Knowledge creation support device, knowledge creation support method, and knowledge creation support program

Abstract

This invention provides a knowledge generation support device, a knowledge generation support method, and a knowledge generation support program that assist in the verification process required when generating knowledge from text data. [Solution] A knowledge generation support device 1100 that supports the generation of business-related knowledge has: a memory unit 17 that stores information about multiple knowledge reviewers; an input unit 12 that accepts text data related to the business; a text mining unit 142 that analyzes the text data; a graph generation unit 144 that generates a graph that is an example of the knowledge based on the analysis results by the text mining unit 142; and a reviewer identification unit 147 that identifies the reviewer of the generated knowledge based on the analyzed text data and the stored information about the reviewer of the knowledge.

Description

[Technical Field]

[0001] The present invention relates to knowledge generation in information processing, and in particular to a technique for supporting knowledge generation using text mining. [Background technology]

[0002] Recent trends in IT (Information Technology) include the spread of cloud environments that allow for the inexpensive storage of large amounts of data, improvements in computer processing performance, and improvements in data processing technologies such as AI (Artificial Intelligence). With regard to data processing technologies, natural language processing technology, especially generative AI, has recently been rapidly developing and becoming widespread.

[0003] The momentum for data utilization among medium-sized companies is growing more than ever, and data utilization is also advancing in various areas of the manufacturing industry. In response to this trend, as mentioned above, with the development and widespread adoption of natural language processing technology in recent years, there has been a trend toward analyzing and utilizing unstructured data, such as free text in documents. This trend can be seen, for example, in quality assurance and related tasks, where there are attempts to analyze quality information such as product defect reports to extract knowledge and provide feedback to processes such as design, manufacturing, and maintenance. In this specification, feedback refers to the transmission of defect and countermeasure information to product design, process design, etc., with the aim of preventing the defect from recurring.

[0004] Regarding the analysis of data containing free text, a method for extracting knowledge using text mining technology has been devised. In Patent Document 1, text mining is used to mechanically extract information from written reports of product failures. In addition, a dictionary is used to extract fault events so that the method can be used to deal with text in which general semantic information alone cannot sufficiently narrow down the causative component.

[0005] Regarding the feedback of extracted knowledge, as described in Patent Document 2, a method has been devised to automatically extract keywords from a document being created by the user, compare them with a list of issues containing the issues and related terms, identify issues for each design stage, and classify and display information according to the issues.

[0006] When applying the analysis of quality information such as defect reports to generate knowledge and feed it back into processes such as design, manufacturing, and maintenance in actual business operations, the following problems may arise. While defect reports will be used as a concrete example of quality information in the following explanation, this invention is also applicable to quality information other than product defects, such as violations of laws and regulations.

[0007] A defect report contains information such as the product, component, report author, and details of the defect. While the product, component, and report author may be structured data, the details of the defect, such as the defect event and its causal relationship, are generally written in free-text format as unstructured data. Users extract information from this unstructured free-text data and concisely combine the product, component, defect event, and causal relationship to create knowledge. It is assumed that extracting knowledge from a single report is burdensome, and this burden increases if there are multiple reports. Furthermore, it is likely that the same defect may be reported multiple times, but users cannot determine the identity of the defect without reading each defect report individually, and aggregating identical defects also places a burden on the user. In this invention, an identical defect refers to a defect in which the product, component, defect event, causal relationship, process causing the defect, and tasks in the process causing the defect are the same. [Prior art documents] [Patent documents]

[0008] [Patent Document 1] Japanese Patent Publication No. 2023-163427 [Patent Document 2] Japanese Patent Publication No. 2011-108067 Summary of the Invention [Problem to be solved by the invention]

[0009] Patent Document 1 discloses a device for identifying faulty parts and technology to support faulty part identification using the same, and Patent Document 1 can support the generation and utilization of knowledge. However, with Patent Document 1, the accuracy of the extracted knowledge must be verified by a human. Furthermore, verifying accuracy requires identifying an appropriate verifier and requesting verification from the target person, but if for some reason the creator or other information cannot be read from the defect report, a different method must be used to identify the verifier. These tasks are considered to be burdensome, but Patent Document 1 does not address this.

[0010] Patent Document 2 discloses a technology that supports the feedback of quality information through the distribution of knowledge. However, Patent Document 2 does not address the identification and verification of the person who will review the generated knowledge. Therefore, the present invention aims to support the generation of knowledge from text data, and in particular, to support the knowledge verification process. [Means for solving the problem]

[0011] To achieve this objective, the present invention identifies the verifier of the knowledge and / or the text data used to generate said knowledge.

[0012] More specifically, the knowledge generation support device is a device that supports the generation of business-related knowledge, and comprises a storage unit that stores information on multiple knowledge verifiers, an input unit that receives text data related to the business, a text mining unit that analyzes the text data, a knowledge generation unit that generates the knowledge based on the analysis results by the text mining unit, and a verifier identification unit that identifies the verifier of the generated knowledge based on the analyzed text data and the stored information on the knowledge verifiers.

[0013] Furthermore, the present invention also includes a knowledge creation support method executed by the knowledge creation support device, a knowledge creation support program for causing the knowledge creation support device to function as a computer, and a storage medium storing the same. [Effects of the Invention]

[0014] According to the present invention, it is possible to support the verification process during knowledge generation, thereby supporting more efficient knowledge generation. [Brief explanation of the drawings]

[0015] [Figure 1A] This is a functional block diagram of a knowledge generation support system including a knowledge generation support device 1100 in one embodiment of the present invention. [Figure 1B] This diagram illustrates the hardware configuration of the knowledge generation support device 1100 in one embodiment of the present invention. [Figure 2] This is a diagram illustrating text mining in one embodiment of the present invention. [Figure 3] This is a diagram illustrating the generation of a knowledge base in one embodiment of the present invention. [Figure 4] This figure schematically shows a graph in one embodiment of the present invention. [Figure 5] FIG. 18 is a diagram showing an excerpt from a table of work record data in a product development project managed by a product development project management system 1800 according to one embodiment of the present invention. [Figure 6] This figure shows a user dictionary 1470 in one embodiment of the present invention. [Figure 7] This figure shows the WF creation instruction data in one embodiment of the present invention. [Figure 8] This figure shows the contents of a knowledge verification request email in one embodiment of the present invention. [Figure 9] This flowchart shows the overall flow from obtaining a defect report to storing the finalized knowledge in one embodiment of the present invention. [Figure 10] 10 is a flowchart illustrating an example of a processing flow including identification of a verifying person in one embodiment of the present invention. [Figure 11] FIG. 10 is a diagram showing a confirmation screen in one embodiment of the present invention. [Figure 12A] 10 is a flowchart illustrating a case where confirmation processing is executed as serial processing in one embodiment of the present invention. [Figure 12B] 10 is a flowchart illustrating a case where confirmation processing is executed as parallel processing in one embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0016] An embodiment of the present invention will be described below with reference to the accompanying drawings. In this embodiment, a reviewer of knowledge and / or text data used to generate the knowledge is identified. Here, the knowledge in this embodiment refers to business-related knowledge, particularly knowledge related to product manufacturing.

[0017] The text data is data related to manufacturing, which is an example of a business, and includes reports (e.g., defect reports), specifications, design documents, etc. The text data in this embodiment may include control codes other than character strings. Knowledge generation refers to identifying knowledge, and is a concept that also includes knowledge extraction. The checker is a person involved in the business, for example, manufacturing, and includes a designer, a quality assurance officer, a manufacturing worker, etc.

[0018] <Schematic configuration of this embodiment> The schematic configuration of this embodiment will be described with reference to FIGS. 1A and 1B.

[0019] <Function block> 1A is a functional block diagram of a knowledge creation support system including a knowledge creation support device 1100 (hereinafter also referred to as the present device 1100) according to this embodiment. The knowledge creation support device 1100 is connected to an input / output device 1600, a defect management system 1700, a product development project management system 1800, and a WF system 1900 (WF: workflow) via a network 1000 to form the knowledge creation support system. The present device 1100 then executes processing to identify a reviewer of knowledge and / or text data used in the generation of the knowledge. To this end, the knowledge creation support device 1100 includes a communication unit 11, an input unit 12, an output unit 13, a knowledge generation processing unit 14, a knowledge reviewer information identification unit 146, a WF creation instruction unit 15, a knowledge distribution unit 16, and a storage unit 17.

[0020] First, the communication unit 11 connects to other devices and systems via the network 1000. The input unit 12 accepts operations from a user. Then, the communication unit 11 and the input unit 12 accept text data such as a defect report. The output unit 13 outputs, for example, displays, processing results of the knowledge creation support device 1100. The input unit 12 and the output unit 13 may be configured as an integrated unit, such as a touch panel. Furthermore, if the knowledge creation support device 1100 is realized as a server or the like, the input unit 12 and the output unit 13 can be omitted.

[0021] The knowledge generation processing unit 14 also executes processing for generating knowledge related to manufacturing operations based on text data. To this end, the knowledge generation processing unit 14 includes a quality information acquisition unit 141, a text mining unit 142, a mining basis generation unit 143, a graph generation unit 144, and an identical quality information aggregation unit 145.

[0022] The quality information acquisition unit 141 acquires defect report data stored in the defect management system 1700 via the communication device 1500. The text mining unit 142 analyzes the text data, i.e., executes text mining processing. As a result, the text mining unit 142 extracts knowledge grounds that can become knowledge. Note that this knowledge ground itself may be treated as knowledge.

[0023] Here, the text mining unit 142 analyzes, for example, a manufacturing defect report as text data, and extracts analysis items such as knowledge basis candidates from the data.

[0024] Furthermore, the mining basis generation unit 143 generates knowledge basis that serves as the basis for generating knowledge in accordance with the text mining process performed by the text mining unit 142. To this end, the mining basis generation unit 143 identifies partial text of the text data that is extracted as the knowledge basis. It is desirable that such knowledge basis be displayed on the output unit 13.

[0025] The graph generation unit 144 is an example of a knowledge generation unit that generates knowledge based on the analysis results of text data. The knowledge generation unit generates knowledge from the knowledge basis as the analysis result. The knowledge generation unit may use the knowledge basis itself as knowledge, or may generate knowledge from the knowledge basis. Here, the graph generation unit 144 generates a graph, which is a graph structure, as knowledge from the knowledge basis.

[0026] Furthermore, the identical quality information aggregating unit 145 aggregates related text data, such as identical text data, for multiple pieces of text data processed by the text mining unit 142. More preferably, the identical quality information aggregating unit 145 aggregates multiple pieces of knowledge based on related text data. Note that the aggregation of knowledge may be performed by a knowledge generating unit such as the graph generating unit 144.

[0027] In this embodiment, defect reports are used as text data. In this case, the identical quality information aggregating unit 145 determines whether there are defect reports about the same defect among multiple defect reports, and aggregates the identical reports into one knowledge.

[0028] Furthermore, the knowledge checker information identification unit 146 identifies the checker who checks the generated knowledge, or in the above example, the graph. At this time, it is desirable to also identify information about the checker. For this purpose, the knowledge checker information identification unit 146 has a checker identification unit 147 and a user ID matching unit 148.

[0029] First, the checker identification unit 147 identifies relevant parties in the work who could be checkers from the text data. In this embodiment, a defect report is used as the text data. To this end, the checker identification unit 147 identifies the work records in the development project for the product using the product, parts, target process, and tasks in the target process that are the subject of the defect report as keys, and identifies the person in charge. Furthermore, the user ID matching unit 148 uses a user dictionary 1470 that indicates information about the checker to identify the user ID that identifies the identified person in charge.

[0030] Furthermore, the WF creation instruction unit 15 creates WF creation instruction data for creating a workflow for having the checker identified by the checker identification unit 147 check the knowledge. Furthermore, the knowledge distribution unit 16 distributes knowledge such as a graph generated by the graph generation unit 144 to persons involved in the work via the communication unit 11. Preferably, the knowledge distribution unit 16 distributes the knowledge confirmed by the checker.

[0031] The storage unit 17 also stores a product configuration dictionary 1450, a project process / task dictionary 1460, a user dictionary 1470, and a database 1480. First, the product configuration dictionary 1450 indicates configuration data of a product to be manufactured. This is an example of data on the object of a business operation.

[0032] Furthermore, the project process / task dictionary 1460 shows the process data for a manufacturing project. This is an example of data related to procedures and steps in business operations. The user dictionary 1470 contains data such as the names and IDs of users, including verifiers. As mentioned above, the user dictionary 1470 is an example of data related to verifiers. The database 1480 shows various data such as generated knowledge, knowledge basis, and text data.

[0033] Furthermore, the input / output device 1600 consists of a keyboard, mouse, display, etc., which the user uses to input data and display the processing results, and can be implemented using a computer such as a PC or tablet. Note that if the knowledge generation support device 1100 is equipped with an input unit 12 and an output unit 13, the input / output device 1600 can be omitted.

[0034] Furthermore, the defect management system 1700 creates and stores a defect report, which is an example of text data. The product development project management system 1800 creates and stores work records in a product development project, which is an example of a business process. The WF system 1900 executes a workflow to allow a reviewer to confirm the knowledge. This concludes the explanation of the functional blocks in this embodiment. Next, an implementation example for realizing each of these functions will be described.

[0035] <Implementation Example (Hardware Configuration)> Figure 1B is a diagram illustrating the hardware configuration of the knowledge generation support device 1100 in this embodiment. The knowledge generation support device 1100 can be implemented on a computer such as a server in the cloud. Therefore, in Figure 1B, the knowledge generation support device 1100 consists of an arithmetic unit 1200, a main memory 1300, and an auxiliary memory 1400. and a communication device 1500, which are connected to each other via a communication path. The knowledge creation support device 1100 is connected to a terminal device 1601, which is an example of an input / output device 1600, via the network 1000. It is connected to a defect management system 1700, a product development project management system 1800, and a WF system 1900.

[0036] First, the arithmetic unit 1200 can be implemented using a processor such as a CPU, and it performs calculations according to the knowledge generation processing program 1410 and the knowledge verifier information identification program 1420 stored in the auxiliary storage device 1400, which will be described later. These programs will be described later.

[0037] 7, and is connected to the network 40 to communicate with other devices.

[0038] Furthermore, the main memory 1300 and the auxiliary memory 1400 correspond to the storage unit 17 in Figure 1A. The main memory 1300 then unpacks the knowledge generation processing program 1410 and the knowledge verifier information identification program 1420, which are stored in the auxiliary memory 1400, as well as information used for processing by the arithmetic unit 1200. The auxiliary memory 1400 can be implemented as so-called storage and stores the knowledge generation processing program 1410, the knowledge verifier information identification program 1420, the WF creation instruction program 1430, the knowledge distribution program 1440, the product configuration dictionary 1450, the project process / task dictionary 1460, the user dictionary 1470, and the database 1480.

[0039] Furthermore, the auxiliary storage device 1400 may be implemented using various storage media such as an external HDD (Hard Disk Drive), SSD (Solid State Drive), or memory card, or it may be implemented as a separate device from the knowledge generation support device 11100, such as a file server.

[0040] Here, the knowledge generation processing program 1410 is composed of a quality information acquisition subprogram 1411, a text mining subprogram 1412, a mining basis generation subprogram 1413, a graph generation subprogram 1414, and an identical quality information aggregation subprogram 1415. Also, the knowledge checker information identification program 1420 is composed of a checker identification subprogram 1421 and a user ID matching subprogram 1422.

[0041] Each of the above subprograms may be implemented as individual programs or combinations thereof, or they may be organized as so-called modules.

[0042] Here, the configuration shown in Figure 1A, which performs the same function as each program and submodule, is as follows: Knowledge generation processing program 1410: knowledge generation processing unit 14 Quality information acquisition subprogram 1411: quality information acquisition unit 141 Text mining subprogram 1412: Text mining unit 142 Mining basis generation subprogram 1413: Mining basis generation unit 143 Graph generation subprogram 1414: Graph generation unit 144 Same quality information aggregation subprogram 1415: Same quality information aggregation unit 145 Knowledge checker information identification program 1420: Knowledge checker information identification unit 146 Checker identification subprogram 1421: Checker identification unit 147 User ID verification subprogram 1422: User ID verification unit 148 WF creation instruction program 1430: WF creation instruction section 15 Knowledge Distribution Program 1440: Knowledge Distribution Department 16 Therefore, the computing unit 1200 executes the processing of the knowledge generation processing unit 14, the knowledge verifier information identification unit 146, the WF creation instruction unit 15, the knowledge distribution unit 16, and each of the units that constitute them, in accordance with the knowledge generation processing program 1410, the knowledge verifier information identification program 1420, the WF creation instruction program 1430, the knowledge distribution program 1440, and the subprograms that constitute them.

[0043] <Function details> Next, the details of the functions of each program and subprogram in this embodiment will be described. These functions are implemented by the arithmetic unit 1200 through each program and its constituent subprograms. Furthermore, each function is executed by the aforementioned programs and subprograms and the parts shown in Figure 1A.

[0044] First, the knowledge generation processing program 1410 analyzes quality information and executes a process to generate knowledge, including information such as events and causes. This process for generating knowledge includes the generation of the knowledge itself. Next, the knowledge verifier information identification program 1420 identifies the verifiers of the created knowledge. Then, the WF creation instruction program 1430 creates instruction data for creating a knowledge verification WF based on the identified verifiers. Finally, the knowledge distribution program 1440 distributes the verified knowledge to users. Next, the knowledge generation processing program 1410 analyzes the defect report as text data and generates knowledge. At this time, the knowledge generation processing program 1410 identifies the knowledge basis, which is the basis for generating the knowledge. Here, the subprograms that make up the knowledge generation processing program 1410 will be described.

[0045] First, the quality information acquisition subprogram 1411 acquires defect reports from the defect management system 1700, which is an external system. Alternatively, defect reports may be acquired from the terminal device 1601 or the product development project management system 1800. Furthermore, the quality information acquisition subprogram 1411 may also acquire quality information related to product quality, such as specifications, other than defect reports, as text data.

[0046] Furthermore, the acquired defect reports are analyzed, or text mined, using the text mining subprogram 1412. Here, the text mining subprogram 1412 performs text mining on the acquired defect reports and extracts the following as analysis items: the target product, the target part, the defect phenomenon, the causal relationship of the defect, the process that caused the defect, the tasks in the process that caused the defect, the affiliation of the knowledge verifier, and the name of the knowledge verifier.

[0047] Here, we will explain text mining using Figure 2. In Figure 2, the extracted analysis items in a defect report, which is an example of text data, are shown with thick borders. Specifically, the signature field 2100 in Figure 2 has been extracted as a candidate for verifier. This signature field 2100 reveals the affiliation and name of the author, verifier, and approver of the defect report. These individuals are suitable as verifiers of the accuracy of the knowledge. Thus, when a defect report has a signature field, it is easy to identify the knowledge verifier. Cases where the verifier cannot be identified from the defect report due to reasons such as the absence of a signature field will be explained later. Note that analysis items are items (words, etc.) extracted from text data by text mining, and include candidate knowledge bases and candidate verifiers.

[0048] Next, regarding the content of the status column, "In the fuel piping of the fuel hose / pipe (model number: P-aaaa)" 2200 and "excessive load" 2500 are written in one sentence, and the following sentence states, "As a result, if continued use is carried out, the piping in question will break and fuel will leak" 2600. From this, it can be seen that the "fuel hose / pipe (model number: P-aaaa)" is the target part, and "excessive load" and "the piping in question will break and fuel will leak" are malfunction events. In other words, these can be extracted as knowledge basis candidates by text mining.

[0049] Additionally, "inappropriate structure" 2300 and "excessive load" 2500 are connected by the expression "because" 2400. From this, it can be seen that "inappropriate structure" is the cause and "excessive load" is the effect. Furthermore, it can be inferred that "Therefore, if continued use is as is, the pipe in question may be damaged, resulting in a fuel leak" 2600 is a continuation of "excessive load" 2500. Therefore, it can be seen that there is a causal relationship with "inappropriate structure." And from the statement "The vehicle model equipped with this device is BBBB (model number: I-bbbb)" 2700, it can be seen that this is the affected product. Furthermore, from the statement "inappropriate structure" 2300, it can be inferred that the causal process is design, and that the task within that process is structural design. As a result, the above statements are extracted as candidate knowledge grounds using text mining.

[0050] Furthermore, the following dictionaries (items 1-4 below) can be used to improve the accuracy of text mining: (Item 1) Prepare a dictionary of defect events (Item 2) Prepare a dictionary of causal relationships (Item 3) Prepare a dictionary to identify the relationships between products and parts. (Item 4) Prepare a dictionary to identify processes and tasks. Each of these dictionaries can be stored in the memory unit 17, and it is desirable to store at least one of them.

[0051] Furthermore, while items 1 and 2 above can be expressed using general terminology, items 3 and 4 are likely to have unique wording and structures depending on the customer and product. Therefore, as a way to improve the accuracy of text mining, it is conceivable to use the product configuration dictionary 1450 and the project process / task dictionary 1460 shown in Figure 1 as dictionaries.

[0052] Furthermore, the mining basis generation subprogram 1413 generates analysis items that serve as the basis for extraction from the text data that has been subjected to text mining.

[0053] Furthermore, the graph generation subprogram 1414 creates a graph from the knowledge candidates extracted by text mining, representing them as an example of structured data known as knowledge. The graph generation subprogram 1414 then stores the created graph in the database 1480 of the storage unit 17.

[0054] Furthermore, the same quality information aggregation subprogram 1415 aggregates related knowledge into a single knowledge entry for the same defect when multiple defect reports have been text-mined. Note that the number of aggregated entries is not limited to one; it must be less than the number of entries before aggregation.

[0055] Furthermore, the quality information acquisition subprogram 1411 acquires the data of the defect report stored in the defect management system 1700 via the communication unit 11 .

[0056] Furthermore, the mining evidence generation subprogram 1413 generates the knowledge evidence described above. The details will be explained below using Figure 3. Figure 3 is a diagram illustrating the generation of knowledge evidence in this embodiment. Expressions extracted from the defect report as analysis items that are candidates for knowledge or candidates for verifiers are highlighted in Figure 3.

[0057] Here, text mining is performed by the text mining unit 142 (text mining subprogram 1412), but the accuracy cannot be guaranteed. It is desirable for the user to check the accuracy of the text mining results, and it is considered that when checking the accuracy, the user may read the report of the original data.

[0058] However, it is thought that it would be a burden for users to read through all of the long text to confirm the knowledge. However, by referring to the knowledge basis, users can easily see where the knowledge is written, such as the target product, target part, defect phenomenon, defect causal relationship, defect cause process, and task in the defect cause process. Then, by reading the defect report focusing on the context of these, it is possible to efficiently confirm the knowledge basis.

[0059] Therefore, in this embodiment, the mining basis generation subprogram 1413 generates a knowledge basis generation screen as shown in Fig. 3 and causes this to be displayed on the terminal device 1601. In other words, the mining basis generation unit 143 generates a knowledge basis generation screen and causes this to be displayed on the output unit 13 and / or on the input / output device 1600 via the communication unit 11.

[0060] In Figure 3, the highlights are shown in grayscale. However, during implementation, it may be possible to create classifications such as product or component names and defects, and set the highlight color for each classification. In this case, it is thought that the knowledge base can be visualized more easily.

[0061] Furthermore, the graph generation subprogram 1414 structures the extracted knowledge candidates to generate a graph, which is an example of knowledge. Then, the graph generation subprogram 1414 stores this in the database 1480. Here, the structure of graph data will be explained using FIG. 4. FIG. 4 is a diagram that schematically shows a graph in this embodiment. In this embodiment, knowledge is handled as data with a graph structure. Here, the graph structure is made up of nodes that store data such as objects and events, and edges that define the relationships between the nodes.

[0062] In addition, in FIG. 4, the relationship details are written near the edges. That is, in FIG. 4, nodes are represented by ellipses and edges by solid lines. The target product, target part, defect event, defect causal process, and task in the defect causal process are nodes. Meanwhile, the relationship between products and parts and the causal relationship of the defect are edges. In this way, the generated knowledge can be used not only for feedback to processes such as design, manufacturing, and maintenance, but also for users to confirm the accuracy of the knowledge. For example, knowledge stored in a relational database is displayed in a table format, but a graph illustrates relationships such as the relationship between products and parts and the causal relationship of the defect. This allows users to intuitively understand the causal relationship of the defect. Therefore, in this embodiment, it is desirable for the graph generation subprogram 1414 to display the graph shown in FIG. 4 on the terminal device 1601. In other words, the graph generation unit 144 displays the graph on the input / output device 1600 via the output unit 13 and / or the communication unit 11.

[0063] Here, this graph is created through the execution of code. In this embodiment, it is assumed that the graph generation subprogram 1414 is implemented by a generation AI, thereby eliminating the burden of the user having to manually write code. Furthermore, by employing a generation AI instead of a conventional AI, it is possible to generate graphs without detailed engineering according to the graph pattern. This also reduces the burden of graph generation. However, the extracted knowledge candidates and graphs, which are examples of generated knowledge, are not necessarily correct. For this reason, human confirmation, such as by a user, is desirable. In particular, there is a risk of hallucination occurring in the output of the generation AI, making human confirmation all the more important.

[0064] In this embodiment, it is assumed that the quality information acquisition subprogram 1411 can acquire multiple defect reports in one process, and in this case, multiple pieces of knowledge are also generated at once. It is conceivable that multiple defect reports with the same content will be acquired and knowledge extracted, and the processing in this case will be explained later.

[0065] Furthermore, when the text mining subprogram 1412 performs text mining on multiple defect reports in a single process, if defect reports for the same defect exist, the same quality information aggregation subprogram 1415 aggregates these into one knowledge. To this end, the same quality information aggregation subprogram 1415 determines whether defect reports for the same defect exist. At this time, in addition to the same defect, related defects (defects in the same location or defects of the same type) may also be aggregated.

[0066] Here, the aggregation process in the identical quality information aggregation subprogram 1415 will be explained using Figure 4. As mentioned above, Figure 4 schematically shows a graph that is an example of knowledge generated by the graph generation subprogram 1414. Here, we will take as an example a case where there are two defect reports, defect report 1 and defect report 2, and the defect is the same. First, the "defect report 1" node 4100 and the "defect report 2" node 4200 are both connected to the "fuel hose / pipe" node. Similarly, they are also connected to the "damaged, fuel leaking" node. By using this graph structure, the identical quality information aggregation subprogram 1415 can aggregate multiple identical defect reports into a single piece of knowledge.

[0067] It is believed that among the defect reports held by a company, there are multiple identical cases. If knowledge is generated for each of the same defect, multiple copies of the same knowledge will be created, which makes it difficult for users to check and utilize the knowledge. Therefore, by consolidating the knowledge into one knowledge as in this embodiment, the burden of checking on the user can be reduced.

[0068] Furthermore, an appropriate checker for the generated graph is identified by the knowledge checker information identification program 1420. To this end, the knowledge checker information identification program 1420 has the following subprograms.

[0069] First, the verifier identification subprogram 1421 identifies the work records in the product development project for the product in question, using the product, part, process, and tasks within the process that are the subject of the defect report as keys, and identifies the person in charge. Then, the user ID matching subprogram 1422 matches the identified username with the user dictionary to uniquely identify the user ID.

[0070] As mentioned above, the accuracy of the knowledge generated by the AI-powered knowledge generation program 1410 must be verified by the user. When the knowledge generation program 1410 performs text mining on the bug report, it can also mine the names of the verifiers from the signature field, etc. However, if the names of those verifiers are not listed in the bug report, the verifiers must be identified by other means, which can be burdensome for the user. In this case, the verifier identification subprogram 1421 can be used to reduce the burden of this task.

[0071] Next, we will describe the details of each of these subprograms. The verifier identification subprogram 1421 uses the product, parts, target process, and tasks in the target process, which have been text-mined by the knowledge generation processing program 1410, as keys to identify the work records in the product development project of the relevant product from the product development project management system 1800. Then, the verifier identification subprogram 1421 identifies the design engineer and quality assurance engineer. Here, the data structure of work records in a product development project and the search logic for work records for that product will be described using FIG. 5. FIG. 5 shows an excerpt from a table of work record data for a product development project managed by the product development project management system 1800 in this embodiment. In FIG. 5, the work record data is assumed to be composed of columns such as a product ID column 5100, a product name column 5200, a part ID column 5300, a part name column 5400, a process name column 5500, a task name column 5600, a designer column 5700, and a quality assurance technician column 5800. Therefore, the checker identification subprogram 1421 can identify the target work record by searching the work record using the above-mentioned data (task) as a key. Next, the checker identification subprogram 1421 can identify the person who checked the knowledge by referencing the data stored in the designer column 5700 and the quality assurance technician column 5800 in the identified work record.

[0072] Furthermore, the user ID matching subprogram 1422 matches the checker name identified by the checker identification subprogram 1421 with the user dictionary 1470 to uniquely identify the user ID of the checker. Here, the identification of the user ID will be described with reference to FIG. 6 . FIG. 6 is a diagram showing the user dictionary 1470 in this embodiment. In FIG. 6 , the user dictionary 1470 is assumed to be composed of columns such as a user ID column 6100, a name column 6200, and an affiliation column 6300. The checker identification subprogram 1421 searches the user dictionary 1470 using the identified checker name as a key and references the data stored in the user ID column 6100, thereby uniquely identifying the checker of the knowledge. Note that if there are multiple users with the same name, the checker identification subprogram 1421 performs a second search using affiliation information as a key to narrow down the user IDs. This reduces the search burden on the user when the checker of the knowledge cannot be easily identified.

[0073] Furthermore, the WF creation instruction program 1430 creates WF creation instruction data for knowledge confirmation and transmits it to the WF system 1900 via the communication device 1500. The contents of the WF creation instruction data will now be described with reference to FIG. 7 . FIG. 7 is a diagram illustrating an excerpt from a table of the WF creation instruction data in this embodiment. In FIG. 7 , the WF creation instruction data is assumed to be configured with columns such as an email subject column 7100, a defect report name column 7200, a checker 1_ID column 7300, a checker 2_ID column 7400, a link column 7500 to the target defect report, a link column 7600 to knowledge, a link column 7700 to the knowledge basis, and a confirmation request email body 7800. The email subject column 7100 stores data output by the generation AI as the subject of the confirmation WF based on the defect report name. The defect report name column 7200 stores the defect report name, which is the name of the defect report, which is text data mined by the knowledge generation processing program 1410. The checker 1_ID column 7300 and the checker 2_ID column 7400 store user ID data identified by the knowledge generation processing program 1410 or the knowledge checker information identification program 1420, respectively. Note that while FIG. 7 illustrates an example in which there are two checkers, the number may be other (for example, three or more). The confirmation request email body 7800 stores the knowledge confirmation request email body generated by the generation AI. Note that in FIG. 7, the WF creation instruction data is described in table format, but the data structure may be determined during implementation in accordance with the specifications of the destination WF system.

[0074] Here, the contents of the knowledge confirmation request email will be explained using Figure 8. In Figure 8, the body of the knowledge confirmation request email is assumed to consist of items such as a request text 8100, a link to the defect report 8200, a link to the knowledge 8300, and a link to the knowledge basis 8400. Note that by having a generation AI generate the email body, it is possible to reduce the burden of creating the email body according to patterns such as the number of defect reports.

[0075] By using the WF creation instruction data and knowledge confirmation request emails shown in Figures 7 and 8 above, the burden of constructing and initiating the knowledge confirmation flow can be reduced, and the burden of identifying reference information can be reduced by providing links to the extraction basis.

[0076] Furthermore, the knowledge distribution program 1440 distributes the knowledge accumulated in the database 1480 via the communication device 1500 to the terminal device 1601 used by the user who is responsible for tasks such as product and part design, production line startup, and product maintenance. Here, the knowledge in this embodiment is stored as a graph, with the target product, target part, defect event, defect cause, defect occurrence process, and defect occurrence task being held as nodes. Meanwhile, the relationship between the product and part and the causal relationship of the defect are held as edges. When data such as the product, part, and process for which the user is responsible is input to the knowledge generation support device 1100 from the terminal device 1601 or the like, the knowledge extracted from the database 1480 is distributed to the terminal device 1601. This allows past knowledge to be checked without omission, and by using it like a checklist for one's own tasks, it is expected that recurrence of defects will be prevented.

[0077] As explained in the knowledge generation processing program 1410, the knowledge in this embodiment has a graph structure that illustrates relationships between products and parts, and the causal relationships of defects. This allows users to intuitively understand the causal relationships of defects from the delivered knowledge without having to devise a display format. It is also conceivable that users would like to check not only the structured knowledge but also the original defect report and the description that served as the basis for extracting the knowledge. In this case, the knowledge basis generated by the mining basis generation subprogram 1413 can be used as auxiliary information. In this case, the knowledge basis generated by the mining basis generation subprogram 1413 can be notified to the terminal device 1601 via the communication device 1500. This allows users to confirm the burden of searching for original information and the burden of searching for important information from the original information, thereby reducing these burdens.

[0078] This concludes the explanation of the configuration of this embodiment. In the network 1000 shown in Figures 1A and 2B of this embodiment, the following types of communication can be used. A general public network, whether wired or wireless, such as the fifth-generation mobile communication system, or so-called 5G (5th Generation), which enables "massive simultaneous connections" and "ultra-low latency," can be used. Furthermore, by taking advantage of the features of new mobile phone systems beyond 5G, various types of data, including knowledge, can be distributed. In addition, even if the functions of the knowledge generation support device 1100 are distributed across more than 1000 of the network, processing delays in these interactions can be suppressed.

[0079] The knowledge creation support system of this embodiment may be implemented as an in-house system, or may be implemented as a so-called cloud system that can be used by multiple organizations (e.g., companies). In the latter case, the input / output device 1600 (e.g., terminal device 1601) is used by each of the multiple organizations.

[0080] <Processing flow> <Overall flow> Next, the processing flow in this embodiment will be explained. To this end, first, the overall flow from acquiring a defect report to storing the final version of knowledge in this embodiment will be explained using Fig. 9. In this case, the subject of each process will be explained using the configuration in Fig. 1A.

[0081] First, the quality information acquisition unit 141 acquires text data of a defect report from the defect management system 1700 via the communication unit 11 (step S9100).

[0082] Furthermore, the text mining unit 142 performs text mining on the acquired defect report to extract analysis items such as the name of the target product or target part, the causal relationship of the defect, the process that caused the defect, the task in the process that caused the defect, the name of the quality assurance person who reported the defect, and the name of the designer (step S9200). In other words, analysis items that are knowledge candidates or checker candidates are extracted. Furthermore, the mining basis generation unit 143 generates mining basis for the defect report, which is text data that has been text-mined, based on the analysis items (step S9300). Here, the mining basis includes data that highlights the defect report (FIG. 3) and the highlighted parts themselves.

[0083] Furthermore, when multiple defect reports are text mined, the identical quality information aggregating unit 145 determines whether identical defect reports exist, and if so, aggregates them into one knowledge (step S9400). Here, in step S9400, the identical quality information aggregating unit 145 aggregates, for example, defect reports with the same status and countermeasures but different defect report management numbers. Here, the identical defect (same defect) refers to a defect with the same or similar results in text mining. Similarity can be used to refer to a defect having a predetermined relationship. Furthermore, defect reports with the same device but different defect report management numbers may be aggregated.

[0084] Furthermore, the graph generating unit 144 structures the extracted knowledge candidates and generates a graph that is an example of knowledge (step S9500). Then, the graph generating unit 144 stores this graph in the storage unit 17.

[0085] Furthermore, the Knowledge Reviewer Information Identification Unit 146 identifies the reviewer for the graph, which is the generated knowledge, and the WF Creation Instruction Unit 15 creates WF creation instruction data (step S9600). The WF Creation Instruction Unit 15 then notifies the WF system 1900 via the communication unit 11 of the WF creation instruction data, including its distribution to the identified reviewer. As a result, the WF system 1900 creates a WF for the reviewer to review the graph according to the WF creation instruction data. Consequently, a review process for the graph (step S9700) is executed according to the WF.

[0086] First, the WF is deployed to the terminal device 1601 of the relevant verifier. In this case, the input / output device 1600 performs a verification process so that the user can view the graph (step S9701). That is, the input / output device 1600 displays a verification screen. This verification screen includes at least one of the generated knowledge graph (Figure 4), the knowledge basis generation screen (Figure 3), the defect report (Figure 2), and the knowledge verification request email. Furthermore, as an example of other displays, the display shown in Figure 11 may be shown. In Figure 11, the display screen 110 consists of a defect report area 111 and a knowledge verification request email area 112.

[0087] These are shown in Figures 2 and 8, respectively, and may be displayed as follows. For example, both may be displayed from the beginning, or the knowledge confirmation request email area 112 may be displayed first. In the latter case, when the user specifies the link destination of the defect report in the knowledge confirmation request email area 112, the corresponding defect report is displayed in the defect report area 111. It is desirable that the input / output device 1600 also displays the corresponding graph when the defect report area 111 is displayed. The display of this graph may be done according to the user's instructions or may be displayed automatically. As a result, the user can check the generated knowledge and the basis for that knowledge.

[0088] Furthermore, the input / output device 1600 receives the user's confirmation result for the relevant graph using the confirmation screen in step S9701 and determines whether approval for the relevant graph is possible (step S9702). If approval is possible (YES), the input / output device 1600 notifies the knowledge generation support device 1100 that approval is possible and proceeds to step S9800. If approval is not possible (NO), the process proceeds to step S9703.

[0089] Furthermore, if approval for the graph is not possible (S9702 is NO), the input / output device 1600 performs a rollback process (step S9703). For this purpose, the input / output device 1600 sends a message to the knowledge generation support device 1100 indicating that approval for the relevant graph is not possible. Then, the knowledge generation support device 1100 proceeds to one of the steps from step S9200 onwards. As a result, the knowledge generation support device 1100 modifies the parameters and other conditions in one of the steps in S9200 and executes it.

[0090] Furthermore, when the graph generation unit 144 of the knowledge creation support device 1100 receives through the communication unit 11 that approval of the graph is possible, the graph that has been approved is stored in the storage unit 17 (step S9800).

[0091] If approved, the input / output device 1600 may accept manual correction instructions for the graph from the user. In this case, in step S9800, the input / output device 1600 transmits the correction instructions or the corrected graph corresponding to the correction instructions to the knowledge generation support device 1100. Upon receiving this, the graph generation unit 144 may update the corresponding graph.

[0092] As a result of the above, in step S9800, the knowledge that has been confirmed by the confirmer and the confirmation record are recorded in DB 1480 of knowledge generation support device 1100.

[0093] <Example of flow for identifying the person who confirmed the information> Next, an example of the processing flow including the identification of the verifier in this embodiment will be described. Specifically, using Figure 10, the detailed flow from when the knowledge generation support device 1100 identifies the knowledge verifier, creates a WF creation instruction, to when the user verifies and confirms the knowledge will be described.

[0094] First, the text mining unit 142 determines whether the verifier's name has been identified (step S10100). If it has been identified (YES), the process proceeds to step S10400; otherwise, it proceeds to step S10200.

[0095] Furthermore, if the name of the verifier is not identified, the verifier identification unit 147 identifies the work record in the development project of the product in question, using the product, part, process, and tasks in the process that are the subject of the defect report as keys (step S10200).

[0096] 9. Furthermore, the checker identification unit 147 identifies the checker name from the work record of the identified development project (step S10300). Furthermore, the user ID matching unit 148 searches the user dictionary 1470 using the identified checker name and affiliation as keys to uniquely identify the user ID of the checker name identified in step S10300 (step S10400). These steps S10100 to S10400 correspond to the identification of the checker by the knowledge checker information identification unit 146 in step S9600 of FIG.

[0097] Furthermore, the WF creation instruction unit 15 creates WF creation instruction data for knowledge verification (step S10500). This step S10500 corresponds to the creation of WF creation instruction data by the WF creation instruction unit 15 in step S9600 of Figure 9.

[0098] Then, similar to step S9600 in Figure 9, the WF creation instruction unit 15 notifies the WF system 1900 of the WF creation instruction data via the communication unit 11.

[0099] Furthermore, the WF system 1900 creates a WF (Write-Flyer) for the verifier to review the graph according to the WF creation instruction data (step S10700). Then, the WF system 1900 deploys the created WF to the verifier's input / output device 1600.

[0100] In response, the input / output device 1600 performs a knowledge verification process similar to step S9700 in Figure 9 (step S10800). Note that the verification processes in steps S9700 and S10800 are processed by the WF system 1900 in response to user (verifier) ​​input to the input / output device 1600. However, in this application, it is described as processing by the input / output device 1600. The verification results are managed by the WF system 1900, and the WF system 1900 receives the user's verification results from the input / output device 1600 to the knowledge generation support device 1100 (step S10900). As a result, similar to step S9800 in Figure 9, the knowledge generation support device 1100 records the knowledge and verification records that have been verified by the user in the storage unit 17 (step S10600).

[0101] <Detailed flow of the confirmation process> Next, we will explain two verification processes (steps S9700 and S10800) when multiple verifiers are identified. First, as a first example, we will explain serial processing. Figure 12A is a flowchart of the verification process in this embodiment when executed using serial processing.

[0102] First, the WF system 1900 executes the following steps S10701 to S10703 as the WF creation process shown in step S10700 in Figure 10. Note that at least one of these steps may be executed by the WF creation instruction unit 15 or the input / output device 1600. The WF system 1900 then determines the order of confirmation for the multiple confirmers identified in steps S10100 to S10400 (step S10701). For this purpose, the WF system 1900 uses, for example, the roles and affiliations of the identified confirmers. In these cases, the order of creation of the defect report, such as creation-confirmation-approval, may be used as the confirmation order, or a priority order may be set for each affiliation, and the confirmation order may be determined according to that priority order.

[0103] Furthermore, the WF system 1900 extracts the highest-ranking verifier from among multiple verifiers (step S10702). Then, the WF system 1900 creates a verification request for the extracted highest-ranking verifier and notifies the corresponding input / output device 1600 of this (step S10703).

[0104] Next, the input / output device 1600 executes steps S10801 to S10804 as step S10800. First, the input / output device 1600 performs a confirmation process similar to step S9701 (step S10801). Also, the input / output device 1600 determines whether approval of the graph is possible, similar to step S9702 (step S10802). If the result of this determination is that approval is possible (YES), the process proceeds to step S10803.

[0105] The input / output device 1600 also determines whether there is a checker next higher than the identified highest-ranking checker. As a result, if there is a checker (YES), the process proceeds to step S10702, where the WF system 1900 extracts the next higher-ranking checker. If there is no checker (NO), the input / output device 1600 notifies the WF system 1900 that approval has been completed, and the process proceeds to step S10900. The input / output device 1600 also executes a return process (step S10804), similar to step S9703. By the above process, the confirmation process can be executed serially for multiple checkers.

[0106] Next, as a second example, an example in which confirmation processing is executed by parallel processing will be described. FIG. 12B is a flowchart when confirmation processing in this embodiment is executed by parallel processing. The WF system 1900 creates confirmation requests for multiple checkers as the WF creation processing in step S10700 shown in FIG. 10, and broadcasts the confirmation requests to the input / output devices 1600 of each checker. Then, in step S10800, the input / output devices 1600 of each checker each execute knowledge confirmation processing similar to steps S9700 and S10801 (step S10805). Then, each input / output device 1600 notifies the WF system 1900 of the confirmation result.

[0107] Furthermore, the WF system 1900 collects the confirmation results from each input / output device 1600 (step S11001). Then, the WF system 1900 determines whether all the confirmers can approve the confirmation results from each input / output device 1600 (step S11002). As a result, if all the confirmers can approve (YES), the process proceeds to step S10600. If there is a confirmer who cannot approve (NO), the process proceeds to step S1103, where the WF system 1900 executes a return process similar to steps S9703 and S10804 (step S11003). Note that step S11001 may be determined as YES if a predetermined number or percentage, such as half, of all the confirmers approve. By the above process, the confirmation process can be executed in parallel for multiple confirmers.

[0108] This concludes the description of this embodiment, but various modifications and applications are possible. For example, it can be applied to business operations other than product manufacturing, such as logistics, sales, and service provision. Furthermore, this embodiment enables a reduction in task load and rapid feedback on quality information. In particular, it can generate knowledge about defects from a large amount of past defect information. Moreover, it enables the identification of verifiers for the accuracy of the generated defect knowledge, the construction of verification processes, and the proper execution of verification work. Furthermore, organizations using this embodiment can contribute to reducing the cost of defect response, improving product quality, and ultimately maintaining their brand image. In addition, this embodiment can provide factory equipment and factory line construction services for improving product quality. [Explanation of symbols]

[0109] 1100: Knowledge generation support device 1200: Arithmetic device 1300: Main memory 1400:Auxiliary storage device 1410: Knowledge generation processing program 1411: Quality Information Acquisition Subprogram 1412: Text mining subprogram 1413: Mining evidence generation subprogram 1414: Graph generation subprogram 1415: Same quality information aggregation subprogram 1420: Knowledge checker information identification program 1421: Verifier Identification Subprogram 1422: User ID Verification Subprogram 1430: WF creation instruction program 1440: Knowledge Distribution Program 1450: Product Configuration Dictionary 1460: Project Process / Task Dictionary 1470: User Dictionary 1480: Database 1500: Communication device 1600: Input / output device 1700: Defect management system 1800: Product Development Project Management System 1900:WF System

Claims

1. A knowledge creation support device that supports the creation of business-related knowledge, a storage unit for storing information about multiple knowledge reviewers; an input unit that accepts text data related to the business; a text mining unit that analyzes the text data; a knowledge generation unit that generates the knowledge based on the analysis result by the text mining unit; The knowledge generation support device includes a verifying party specifying unit that specifies a verifying party of the generated knowledge based on the analyzed text data and the stored information on the verifying party of the knowledge.

2. 2. The knowledge creation support device according to claim 1, The knowledge creation support device further includes an identical quality information aggregation unit that aggregates multiple pieces of knowledge based on related text data.

3. 2. The knowledge creation support device according to claim 1, moreover, a mining basis generation unit that generates knowledge basis that serves as a basis for generating the knowledge; A knowledge creation support device having an output unit that outputs the knowledge basis.

4. 2. The knowledge creation support device according to claim 1, A knowledge creation support device in which a creation AI is implemented in the knowledge creation unit, and the knowledge is created by the creation AI.

5. 2. The knowledge creation support device according to claim 1, The knowledge generation unit is a graph generation unit that generates a graph having a graph structure as the knowledge.

6. A knowledge creation support method executed by a knowledge creation support device that supports the creation of business knowledge, comprising: storing information about multiple knowledge reviewers in a storage unit; an input unit that receives text data related to the business; a text mining unit that analyzes the text data; a knowledge generation unit that generates the knowledge based on the analysis result by the text mining unit; The knowledge generation support method further comprises a verifying party specifying unit specifying a verifying party of the generated knowledge based on the analyzed text data and stored information relating to the verifying party of the knowledge.

7. 7. The knowledge creation support method according to claim 6, Furthermore, the knowledge generation support method includes an identical quality information aggregation unit aggregating multiple pieces of knowledge based on related text data.

8. 7. The knowledge creation support method according to claim 6, moreover, a mining basis generation unit that generates a knowledge basis that serves as a basis for generating the knowledge; An output unit outputs the knowledge basis.

9. 7. The knowledge creation support method according to claim 6, A knowledge generation support method in which a generation AI is implemented in the knowledge generation unit, and the knowledge is generated by the generation AI.

10. 7. The knowledge creation support method according to claim 6, The knowledge generation support method, wherein the knowledge generation unit is a graph generation unit that generates a graph having a graph structure as the knowledge.

11. A knowledge generation support device that supports the generation of knowledge related to a certain business using a computer, a storage unit for storing information about multiple knowledge reviewers; an input unit that accepts text data related to the business; a text mining unit that analyzes the text data; a knowledge generation unit that generates the knowledge based on the analysis result by the text mining unit; a knowledge generation support program that functions as a verifyer specifying unit that specifies a verifyer of the generated knowledge based on the analyzed text data and the stored information on the verifyer of the knowledge;

12. 12. The knowledge creation support program according to claim 11, A knowledge creation support program that causes the knowledge creation support device to further function as an identical quality information aggregation unit that aggregates a plurality of pieces of knowledge based on related text data.

13. 12. The knowledge creation support program according to claim 11, The knowledge creation support device further comprises: a mining basis generation unit that generates knowledge basis that serves as a basis for generating the knowledge; a knowledge creation support program that functions as an output unit that outputs the knowledge basis;

14. 12. The knowledge creation support program according to claim 11, A knowledge generation support program in which a generation AI is implemented in the knowledge generation unit, and the knowledge is generated by the generation AI.

15. 12. The knowledge creation support program according to claim 11, The knowledge generation support program is a graph generation unit that generates a graph having a graph structure as the knowledge.

Images