graph data is data representing a graph after the structured graph has been updated; the update definition data includes master-slave relationship data representing a master-slave relationship between document parts or documents, and update method data representing a graph update method for the subordinate document part or document according to a difference between the master document part or document and the subordinate document part or document, If the graph data corresponds to the main document part or document, the structural update process: Identifying the subordinate document portion or document corresponding to the master document portion or document from the master-slave relationship data; updating the structured graph represented by the copy of the graph data in accordance with the graph update method represented by the update method data, thereby generating the updated graph data; Including, Data processing device.

8. A storage device storing graph data for a document representing a plurality of procedures; a computing device that is connected to the storage device and performs a structured update process that generates updated graph data; Equipped with the graph data is data representing a graph including a plurality of entity nodes and one or a plurality of edges; each of the plurality of entity nodes is a node representing an entity in a document; the structured update process includes updating the structured graph based on update definition data, which is data that defines an update of at least one of a node and an edge using an expression using a taxonomy, and a taxonomy of at least one entity node in the structured graph, which is a graph represented by the graph data or a copy thereof; the updated graph data is data representing a graph after the structured graph has been updated; the update definition data includes name matching definition data representing an entity before name matching and a taxonomy after name matching for each name matching type; the structured update process includes updating an entity node corresponding to the entity before the name matching in the structured graph represented by the copy of the graph data to a node representing the taxonomy after the name matching, thereby generating the updated graph data; Data processing device.

9. A storage device storing graph data for a document representing a plurality of procedures; a computing device that is connected to the storage device and performs a structured update process that generates updated graph data; Equipped with the graph data is data representing a graph including a plurality of entity nodes and one or a plurality of edges; each of the plurality of entity nodes is a node representing an entity in a document; the structured update process includes updating the structured graph based on update definition data, which is data that defines an update of at least one of a node and an edge using an expression using a taxonomy, and a taxonomy of at least one entity node in the structured graph, which is a graph represented by the graph data or a copy thereof; the updated graph data is data representing a graph after the structured graph has been updated; The computing device extracting expressions related to the plurality of procedures as entities from the document; classifying categories of said entities; generating a plurality of entity groups each consisting of one or more of the entities and corresponding to one of the procedures; for each of the entity groups, identifying a primary entity that is the entity that characterizes the procedure corresponding to the entity group based on a category of one or more of the entities included in the entity group; execute a first order determination process that determines an order of the plurality of procedures based on the relationship between the primary entities; determining an order of the plurality of procedures based on a result of the first order determination process; generating, as the graph data, information about the ordered entity groups; Data processing device.

10. a computer performs a structural update process that generates updated graph data for the document representing a plurality of procedures; the graph data is graph data representing a graph including a plurality of entity nodes and one or a plurality of edges; each of the plurality of entity nodes is a node representing an entity in a document; the structured update process includes updating the structured graph based on update definition data, which is data that defines an update of at least one of a node and an edge using an expression using a taxonomy, and a taxonomy of at least one entity node in the structured graph, which is a graph represented by the graph data or a copy thereof; the updated graph data is data representing a graph after the structured graph has been updated; A taxonomy is either an entity taxonomy, which is a taxonomy of entities, or a relationship taxonomy, which is a taxonomy of entity taxonomies; the update definition data includes graph pattern data representing one or more graph patterns, each of the one or more graph patterns includes a pattern condition that is a graph structure corresponding to the graph pattern or a summary thereof, and a completion process for the corresponding graph structure; the structural update process includes a graph generalization process and a graph update process; the graph generalization process includes converting the graph represented by the graph data into a generalized graph including a plurality of generalized nodes and one or a plurality of edges; each of the plurality of generalization nodes is a node representing an entity taxonomy or a relationship taxonomy of an entity represented by the entity node, or a copy of the entity node; The graph update process includes: updating the generalized graph by performing a complementation process on the generalized graph in a graph pattern including a pattern condition to which the generalized graph conforms; generating the updated graph data by making the graph structure of the structured graph conform to the graph structure of the updated generalized graph; Including, Data processing methods.

11. A method for generating graph data after updating graph data for a document representing a plurality of procedures, comprising: the graph data is graph data representing a graph including a plurality of entity nodes and one or a plurality of edges; each of the plurality of entity nodes is a node representing an entity in a document; the structured update process includes updating the structured graph based on update definition data, which is data that defines an update of at least one of a node and an edge using an expression using a taxonomy, and a taxonomy of at least one entity node in the structured graph, which is a graph represented by the graph data or a copy thereof; the updated graph data is data representing a graph after the structured graph has been updated; the update definition data includes master-slave relationship data representing a master-slave relationship between document parts or documents, and update method data representing a graph update method for the subordinate document part or document according to a difference between the master document part or document and the subordinate document part or document, If the graph data corresponds to the main document part or document, the structural update process: Identifying the subordinate document portion or document corresponding to the master document portion or document from the master-slave relationship data; updating the structured graph represented by the copy of the graph data in accordance with the graph update method represented by the update method data, thereby generating the updated graph data; Including, Data processing methods.

12. A computer performs a structured update process that generates updated graph data for graph data on a document that represents multiple procedures, the graph data is graph data representing a graph including a plurality of entity nodes and one or a plurality of edges; each of the plurality of entity nodes is a node representing an entity in a document; the structured update process includes updating the structured graph based on update definition data, which is data that defines an update of at least one of a node and an edge using an expression using a taxonomy, and a taxonomy of at least one entity node in the structured graph, which is a graph represented by the graph data or a copy thereof; the updated graph data is data representing a graph after the structured graph has been updated; the update definition data includes name matching definition data representing an entity before name matching and a taxonomy after name matching for each name matching type; the structured update process includes updating an entity node corresponding to the entity before the name matching in the structured graph represented by the copy of the graph data to a node representing the taxonomy after the name matching, thereby generating the updated graph data; Data processing methods.

13. A method for generating graph data after updating graph data for a document representing a plurality of procedures, comprising: the graph data is graph data representing a graph including a plurality of entity nodes and one or a plurality of edges; each of the plurality of entity nodes is a node representing an entity in a document; the structured update process includes updating the structured graph based on update definition data, which is data that defines an update of at least one of a node and an edge using an expression using a taxonomy, and a taxonomy of at least one entity node in the structured graph, which is a graph represented by the graph data or a copy thereof; the updated graph data is data representing a graph after the structured graph has been updated; The computer extracting expressions related to the plurality of procedures as entities from the document; classifying categories of said entities; generating a plurality of entity groups each consisting of one or more of the entities and corresponding to one of the procedures; for each of the entity groups, identifying a primary entity that is the entity that characterizes the procedure corresponding to the entity group based on a category of one or more of the entities included in the entity group; execute a first order determination process that determines an order of the plurality of procedures based on the relationship between the primary entities; determining an order of the plurality of procedures based on a result of the first order determination process; generating, as the graph data, information about the ordered entity groups; Data processing methods.

14. causing a computer to perform a structured update process which is a process of generating updated graph data for graph data on a document representing a plurality of procedures; the graph data is graph data representing a graph including a plurality of entity nodes and one or a plurality of edges; each of the plurality of entity nodes is a node representing an entity in a document; the structured update process includes updating the structured graph based on update definition data, which is data that defines an update of at least one of a node and an edge using an expression using a taxonomy, and a taxonomy of at least one entity node in the structured graph, which is a graph represented by the graph data or a copy thereof; the updated graph data is data representing a graph after the structured graph has been updated; A taxonomy is either an entity taxonomy, which is a taxonomy of entities, or a relationship taxonomy, which is a taxonomy of entity taxonomies; the update definition data includes graph pattern data representing one or more graph patterns, each of the one or more graph patterns includes a pattern condition that is a graph structure corresponding to the graph pattern or a summary thereof, and a completion process for the corresponding graph structure; the structural update process includes a graph generalization process and a graph update process; the graph generalization process includes converting the graph represented by the graph data into a generalized graph including a plurality of generalized nodes and one or a plurality of edges; each of the plurality of generalization nodes is a node representing an entity taxonomy or a relationship taxonomy of an entity represented by the entity node, or a copy of the entity node; The graph update process includes: updating the generalized graph by performing a complementation process on the generalized graph in a graph pattern including a pattern condition to which the generalized graph conforms; generating the updated graph data by making the graph structure of the structured graph conform to the graph structure of the updated generalized graph; Including, Computer program.

15. A method for causing a computer to perform a structured update process, which is a process for generating updated graph data for graph data on a document representing a plurality of procedures, the graph data is graph data representing a graph including a plurality of entity nodes and one or a plurality of edges; each of the plurality of entity nodes is a node representing an entity in a document; the structured update process includes updating the structured graph based on update definition data, which is data that defines an update of at least one of a node and an edge using an expression using a taxonomy, and a taxonomy of at least one entity node in the structured graph, which is a graph represented by the graph data or a copy thereof; the updated graph data is data representing a graph after the structured graph has been updated; the update definition data includes master-slave relationship data representing a master-slave relationship between document parts or documents, and update method data representing a graph update method for the subordinate document part or document according to a difference between the master document part or document and the subordinate document part or document, If the graph data corresponds to the main document part or document, the structural update process: Identifying the subordinate document portion or document corresponding to the master document portion or document from the master-slave relationship data; updating the structured graph represented by the copy of the graph data in accordance with the graph update method represented by the update method data, thereby generating the updated graph data; Including, Computer program.

16. A method for causing a computer to perform a structured update process, which is a process for generating updated graph data for graph data on a document representing a plurality of procedures, the graph data is graph data representing a graph including a plurality of entity nodes and one or a plurality of edges; each of the plurality of entity nodes is a node representing an entity in a document; the structured update process includes updating the structured graph based on update definition data, which is data that defines an update of at least one of a node and an edge using an expression using a taxonomy, and a taxonomy of at least one entity node in the structured graph, which is a graph represented by the graph data or a copy thereof; the updated graph data is data representing a graph after the structured graph has been updated; the update definition data includes name matching definition data representing an entity before name matching and a taxonomy after name matching for each name matching type; the structured update process includes updating an entity node corresponding to the entity before the name matching in the structured graph represented by the copy of the graph data to a node representing the taxonomy after the name matching, thereby generating the updated graph data; Computer program.

17. A method for causing a computer to perform a structured update process, which is a process for generating updated graph data for graph data on a document representing a plurality of procedures, the graph data is graph data representing a graph including a plurality of entity nodes and one or a plurality of edges; each of the plurality of entity nodes is a node representing an entity in a document; the structured update process includes updating the structured graph based on update definition data, which is data that defines an update of at least one of a node and an edge using an expression using a taxonomy, and a taxonomy of at least one entity node in the structured graph, which is a graph represented by the graph data or a copy thereof; the updated graph data is data representing a graph after the structured graph has been updated; The computer, extracting expressions related to the plurality of procedures as entities from the document; classifying categories of said entities; generating a plurality of entity groups each consisting of one or more of the entities and corresponding to one of the procedures; for each of the entity groups, identifying a primary entity that is the entity that characterizes the procedure corresponding to the entity group based on a category of one or more of the entities included in the entity group; execute a first order determination process that determines an order of the plurality of procedures based on the relationship between the primary entities; determining an order of the plurality of procedures based on a result of the first order determination process; generating, as the graph data, information about the ordered entity groups; A computer program that makes things happen.

Images