System, method executed by system, and program

Abstract

To realize updating based on context of a knowledge graph when a document triggers updating of the knowledge graph, or to realize updating even if the document has been subjected to omission or the like.SOLUTION: A document information accepting section 102 accepts document information 162, which is information included in a document 161 that triggers updating of a knowledge graph 181. A knowledge graph information accepting section 107 accepts a knowledge graph information 167 from a knowledge graph system 180. A prompt constructing section 108 constructs a prompt 168 on the basis of the document information 162 and knowledge graph information 167. A response information accepting section 110 accepts response information 170 indicating a response from a language model system 190 to the prompt 168. An update query constructing section 111 constructs an update query 171 on the basis of the response information 170. The update query presenting section 112 presents the update query 171 to the knowledge graph system 180.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present disclosure relates to a technique for updating a knowledge graph.

Background Art

[0002] When managing and operating the infrastructure (infrastructure) of society and economy, including facilities (such as plants and factories), it has often been premised on documents and manual labor in the past. For example, knowledge about facilities and infrastructure is recorded in documents, or such knowledge is personally grasped by experts on the facilities and infrastructure. Moreover, the design, operation, and maintenance operations related to the facilities and infrastructure may be carried out based on documents or may be implemented depending personally on experts, which was the situation at that time. On the other hand, the sophistication of technologies used in facilities and infrastructure (improvement in the quality of knowledge) and the complication of system configurations in the facilities and infrastructure (increase in the amount of knowledge) are progressing. Therefore, it is expected that it will be difficult to appropriately carry out the design, operation, and maintenance operations related to the facilities and infrastructure by the method of managing knowledge about the facilities and infrastructure in documents. In addition, against the backdrop of the decline in the working population due to the progress of the declining birthrate and aging population, a shortage of experts is a concern in various fields, and it is also expected that the transfer of personal knowledge will become difficult.

[0003] In recent years, there has been an exploration of using a language model trained by machine learning (a language model with a huge number of model parameters and a large number of learning data used for training, also called a large language model (LLM)) in business and the like. When a prompt including the content of a question is input into the language model, the language model presents an answer to the question. It is expected that the language model will be utilized in the design, operation, and maintenance operations related to facilities and infrastructure to improve the efficiency of the operations. As methods for adapting a language model to be suitable for some business (adapting the language model to be suitable for queries (prompts) regarding specific domain knowledge), there are fine tuning and retrieval augmented generation (RAG). The fine tuning method requires the cost of re-learning in the language model (for example, the cost of re-learning for learning specific domain knowledge). In cases where the amount of training data used for re-learning is huge or the training data is added (changed) at any time, the fine tuning method is difficult to adopt. In the retrieval augmented generation (RAG) method, instead of re-learning the language model, the language model refers to external knowledge (for example, external knowledge including specific domain knowledge). As external knowledge referred to by the language model, there can be a vector store or a knowledge graph. Since one knowledge graph can contain knowledge included in multiple documents, it is suitable for responding to queries (prompts) regarding knowledge spanning multiple documents.

[0004] As a prior art document regarding a knowledge graph, there is Patent Document 1. Patent Document 1 (see, for example, FIG. 14) discloses a technique in which an operator inputs document data into a knowledge model creation support device, the knowledge model creation support device displays a plurality of terms in a GUI window based on the document data, the operator inputs a selection result of the terms via the GUI window, and when a predetermined condition is satisfied, the knowledge model creation support device displays the knowledge model (knowledge graph) in the GUI window in a manner of newly including the terms in the knowledge model (knowledge graph) and drawing them.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0006] When a language model using the Retrieval-Augmented Generation (RAG) method that externally references a knowledge graph is utilized in the business of designing, operating, and maintaining facilities and infrastructure, and the business efficiency is improved, the knowledge graph will be updated according to the update of the knowledge related to the facilities and infrastructure. Here, if the knowledge graph is updated manually based on the content of a document (such as a maintenance document) that shows the content of the update of the knowledge related to the facilities and infrastructure, it will incur human costs and time costs. For example, although the technology disclosed in Patent Document 1 supports the update of a knowledge graph (knowledge model in the terms of Patent Document 1) according to the content of document data, it requires judgment and labor by the operator. From the perspective of the continuous operation of the knowledge graph, it is desirable to reduce the proportion of human involvement.

[0007] Therefore, it can be considered to utilize a language model for updating the knowledge graph. Specifically, it can be considered to automatically construct an update query that acts on the knowledge graph, which reflects the content of a document (such as a maintenance document) that shows the content of the update of the knowledge related to the facilities and infrastructure and conforms to the specifications of the knowledge graph, by utilizing a language model (as much as possible). However, there may be several problems in constructing an update query based on the content of a document by utilizing a language model.

[0008] First, in order for a language model to generate an answer with the expected content (an answer suitable for the purpose of constructing an update query), it is often necessary to devise the content of the prompt input to the language model. For example, when the language model has not been trained by machine learning specialized in a specific knowledge domain, it is often desirable that the prompt input to the language model includes knowledge (hints) related to the specific knowledge domain. Here, when updating the knowledge graph in response to the update of knowledge related to facilities and infrastructure, depending on the type of knowledge domain related to facilities and infrastructure and the knowledge graph to be updated, the knowledge (hints) that are appropriate to include in the prompt may be determined (in some cases, the content of the prompt other than knowledge (hints) may be determined). When constructing the prompt manually, there is a risk that the creator will create the prompt through trial and error. It is desirable that the construction of the prompt with content according to the type of knowledge domain and the knowledge graph to be updated is realized while maintaining a high degree of automation of the process. And it is desirable that the update of the knowledge graph is realized along with the context of the knowledge graph.

[0009] In addition, a document (for example, a preservation document) indicating the content of the update of knowledge related to facilities and infrastructure may not be created from the viewpoint of having appropriate and sufficient description content for appropriately updating the knowledge graph. Specifically, the description content of the document (for example, a preservation document) may be short or fragmentary. In such a document, the information regarding the nodes and edges in the knowledge graph to be updated (for example, the information regarding the assets included in the facilities and infrastructure) may not be appropriately and sufficiently described (some descriptions may be omitted). For example, there may be a situation where the description regarding the relationship between assets is missing in the document, or a situation where the type of the asset is described in the document but the description of the identifier of the asset is missing. In addition, for example, when there are multiple persons who can create a document (for example, a preservation document), there may be variations in the notation of terms described in the document. For example, one person may describe the official name of the asset included in the facilities and infrastructure in Japanese in the document, another person may describe the alias of the asset in Japanese in the document, and still another person may describe the abbreviation of the asset in alphabet in the document. Furthermore, for example, in a document (e.g., a preservation document), it may be the case that symptoms occurring in assets included in a facility or infrastructure, the causes of such symptoms, and measures taken for such symptoms are not set forth in a set. With only such a document, it may be the case that symptoms occurring in the asset, the causes of such symptoms, and measures taken for such symptoms are not appropriately grasped. It is desirable that, while appropriately supplementing or correcting the content of a document (e.g., a preservation document) showing the content of the update of knowledge regarding a facility or infrastructure, the construction of a prompt with appropriate content is realized while maintaining a high degree of automation of processing. Note that the technology disclosed in Patent Document 1 does not construct a prompt after automatically supplementing or correcting the content of document data.

[0010] As described above, the knowledge graph has been regarded as the destination to which a language model makes an external reference by the method of Retrieval-Augmented Generation (RAG). Also, it has been mentioned that the knowledge graph is updated according to a document (e.g., a preservation document) showing the content of the update of knowledge regarding a facility or infrastructure. However, the knowledge graph to be updated may be other than for the purpose of external reference by the language model by the method of Retrieval-Augmented Generation (RAG). Also, the knowledge graph may be utilized in a digital twin or the like. Also, the target of the knowledge represented by the knowledge graph to be updated may be other than a facility or infrastructure. The document may be any as long as it triggers the update of the knowledge graph. Generally, it is something that updates the knowledge graph (while maintaining a high degree of automation of the update) triggered by a document, and it can be an issue to realize the update of the knowledge graph along with the context of the knowledge graph, or to realize the update of the knowledge graph even if there are omissions or the like in the description content of the document. In such a case, the use of the knowledge graph, the target of the knowledge represented by the knowledge graph, and the type of document that triggers the update of the knowledge graph may be arbitrary.

[0011] Based on the above, when the knowledge graph is updated triggered by a document, one of the objectives of the present disclosure is to maintain a high degree of automation of the update and to realize the update in accordance with the context of the knowledge graph or to realize the update even if there are omissions or the like in the description content of the document.

Means for Solving the Problem

[0012] In order to achieve at least one of the above objectives, the features that the present disclosure may include are as follows, for example. One aspect of the present disclosure is a system. The system includes a document information reception unit, a knowledge graph information reception unit, a prompt construction unit, a prompt presentation unit, an answer information reception unit, an update query construction unit, and an update query presentation unit. The document information reception unit receives document information, which is information included in a document that triggers an update of the knowledge graph. The knowledge graph information reception unit receives knowledge graph information, which is information included in the knowledge graph, from a knowledge graph system that manages the knowledge graph. The prompt construction unit constructs a prompt based on the document information and the knowledge graph information. The prompt is for requesting the presentation of information used for constructing an update query for updating the knowledge graph. The prompt presentation unit presents the prompt to a language model system that manages a language model. The answer information reception unit receives answer information indicating an answer from the language model system to the prompt. The update query construction unit constructs an update query based on the answer information. The update query presentation unit presents the update query to the knowledge graph system.

Advantages of the Invention

[0013] As described above, the present disclosure constructs a prompt based on document information, which is information included in a document that triggers an update of the knowledge graph, and knowledge graph information, which is information included in the knowledge graph. Here, since the knowledge graph information indicates the context of the knowledge graph, the prompt constructed based on the knowledge graph information can be appropriate along with the context of the knowledge graph. Or, even if there are omissions or the like in the description of the document indicated by the document information, based on the knowledge graph information, it is possible to construct an appropriate prompt after substantially supplementing or correcting the part where the omission or the like has been made. Answer information and update queries corresponding to such appropriate prompts are also appropriate. That is, the update of the knowledge graph is also appropriately realized.

[0014] Further, in response to receiving document information, which is information included in a document that triggers an update of the knowledge graph, the present disclosure performs a series of processes including receiving knowledge graph information from the knowledge graph system, constructing a prompt, presenting the prompt to the language model system, receiving answer information from the language model system, constructing an update query, and presenting the update query to the knowledge graph system, and maintains a high degree of automation of the processes from receiving the document information to updating the knowledge graph.

[0015] As described above, when the knowledge graph is updated triggered by a document, the present disclosure maintains a high degree of automation of the update and realizes the update along with the context of the knowledge graph or realizes the update even if there are omissions or the like in the description of the document.

[0016] A method or program that realizes the same thing as the process realized by the above system can also obtain the same operational effects as the above system. In the case of a program aspect, costs are often reduced. In a program, it is also easy to make design changes related to the process. Features that the present disclosure may have other than those described above, and the effects corresponding to such features are disclosed in this specification, the claims, or the drawings.

Brief Description of the Drawings

[0017]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Figure 19

Figure 20

Figure 21

Figure 22

Figure 23

Figure 24

Figure 25

Figure 26

Figure 27

Figure 28

Figure 29

Mode for Carrying Out the Invention

[0018] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. Note that the embodiments described below do not limit the disclosure according to the claims, and not all of the elements and combinations thereof described in the embodiments are essential for the solution means of the present disclosure. The following description and drawings are examples for explaining the present disclosure, and for the sake of clarity of explanation, appropriate omissions and simplifications are made. The present disclosure can be implemented in various other forms. Unless otherwise limited, each component may be singular or plural. The positions, sizes, shapes, ranges, etc. of the components shown in the drawings may not represent the actual positions, sizes, shapes, ranges, etc. in order to facilitate the understanding of the invention. For this reason, the present disclosure is not necessarily limited to the positions, sizes, shapes, ranges, etc. disclosed in the drawings. Each of the systems, apparatuses, or functional units of the present disclosure may be integrated into one piece in terms of hardware, or may be divided into a plurality of parts that cooperate with each other to perform a role. Some systems, apparatuses, or functional units may be integrated in terms of hardware. Each of the systems, apparatuses, or functional units may be realized by causing a computer to execute software (program) (as shown in FIG. 2). Some of the functions of the systems, apparatuses, or functional units may be realized by hardware (for example, hardwired logic or a field programmable gate array (FPGA)), and the remaining functions may be realized by executing software (program). All of the functions of each of the systems, apparatuses, or functional units may be realized in terms of hardware. Some or all of the steps shown in the flowcharts and the like described in the present disclosure may be realized in terms of hardware. Each of the systems, apparatuses, or functional units of the present disclosure may be realized from one or more hardware resources. For this purpose, each of the systems, apparatuses, or functional units of the present disclosure may be virtually realized. For example, virtual computer or container techniques may be used. The program of the present disclosure is included in the concept of what generally corresponds to the software such that a unique information processing system (system) or its operation method according to the purpose of use is constructed by the cooperation of software and hardware resources. That is, the program of the present disclosure is not limited to a specific type or form of program. Also, the program may initially be recorded in a compressed form. Those in which the same reference numerals are used in a plurality of drawings are similar to each other. In the drawings showing flowcharts, rectangular boxes indicate processing steps, and hexagonal boxes indicate conditional branch steps. In the drawings showing flowcharts, "step" is abbreviated as "S". Also, the display or output modes shown by the drawings are merely examples. Within the scope of the gist of the present disclosure, the display or output modes may be different from those shown in the drawings.

[0019] 1. Basic functional configuration (Figure 1) Figure 1 shows the basic functional configuration 100 (and the information handled) of the system according to an embodiment of the present disclosure. Note that not all the functional configurations shown in Figure 1 are essential. Also, it is not precluded that there are functional configurations other than those shown in Figure 1.

[0020] System 101 may be communicable with knowledge graph system 180 and language model system 190. Knowledge graph system 180 manages knowledge graph 181. Knowledge graph system 180 responds to requests for information that knowledge graph 181 has. Also, knowledge graph system 180 updates knowledge graph 181 in response to update query 171 for knowledge graph 181. Here, the update of knowledge graph 181 may be an update of information related to nodes, edges, and subgraphs already existing in knowledge graph 181, or may be an update to add new nodes, edges, and subgraphs to knowledge graph 181. Knowledge graph 181 may represent, for example, knowledge about facilities and infrastructure, but the content of the knowledge represented by knowledge graph 181 is not particularly limited. Language model system 190 manages language model 191. When prompt 168 including the content of a question is presented, language model 191 generates (outputs) answer information 170 indicating an answer to the question. Language model 191 is constructed by training using machine learning. However, language model 191 is not necessarily trained by machine learning specialized in domain knowledge to which the knowledge represented by knowledge graph 181 belongs. When the number of model parameters that language model 191 has is huge or when the number of learning data used for training language model 191 is huge, language model 191 may be called a large language model (LLM). Note that in Figure 1, knowledge graph 181 and language model 191 are shown as existing outside system 101, but one or both of knowledge graph 181 and language model 191 may be inside system 101.

[0021] System 101 presents an update query 171 that requests an update to the knowledge graph 181 in response to a document 161 that triggers an update to the knowledge graph 181. Here, the document 161 may be, for example, a security document for a facility or infrastructure, but the type of the document 161 is not particularly limited. For example, it may be other types of documents for a facility or infrastructure, or it may be a document for something other than a facility or infrastructure. As functional units, system 101 may include a document information reception unit 102, a knowledge graph information reception unit 107, a prompt construction unit 108, a prompt presentation unit 109, a response information reception unit 110, an update query construction unit 111, and an update query presentation unit 112. Each of these functional units may be realized by executing a program, or may be more hardware-implemented.

[0022] The document information reception unit 102 receives document information 162 that is information included in the document 161. Here, the contents of the document 161 and the document information 162 may be, for example, as shown in FIGS. 9, 11, 13, or 15 described later. Note that the entity that provides the document information 162 to the system 101 may be any client system that can communicate with the system 101. Alternatively, a person (e.g., an operator) who handles a facility, infrastructure, etc. related to the knowledge represented by the knowledge graph 181 may manually provide the document information 162 (document 161) directly to the system 101 via the input device 206 (see FIG. 2).

[0023] The knowledge graph information reception unit 107 receives knowledge graph information 167 that is information included in the knowledge graph 181 from the knowledge graph system 180. The contents of the knowledge graph information 167 may be, for example, information related to the nodes, edges, and subgraphs of the knowledge graph 181. The knowledge graph information 167 may be, for example, in the form of hint information or hint set information. The contents of the knowledge graph information 167 may be, for example, as shown in FIGS. 3, 4, 5, 9, 11, 13, or 15 described later.

[0024] The prompt construction unit 108 constructs a prompt 168 based on the document information 162 and the knowledge graph information 167. The prompt 168 is for requesting the presentation of information (answer information 170) used for constructing an update query 171 for updating the knowledge graph 181 to the language model 191. The prompt 168 is based not only on the document information 162 which is the information included in the document 161 that triggers the update of the knowledge graph 181, but also on the knowledge graph information 167. By constructing the prompt 168 based on the knowledge graph information 167, it is expected that the prompt 168 along the context of the knowledge graph 181 is constructed, or the prompt 168 is constructed after supplementing or correcting the parts omitted etc. in the document information 162. The prompt construction unit 108 may construct the prompt 168, for example, by inserting the document information 162 and the knowledge graph information 167 (hint set information 701 in FIG. 9) into the corresponding part of the prompt template 901 as shown in FIG. 9 described later. However, as long as the document information 162 and the knowledge graph information 167 are used to construct the prompt 168, the construction method of the prompt 168 is not limited.

[0025] The prompt presentation unit 109 presents the prompt 168 to the language model system 190. Accordingly, the language model system 190 inputs the prompt 168 to the language model 191. The language model 191 generates an answer to the question indicated by the prompt 168 and outputs it as the answer information 170. The language model system 190 provides the answer information 170 to the system 101. The response information reception unit 110 receives the response information 170 from the language model system 190. The response information 170 may be information used for constructing the update query 171, and the format of the response information 170 may be arbitrary. That is, the format of the response information 170 does not have to be the same as the format of the update query 171 itself. The response information 170 may be, for example, in a form indicating classes in the PlantUML format as shown in FIG. 9 described later (e.g., a class diagram), or may be in a form of a sentence or a notation using symbols as shown in FIGS. 11, 13, or 15 described later.

[0026] The update query construction unit 111 constructs the update query 171 based on the response information 170. The update query construction unit 111 constructs the update query 171 in a format acceptable to the knowledge graph system 180. For example, if the knowledge graph system 180 is in the form of a certain database system, the update query 171 may be a sentence in a format acceptable to the database system (e.g., an SQL (Structured Query Language) sentence). Also, for example, if the knowledge graph system 180 includes a control device that controls the recording medium in which the knowledge graph 181 is stored and does not perform complex control, the update query 171 may be in a form that instructs a direct operation on the internal structure of the knowledge graph 181 (a structure such as a table that holds information related to nodes, edges, subgraphs, etc.). In addition, when the format of the response information 170 is the same as the format of the update query 171, the update query construction unit 111 may not exist, or the update query construction unit 111 may simply pass through the response information 170 and treat it as the update query 171.

[0027] The update query presentation unit 112 presents an update query 171 to the knowledge graph system 180. The knowledge graph system 180 updates the knowledge graph 181 according to the update content indicated by the update query 171. The update of the knowledge graph 181 may be an update of information related to nodes, edges, and subgraphs already existing in the knowledge graph 181 (such as in FIG. 11 described later), or may be an update indicating the addition of new nodes, edges, and subgraphs to the knowledge graph 181 (such as in FIG. 13 or FIG. 15 described later). As described above, since it is expected that an appropriate prompt 168 can be constructed, the response information 170, update query 171 obtained based on the prompt 168, and the update of the knowledge graph 181 based on the update query 171 are also expected to be in line with the context of the knowledge graph 181, or to reflect the supplementation or correction of parts omitted in the document information 162.

[0028] Since the system 101 in the embodiment of the present disclosure has the above functional configuration, it can achieve the effects shown in the above [Effects of the Invention].

[0029] 2. Computer Architecture for Implementing the Embodiment of the Present Disclosure (FIG. 2) FIG. 2 shows a computer architecture 200 for implementing the system 101 of the embodiment of the present disclosure. To implement the system 101, part or all of an information processing device 201, a storage device 202, a non-volatile recording medium (recording device) 203, an external recording medium drive 204, an input device 206, a display or output device 207, a communication device 208, an external input / output port 209, and a reading device 210 may be interconnected by an interconnecting unit 211. (Note that part or all of the interconnecting unit 211 may be a network. In that case, the system 101 is implemented by a plurality of devices via the network.) The information processing device 201 may be, for example, a processor. Examples of this processor include a CPU, an MPU, or a GPU. Alternatively, the processor referred to here may be another semiconductor device as long as it is a main body that executes predetermined processing. Further, the information processing device 201 may be one or more (micro)processors. The storage device 202 may be, for example, a memory. The non-volatile recording medium (recording device) 203 may be, for example, a non-volatile memory (e.g., flash memory) or a non-volatile disk device. The external recording medium drive 204 may be, for example, a disk drive. The input device 206 may be, for example, a mouse, a keyboard, an imaging device, a sensor, a touch panel, or a pointing device. The display or output device 207 may be, for example, a display, a printer, or a speaker. The communication device 208 may be, for example, a communication device for wired communication or a communication device for wireless communication. The communication device 208 may be a network interface device (NIC) that controls communication with other systems, devices, terminals, or servers according to a predetermined protocol. The interconnecting section 211 may be, for example, a bus or a crossbar switch. (As described above, part or all of the interconnecting section 211 may be a network.)

[0030] In the non-volatile recording medium (recording device) 203, various programs included in the program group 231 (for example, programs for realizing the functional configurations according to the present disclosure. For example, various programs for implementing each of the functional units realized in the system 101.), various data groups included in the data group 232, or various information 233 may be recorded. The program group 231 may include each of various programs for realizing each of the functional units referred to as "units" in the functional configuration diagrams or flowcharts of FIGS. 1, 6, 7, 8, 16, 17, 18, 20, 22, 24, 26, 28, or 29. Note that some of the above programs may be integrated into one program. Also, any of the above programs may be divided into a plurality of programs. The data group 232 may include information (data, etc.) handled by the above functional units. Alternatively, various programs included in the program group 231 described above, various data groups included in the data group 232, or part or all of the information in the various information 233 may be acquired from outside the configuration shown in FIG. 2.

[0031] The external recording medium drive 204 can connect an external recording medium 205. The external recording medium 205 may be, for example, a portable recording disk (such as a DVD), an IC card, an SD card, a non-volatile memory (such as a flash memory), or a portable hard disk. Incidentally, from this external recording medium 205, various programs included in the program group 231, various data included in the data group 232, etc., or information similar to the information in the various information 233 may be transferred and stored in the non-volatile recording medium (recording device) 203 or the storage device 202. The external recording medium 205 may be used to record programs and data handled in the system 101. The external recording medium drive 204 and the external recording medium 205 may be connected to the system 101 shown in FIG. 2 via a network. Various programs included in the program group 231, various data such as those included in the data group 232, or the information in the various information 233 may be provided via the communication device 208, the external input / output port 209, the input device 206, and the reading device 210, and recorded or stored in the non-volatile recording medium (recording device) 203 or the storage device 202.

[0032] For the architecture of FIG. 2 to function as system 101, each functional unit within system 101, or a part of each functional unit (executing one or a series of processes (steps)), each of the various programs included in program group 231 may be loaded into storage device 202 (e.g., from non-volatile recording medium (recording device) 203). The program after loading is indicated by 221 in FIG. 2. Then, information processing device 201 may execute program 221 (using various data etc. included in data group 232 existing in non-volatile recording medium (recording device) 203 etc. as necessary, or information of various information 233). By executing program 221, the functions of system 101, each functional unit within system 101, or a part of each functional unit are realized (one or a series of processes (steps) are executed). Various buffers 223 temporarily formed in storage device 202 at this time may also be used as appropriate.

[0033] 3. Examples of the Structure of the Knowledge Graph (FIGS. 3 to 5) Before the processing performed by the embodiments of the present disclosure is described in detail, examples of the structure of the knowledge graph that is the target updated by the present disclosure will be described below. Knowledge graph 181 may be composed of a schema and an instance group. The schema may indicate a pre-definition of nodes and edges that may be included in knowledge graph 181, and a pre-definition of the relationships that may exist between nodes and edges. The schema may be, for example, as shown in FIGS. 10, 12, or 14 described later. On the other hand, the instance group may indicate the current situation etc. actually grasped in the object of knowledge represented by knowledge graph 181 (e.g., facilities and infrastructure). The instance group may be, for example, as shown in FIG. 3, FIGS. 10, 12, or 14 described later.

[0034] 3.1. Examples of the Types of Nodes and Edges Included in the Knowledge Graph Figure 3 shows an example 300 of the structure that the knowledge graph 181 may have. The example in Figure 3 shows an instance group 301 that represents knowledge about a certain water supply mechanism. Although not shown in Figure 3, there may be a schema (for example, the schema 1401 shown in Figure 14) that shows the predefined definitions for the nodes and edges included in this instance group 301. As shown in Figure 3, the knowledge graph 181 may include one or more nodes and one or more edges that indicate the connection relationships between the nodes. In the example of Figure 3, the nodes are represented by square frames, and the edges are represented by arrows. (The edges may be those that connect between nodes without directionality (undirected edges).) In a knowledge graph for representing knowledge about a facility or infrastructure, there may be asset nodes, symptom nodes, and failure mode nodes as types of nodes.

[0035] Asset nodes indicate assets (assets, valuable items) that exist in a facility or infrastructure. In the example of Figure 3, there are asset node 311, asset node 312, and asset node 313. For example, asset node 311 may represent a specific water supply mechanism that exists in a facility or infrastructure. Asset node 312 may represent a specific pump that the specific water supply mechanism has. Asset node 313 may represent a specific impeller (or motor) that the specific pump has. The edges between asset nodes may indicate an inclusion relationship (hierarchical relationship), like the "has_a (owns)" type of edge in Figure 3. In the example of Figure 3, the specific water supply mechanism represented by asset node 311 includes the specific pump represented by asset node 312. Also, in the example of Figure 3, the specific pump represented by asset node 312 includes the specific impeller represented by asset node 313. As the edges between asset nodes, there may be other types of edges. For example, the fact that assets are connected to each other may be indicated by an undirected edge.

[0036] Symptom nodes represent the symptoms that an asset can have. Symptom nodes included in an instance group may represent symptoms that have actually occurred in the past or present, or may occur in the future, in a specific asset represented by an asset node included in the same instance group. (In contrast, symptom schema nodes included in a schema may comprehensively pre-define the symptoms that can occur in an asset.) In the example of FIG. 3, symptom node 314 may indicate the symptom of water supply function stop (in the water supply mechanism). In other examples, symptom node 314 may indicate the symptom of water supply function degradation (in the water supply mechanism). Here, the asset node 311 indicating a specific water supply mechanism and the symptom node 314 indicating the symptom of the specific water supply mechanism may be connected by an edge of the type "symptom (symptom)". Similarly, in the example of FIG. 3, symptom node 315 may indicate the symptom of pump stop (in the pump). In other examples, symptom node 315 may indicate the symptom of reduced discharge volume (in the pump). Here, the asset node 312 indicating a specific pump and the symptom node 315 indicating the symptom of the specific pump may be connected by an edge of the type "symptom (symptom)".

[0037] Failure mode nodes represent the failure modes that an asset can assume. Failure mode nodes included in an instance group may represent failure modes that have actually occurred in the past or present, or may occur in the future, in a specific asset represented by an asset node included in the same instance group. (In contrast, failure mode schema nodes included in a schema may comprehensively pre-define the failure modes that an asset can assume.) In the example of FIG. 3, failure mode node 316 may indicate the failure mode of impeller breakage (in the impeller). Here, the asset node 313 indicating a specific impeller and the failure mode node 316 indicating the failure mode of the specific impeller may be connected by an edge of the type "at (location)".

[0038] Edges may also be set between symptom nodes and fault mode nodes. In the example of FIG. 3, a symptom node 314 representing a symptom of a water supply mechanism and a fault mode node 316 representing a fault mode of an impeller may be connected by an edge of the type "reason". For example, the edge of the type "reason" represents that one of the possible reasons for the symptom of the water supply function stop in the water supply mechanism is the fault mode of the impeller breakage in the impeller of the water supply mechanism. Similarly, in the example of FIG. 3, a symptom node 315 representing a symptom of a pump and a fault mode node 316 representing a fault mode of an impeller may be connected by an edge of the type "reason". For example, the edge of the type "reason" represents that one of the possible reasons for the symptom of the reduced discharge amount in the pump is the fault mode of the impeller breakage in the impeller of the pump. In addition, edges of the type "reason" may also be set between symptom nodes or between fault mode nodes. Also, depending on the object of knowledge (for example, a facility or infrastructure) represented by the knowledge graph 181, the distinction between regarding an event in an asset as a "symptom" or regarding the event in the asset as a "fault mode" may become ambiguous. In that case, the knowledge graph 181 may be constructed without clearly distinguishing the node types between the symptom node and the fault mode node.

[0039] As described above, the knowledge graph can include asset nodes, symptom nodes, fault mode nodes, and edges indicating the relationships between these nodes (relationships between nodes of the same type and relationships between nodes of different types). Therefore, for one or more elements (for example, assets) included in the object (for example, a facility or infrastructure) represented by the knowledge graph, the knowledge graph can represent specific knowledge, including the relationships between the elements (assets) and the symptoms and fault modes that can occur in the elements (assets).

[0040] 2. Examples of Node Information and Edge Information in 3 Each node and each edge included in the knowledge graph 181 (schema or instance group) may be accompanied by respective node information which is information regarding the node and respective edge information which is information regarding the edge. How the node information and the edge information are recorded on the recording medium possessed by the knowledge graph system 180 is arbitrary. For example, as shown in FIGS. 4 and 5, the node information and the edge information may be recorded in the form of a table having records for each node and each edge. Or, the node information and the edge information may be managed on the recording medium possessed by the knowledge graph system 180 in such a manner that respective pieces of node information (edge information) regarding the same node (edge) are connected to each other by links.

[0041] FIG. 4 shows an example 400 of node information. FIG. 4 corresponds to the example of the node shown in FIG. 3. As shown in FIG. 4, the node information may be recorded in the form of a table. The table in FIG. 4 may have a record (one row in the table) for each node. One record may have a "node ID" 401 and a "node type / node name" 402 and may have various pieces of information. The "node ID" 401 is an identifier capable of identifying which node in the knowledge graph 181. In the examples of FIGS. 3 and 4, N1 to N6 are used as the "node ID". The "node type / node name" 402 indicates the type of the node and the name of the node. For example, as the type of the node, it may be possible to identify whether it is an asset node, a symptom node, or a failure mode node. Also, for the asset node, an asset identifier (P001, P002B, P003B in the example of FIG. 4) may be included in the "node name". In the example of FIG. 4, as the various pieces of information included in one record, "various information 1" 403 and "various information 2" 404 are shown. However, there is no limit to the number of the various pieces of information included in one record. In the example of FIG. 4, for asset nodes with node IDs N1, N2, and N3, measurement values related to the asset may exist as "various information 1" 403. As shown in FIG. 4, for example, a measurement value x of the water supply capacity may exist for the water supply mechanism, which is an asset. A measurement value y of the discharge volume and a cumulative operation time m may exist for the pump, which is an asset. A measurement value z of the rotation speed and a cumulative operation time n may exist for the impeller, which is an asset. Note that various types of information may exist as the various information for the asset node. Examples of the various information (information included in the asset node information) for the asset node will be described later. In the example of FIG. 4, for symptom nodes or failure mode nodes with node IDs N4, N5, and N6, judgment criterion information for determining whether or not it corresponds to a symptom or a failure mode may exist as "various information 1" 403. As shown in FIG. 4, for example, information such as "the water supply capacity value is less than X" may exist as judgment criterion information for determining that the water supply mechanism has a symptom of a decrease in water supply capacity. Information such as "the discharge volume is less than Y" may exist as judgment criterion information for determining that the pump has a symptom of a decrease in discharge volume. Information such as "the rotation speed is less than Z" may exist as judgment criterion information for determining that the impeller is in a failure mode of impeller damage. Note that various types of information may exist as the various information for the symptom node or the failure mode node.

[0042] FIG. 5 shows an example 500 of edge information. FIG. 5 corresponds to the example of the edge shown in FIG. 3. As shown in FIG. 5, the edge information may be recorded in the form of a table. The table in FIG. 5 may have a record (one row in the table) for each edge. One record may have an "Edge ID" 501, an "Edge Type" 502, a "FromID" 503, a "ToID" 504, and may have various types of information. The "Edge ID" 501 is an identifier capable of identifying any edge in the knowledge graph 181. In the examples of FIGS. 3 and 5, E1 to E7 are used as the "Edge ID". The "Edge Type" 502 indicates the type of the edge. For example, as the type of the edge, at least any one of "has_a (ownership)", "symptom", "reason", "at (location)" may be identifiable. As the type of the edge, "is_a (instantiation)" may also be identifiable. The "FromID" 503 indicates the identifier of the node that is the starting point of the edge. The "ToID" 504 indicates the identifier of the node that is the arrival point of the edge. In the example of FIG. 5, for the edge with the edge ID of E1, the starting point of the edge is the node with the node ID of N1, and the arrival point of the edge is the node with the node ID of N2. In addition, when the edge is undirected, the "FromID" 503 and the "ToID" 504 may indicate the identifiers of the nodes corresponding to the two endpoints of the undirected edge. In addition, as various types of information for the edge, various types of information may exist. Also, such various types of information may be called edge hint information.

[0043] 3. Information that an asset node can have (asset node information) of 3 Various types of node information (asset node information) may be associated with an asset node for expressing knowledge about the asset. Examples of node information (asset node information) that can be associated with each of the asset nodes are shown in the node information group 321 of FIG. 3. The asset node information may be, for example, information related to asset attributes, information related to asset configurations, node hint information, hint set information, search range information, past document information, or past difference information.

[0044] The information regarding asset attributes may be, for example, information on the specifications (performance and functions) of the asset, information on measurement values in the asset, information on the criteria for replacing the asset, or alias list information of the asset name to cope with variations in the notation of the asset name. By having information regarding asset attributes as asset node information, the knowledge graph can hold knowledge about various attributes of the asset individually and specifically. The information regarding asset composition may be, for example, information indicating the relationship between assets. Note that the information indicating the relationship between assets interconnected by an edge may be indicated by a pair of "FromID" 503 and "ToID" 504, which is the edge information corresponding to the edge, or may be indicated by "information regarding asset composition", which is one of the asset node information. For example, the "information regarding asset composition", which is the asset node information related to the asset node indicating a pump, may be the information of the sentence "A pump is composed of a motor, an impeller, and a coupling." Also, in the knowledge graph 181, the "information regarding asset composition", which is the asset node information related to an asset node in a relatively higher hierarchy, may indicate the information indicating the relationship between assets in a group of asset nodes in a relatively lower hierarchy. In this way, the knowledge graph can not only hold the information indicating the relationship between assets as edge information but also hold it as asset node information, so it has flexibility regarding the mode of holding information.

[0045] The node hint information is information about the object (in this case, the asset) represented by the node (in this case, the asset node) associated with the node hint information. The node hint information may be any information as long as it serves as a hint for the language model 191 when the language model 191 generates the response information 170, which is the response to the prompt 168. For example, the "node hint information", which is asset node information related to an asset node indicating a pump, may be information in the sentence "Symptoms that may occur in the pump are pump stop, decrease in discharge volume, and abnormal vibration." (This information may also be edge information related to an edge between the asset node and the symptom node.) Also, for example, the "node hint information", which is asset node information related to an asset node indicating a pump or an impeller (which are in an inclusion relationship (hierarchical relationship) with each other), may be information in the sentence "If the impeller is damaged, the discharge volume of the pump will decrease." (This information may also be edge information related to an edge between the asset node indicating the pump and the asset node indicating the impeller. Or, this information may also be edge information between the symptom node indicating a decrease in discharge volume and the failure mode node indicating impeller damage.) Furthermore, for example, the "node hint information", which is asset node information related to any one of two asset nodes that are in a connection relationship with each other (connected by an edge indicating a connection relationship), may be information indicating the relationship between the two assets. (This information may also be edge information related to the edge between the two asset nodes.) In addition, for example, regarding an asset schema node indicating a pump and an asset schema node indicating a rotating body in a schema (which are in an inheritance relationship with each other (connected by an edge indicating an inheritance relationship)), the "node hint information", which is asset node information related to the asset schema node indicating the rotating body, may be information in the sentence "One of the failure modes of the rotating body is abnormal vibration due to shaft misalignment." (This information may also be edge information related to an edge between the asset schema node indicating the rotating body and the failure mode schema node indicating abnormal vibration due to shaft misalignment.) Node hint information (including attribute information) may be set not only for asset nodes but also for symptom nodes and failure modes.

[0046] Hint set information is information formed by aggregating node information related to each of a plurality of nodes (or edge information related to each of a plurality of edges). For example, hint set information may be held as one piece of node information associated with any one of the plurality of nodes or a node in a relatively higher hierarchy as viewed from the plurality of nodes. As will be described later, a hint set information registration instruction unit 113 (see FIGS. 6, 7, and 8) or a conformity information registration instruction unit 130 (see FIGS. 16 and 17) in the system 101 may instruct to register hint set information in the knowledge graph 181. Therefore, at the beginning when the operation of the knowledge graph 181 is started, hint set information may not be registered in the knowledge graph 181. Search range information is information that determines which nodes and edges within a certain range including a node (in this case, an asset node) associated with the search range information are to be targeted for collecting node information (e.g., node hint information) and edge information (e.g., edge hint information), when the node associated with the search range information (in this case, the node of the asset node) is the node specified by a knowledge graph information request 166 (see FIGS. 6, 7, and 8) described later (in this case, the node of the specific asset 182). For example, if the search range information regarding a certain node is defined as "2 hops", in response to the knowledge graph information request 166 described later that specifies the certain node, the nodes and edges within the range of "2 hops" (the range reachable through two edges) starting from the certain node may be defined as the search range. The knowledge graph system 180 may collect node information and edge information related to the nodes and edges included in the search range and present the collected information to the knowledge graph information reception unit 107 of the system 101 as knowledge graph information 167.

[0047] Past document information is document information 162 indicating the content of a document 161 that triggered the update of the knowledge graph 181 in the past. Past document information is used during the optimization of hint set information and the like (see FIGS. 16 and 17) described later. Note that if the optimization process of hint set information and the like is not performed, the knowledge graph system 180 may not hold past document information. The past difference information indicates the content of the update result of the knowledge graph 181 (the difference information (difference sub-graph information) of the knowledge graph 181 before and after the update) when the knowledge graph 181 was updated in the past. The difference information (difference sub-graph information) of the knowledge graph 181 before and after the update may include one or both of (1) the update information of the node information and edge information related to the nodes and edges that already existed in the knowledge graph 181 before the update, and (2) the information related to the nodes and edges that did not exist in the knowledge graph 181 before the update and were added after the update. The past document information and the past difference information form a pair. That is, the past difference information indicates the result of updating the knowledge graph 181 based on the document information indicated by the past document information. If the knowledge graph 181 was updated multiple times in the past, the knowledge graph system 180 may record multiple pairs of the past document information and the past difference information. Note that if no optimization process such as hint set information is performed, the knowledge graph system 180 does not necessarily have to hold the past difference information.

[0048] 4. Processes Performed by Embodiments of the Present Disclosure Hereinafter, the processes executed by the embodiments of the present disclosure will be described. Note that it is not essential to implement all of the functional configurations described below and perform all of the processes. Also, the implementation of functional configurations and the execution of processes other than those described below are not precluded. In the following, while mainly referring to FIGS. 6 to 9, processing such as prompt construction using knowledge graph information (hint information or hint set information) will be described. While mainly referring to FIGS. 10 to 15, some examples of the knowledge graph, document information, knowledge graph information, and answer information (update query) will be described. While mainly referring to FIGS. 16 to 17, processing for optimizing hint set information and the like held in the knowledge graph 181 will be described. While mainly referring to FIGS. 18 to 19, processing for displaying a supplementary document formed by supplementing or correcting document information 162 based on knowledge graph information 167 will be described. While mainly referring to FIGS. 20 to 21, processing for displaying an inquiry regarding the uniqueness of the update of the knowledge graph 181 based on document information 162 will be described. While mainly referring to FIGS. 22 to 23, processing for editing the knowledge graph information (hint information) held in the knowledge graph 181 will be described.

[0049] 4.1. Processing such as prompt construction using knowledge graph information (FIGS. 6 to 9) While already referring to FIG. 1, the basic configuration of the embodiment of the present disclosure involved from when the system 101 receives document information 162 showing the content of document 161 that triggers the update of the knowledge graph 181 until the knowledge graph 181 is updated has been described. In the following, once again, while mainly referring to FIGS. 1, 3, 6, 7, 9, and 25 as appropriate, the processing performed in the embodiment of the present disclosure from when the system 101 receives document information 162 showing the content of document 161 that triggers the update of the knowledge graph 181 until the knowledge graph 181 is updated will be described in the order of the processing steps included in the flowchart of FIG. 8. In particular, in the following, the processing until the system 101 receives the knowledge graph information 167, the processing for constructing the prompt 168 according to the content of the received knowledge graph information 167, and the update processing additionally performed on the knowledge graph 181 accompanying the update of the knowledge graph 181 intended by the document 161 will be described in more detail. Before entering the description of the flowchart of FIG. 8, an overview of the drawings related to the following description will be described. Since FIGS. 1 and 3 have already been described, the description will be omitted here.

[0050] FIG. 6 focuses on the functional configuration (or information handled) leading up to the construction of the prompt 168 when the content of the knowledge graph information 167 used for the construction of the prompt 168 among the functional configurations (or information handled) in the embodiment of the present disclosure is one or more hint information. In FIG. 6, the notation of the functional configuration (or information handled) related to the processing after the prompt 168 is constructed is omitted. (For the functional configuration (or information handled) related to the processing after the prompt 168 is constructed, refer to FIG. 1.) Hereinafter, the "hint information" may be any information as long as it is information held by the knowledge graph 181 and is not the "hint set information" itself described in FIG. 3. The "hint information" may be node information related to nodes included in the knowledge graph 181 (other than the "hint set information"), or edge information related to edges included in the knowledge graph 181. That is, among all the information shown in FIGS. 3, 4, or 5, information other than the "hint set information" can be the "hint information". Note that the "hint information" should not be interpreted limitedly only to the "node hint information" or "edge hint information" shown in FIG. 3. In the upper part of FIG. 6, what is shown is a simplified display of the information held by the knowledge graph 181. In the upper part of FIG. 6, it is shown by two-way arrows between the asset nodes (this two-way arrow is not the edge itself in FIG. 3) that the asset nodes are in an inclusion relationship (vertical relationship) or a connection relationship. And in the upper part of FIG. 6, it is shown that hint information is associated with each of the asset nodes. For example, hint information (1) 601 is associated with the asset node (1) 311, hint information (2) 602 is associated with the asset node (2) 312, and hint information (3) 603 is associated with the asset node (3) 313. Also, in the upper part of FIG. 6, it is shown that hint information is associated with a set of nodes (for example, a pair of nodes) consisting of a plurality of asset nodes. For example, hint information (1 + 2) 611 is associated with the set of nodes consisting of the asset node (1) 311 and the asset node (2) 312. Also, hint information (2 + 3) 612 is associated with the set of nodes consisting of the asset node (2) 312 and the asset node (3) 313. Hint information (1 + 2) 611 and hint information (2 + 3) 612 may be, for example, edge information. In FIG. 6, what has a solid-line frame and has the word "section" in its name is a functional section. Each of the functional sections may be realized by a program being executed, or may be realized more in a hardware manner. As shown in FIG. 6, in addition to the functional sections already shown in FIG. 1, the system 101 may include an asset identification request prompt construction section 103, an asset identification request prompt presentation section 104, a specific asset response information reception section 105, a knowledge graph information request section 106, and a hint set information registration instruction section 113. The roles played by these functional sections will be described later together with the processing steps included in the flowchart of FIG. 8. In FIG. 6, what has a dotted-line frame indicates some kind of information (data). These information will also be described later together with the processing steps included in the flowchart of FIG. 8.

[0051] FIG. 7 focuses on the functional configuration (and the information handled) leading up to the construction of prompt 168 when the content of the knowledge graph information 167 used for the construction of prompt 168 among the functional configuration (and the information handled) in the embodiment of the present disclosure is hint set information. In FIG. 7, the notation of the functional configuration (and the information handled) related to the processing after the construction of prompt 168 is omitted. (Refer to FIG. 1 for the functional configuration (and the information handled) related to the processing after the construction of prompt 168.) Also, in FIG. 7, the notation of the functional configuration (and the information handled) related to the processing prior to the process in which the knowledge graph information request unit 106 presents the knowledge graph information request 166 to the knowledge graph system 180 is omitted. (Refer to FIG. 6 for the functional configuration (and the information handled) related to the processing prior to the process in which the knowledge graph information request unit 106 presents the knowledge graph information request 166 to the knowledge graph system 180.) The embodiment of the present disclosure may have the functional configurations of both FIG. 6 and FIG. 7 (in addition to the basic functional configuration shown in FIG. 1). (Note that there may also be a modified example having only one of the functional configurations of FIG. 6 that handles hint information or FIG. 7 that handles hint set information (in addition to the basic functional configuration shown in FIG. 1).) The upper part of FIG. 7 is the same as the upper part of FIG. 6 in that it shows the aspect of the knowledge graph 181. However, the upper part of FIG. 7 shows the hint set information 701 related to the asset node (1) 311. The hint set information 701 is a set of hint information composed of hint information related to a plurality of nodes. That is, the hint set information 701 may include information related to nodes other than the asset node (1) 311. In FIG. 7, those having a solid frame and having "section" in the name are functional sections. Each of the functional sections may be realized by a program being executed or may be realized more in a hardware manner. The roles played by these functional sections will be described later together with the processing steps included in the flowchart of FIG. 8. In FIG. 7, those with a dotted-line frame indicate some kind of information (data). These pieces of information will also be described later together with the processing steps included in the flowchart of FIG. 8.

[0052] FIG. 9 shows an example 900 of the construction of a prompt performed by the prompt construction unit 108. As shown in FIG. 9, the prompt construction unit 108 may insert, for example, into the location denoted as {hint} in the prompt template 901, a hint set information constructed from one or more pieces of hint information (see FIG. 6) received as the knowledge graph information 167 from the knowledge graph system 180, or the hint set information 701 (see FIG. 7) received as the knowledge graph information 167 from the knowledge graph system 180. Also, the prompt construction unit 108 may insert, for example, the document information 162 into the location denoted as {text} in the prompt template 901. Through these insertion processes, the prompt construction unit 108 may construct the prompt 168. Note that the construction method of the prompt 168 in the prompt construction unit 108 is not limited to that shown in FIG. 9. Note that FIG. 9 shows an example of the response information 170 which is the response of the language model 191 to which the prompt 168 is input (for example, the class diagram 970 in the PlantUML format). In this case, the "format X" in FIG. 9 may be, for example, the "PlantUML format". FIG. 9 will also be described later together with the processing steps included in the flowchart of FIG. 8.

[0053] FIG. 25 shows the asset identification request prompt template 2500 (although it is in a modified example to be described later). By inserting the asset name list information 2415 (in the modified example to be described later) into the location denoted as {hint} in the asset identification request prompt template 2500 and inserting the document information 162 into the location denoted as {text} in the asset identification request prompt template 2500, the asset identification request prompt 163 is constructed. Here, in the asset identification request prompt template for the embodiment shown in FIG. 6, the portion displayed as {hint} in FIG. 25 (and the description for two lines immediately before the portion displayed as {hint}) may not be present. Note that "Format Y" in FIG. 25 may be, for example, "CSV format".

[0054] Hereinafter, the processing steps included in the flowchart of FIG. 8 will be described in order. In step 801 (document information reception step) of FIG. 8, the document information reception unit 102 receives the document information 162. The processing performed by the document information reception unit 102, the document information 162, the entity presenting the document information 162, etc. have already been described with reference to FIG. 1.

[0055] In step 802 of FIG. 8, the asset identification request prompt construction unit 103 constructs an asset identification request prompt 163 based on the document information 162. The asset identification request prompt 163 is for requesting the language model 191 to identify the names of elements (such as assets (assets, valuable items)) that constitute the objects of knowledge (such as facilities and infrastructure) represented by the knowledge graph 181 from the content of the document 161 indicated by the document information 162. For example, from the document information 162 showing the content of the document 161 (a preservation document) "In the inspection of the water supply mechanism P001 on October 10, 2022, the impeller P003B of the pump P002B was replaced." illustrated in FIG. 9, the asset identification request prompt 163 is for requesting the language model 191 to identify the names of the elements (assets) "water supply mechanism P001", "pump P002B", and "impeller P003B". The asset identification request prompt construction unit 103 may use the asset identification request prompt template 2500 shown in FIG. 25, which has been mentioned above, or an asset identification request prompt template in a similar format (for example, a template in a format where the part related to the asset name list information 2415 is omitted from the format of FIG. 25). The asset identification request prompt construction unit 103 may construct the asset identification request prompt 163 by inserting the document information 162 into the corresponding part of the asset identification request prompt template (for example, the {text} part in FIG. 25). (In addition to the document information 162, a modified example of constructing the asset identification request prompt 163 based on the asset name list information 2415 will be described later.) In step 803 of FIG. 8, the asset identification request prompt presentation unit 104 presents the asset identification request prompt 163 to the language model system 190. The language model system 190 inputs the presented asset identification request prompt 163 into the language model 191. The language model 191 identifies a name included in the document information 162 inserted in the asset identification request prompt 163, which is the name of an element (asset) that constitutes an object of knowledge (e.g., equipment or infrastructure) represented by the knowledge graph 181. The language model 191 outputs a list of the names of the identified elements (assets) as the specific asset response information 164. For example, based on the document information 162 showing the content of the document 161 (preservation document) "In the inspection of the water supply mechanism P001 on October 10, 2022, the impeller P003B of the pump P002B was replaced." illustrated in FIG. 9, the language model 191 outputs the specific asset response information 164 including the list of the names of the elements (assets) "water supply mechanism P001", "pump P002B", and "impeller P003B". When the format of the list of the names of the elements (assets) is specified in advance, such as the asset identification request prompt template shown in FIG. 25, the language model 191 generates (outputs) a list of elements (assets) according to the specified format. In FIG. 25, the format is "Format Y", but it may be, for example, the CSV format. The language model system 190 presents the specific asset response information 164 to the system 101. In step 804 of FIG. 8, the specific asset response information receiving unit 105 receives the specific asset response information 164 from the language model system 190. Either the specific asset response information receiving unit 105 or the knowledge graph information requesting unit 106 may handle each piece of information indicating the name of the element (asset) included in the specific asset response information 164 as "specific asset information 165". In the above example, each of the information indicating "water supply mechanism P001", the information indicating "pump P002B", and the information indicating "impeller P003B" may be handled as the specific asset information 165. Also, "water supply mechanism P001", "pump P002B", and "impeller P003B" may be called "specific assets 182" in the sense that they are elements (assets) specified from the document information 162. As described above, since the language model 191 is utilized to obtain a list of elements (assets) from the document information 162, the generation of the specific asset information 165 can be automated.

[0056] In step 807 of FIG. 8, the knowledge graph information request unit 106 presents a knowledge graph information request 166 to the knowledge graph system 180. The knowledge graph information request 166 includes the specific asset information 165. The knowledge graph information request 166 requests to obtain, as the knowledge graph information 167, knowledge regarding the specific asset 182 indicated by the specific asset information 165 among the knowledge represented by the knowledge graph 181, or knowledge regarding an associated asset 183 that is an asset associated with the specific asset 182 (if it exists). The knowledge graph system 180 extracts knowledge regarding the specific asset 182 indicated by the presented knowledge graph information request 166 and knowledge regarding the associated asset 183 (if it exists) from the knowledge graph 181. The knowledge graph system 180 presents the information of the extracted knowledge to the system 101 as the knowledge graph information 167. The knowledge regarding the asset here corresponds to any information shown in FIG. 3, FIG. 4, or FIG. 5, but in FIG. 6 and FIG. 7, it is simply classified into "hint information (any information other than hint set information)" or "hint set information". In the above example, each of "water supply mechanism P001", "pump P002B", and "impeller P003B" corresponds to the specific asset 182. If it exists and is necessary, an associated asset 183 such as "motor P004B" may also be targeted for knowledge extraction. Here, the method by which the knowledge graph system 180 determines the associated asset 183 may be arbitrary. For example, the knowledge graph system 180 may determine the search range of the knowledge graph 181 when determining the associated asset 183 based on the "search range information (search hop count information, refer to FIG. 3)" regarding any of the specific assets 182. In step 808 (knowledge graph information reception step) of FIG. 8, the knowledge graph information reception unit 107 receives knowledge graph information 167 from the knowledge graph system 180. The received knowledge graph information 167 is roughly classified into (1) a case that includes hint information (group) regarding the specific asset 182 or the related asset 183 but does not include hint set information 701, and (2) a case that includes hint set information 701 regarding either the specific asset 182 or the related asset 183.

[0057] In step 810 of FIG. 8, the knowledge graph information reception unit 107 (or the prompt construction unit 108) determines whether the received knowledge graph information 167 includes hint set information 701. If the determination result in step 810 is affirmative (if the received knowledge graph information 167 includes hint set information 701), the control may transition to step 813. That is, the processes of steps 811 and 812 may be omitted. If the determination result in step 810 is negative (if the received knowledge graph information 167 does not include hint set information 701), the control transitions to step 811. Note that FIG. 6 mainly shows a case where the received knowledge graph information 167 does not include hint set information 701. On the other hand, FIG. 7 mainly shows a case where the received knowledge graph information 167 includes hint set information 701.

[0058] In step 811 of FIG. 8, the knowledge graph information reception unit 107 (or the prompt construction unit 108) constructs hint set information using the hint information (group) included in the received knowledge graph information 167. For example, the hint set information may be constructed in a manner of listing each of the hint information included in the received knowledge graph information 167.

[0059] In step 812 of FIG. 8, the hint set information registration instruction unit 113 may present a hint set information registration instruction 173 (see FIG. 7) for instructing to register the hint set information constructed in step 811 to the knowledge graph system 180. The knowledge graph system 180 may register (record) the hint set information constructed in step 811 so as to be associated with the nodes (for example, asset nodes in a relatively higher hierarchy) included in the knowledge graph 181 according to the presented hint set information registration instruction 173. In this way, the hint set information 701 can be automatically registered (recorded) in the knowledge graph 181. Therefore, it can be expected that the opportunity to construct an efficient prompt 168 using the hint set information 701 will increase. In addition, the hint set information registration instruction unit 113 may not only operate for automatic registration (recording) of the hint set information, but also operate when there is a request for manual registration (recording) of the hint set information. In this way, hint set information that is the result of human trial and error can also be registered (recorded) in the knowledge graph 181. Also, the processing step shown as step 812 in FIG. 8 may be executed at a timing later than the timing shown in FIG. 8. For example, the processing step shown as step 812 may be executed before or after the timing at which steps 818 and 819, which are the timings near the end in FIG. 8, are performed.

[0060] In step 813 (prompt construction step) of FIG. 8, the prompt construction unit 108 constructs a prompt 168 based on the document information 162 and the knowledge graph information 167. The processing performed by the prompt construction unit 108 and the form of the prompt 168 have already been described using FIG. 1 or have already been described using FIG. 9. Some supplementary explanations will be given below. As shown in FIG. 9, at the beginning of the prompt template 901 used by the prompt construction unit 108, a common request sentence is shown when requesting the language model 191 to generate (output) the answer information 170, which is the information used when constructing the update query 171 for updating the knowledge graph 181. As shown in FIG. 9, for example, a request sentence such as "You are an expert in format X. Using the following text, construct a class in format X and output the class diagram in the format of X. Please follow the following rules for creating the class diagram." may be shown at the beginning of the prompt template 901. In the prompt template 901, after the beginning part, the position where the knowledge graph information 167 is inserted as {hint}, the position where "#Article:" is described, the position where the document information 162 is inserted as {text}, and the position where "#Class diagram in format X" is described (the answer information 170 from the language model 191 is inserted directly below this position.) may be shown in order.

[0061] The knowledge graph information 167 inserted at the {hint} position of the prompt template 901 may be the hint set information constructed in step 811 based on one or more hint information received as the knowledge graph information 167 from the knowledge graph system 180, or the hint set information 701 received as the knowledge graph information 167 from the knowledge graph system 180. The hint set information shown in the example of FIG. 9 consists of "The water supply mechanism P001, the pump P002B, and the impeller P003B are assets.", "Assets have the following relationships.", "Water supply mechanism P001 → Pump P002B", and "Pump P002B → Impeller P003B". If this hint set information is constructed in step 811, the individual hint information constituting this hint set information may be associated with and held for each of the nodes and each of the edges in the knowledge graph 181. For example, in the hint set information shown in FIG. 9, the portion "The water supply mechanism P001, the pump P002B, and the impeller P003B are assets." may be hint information associated with the asset node indicating the "water supply mechanism P001", the asset node indicating the "pump P002B", or the asset node indicating the "impeller P003B". The portion "The asset has the following relationships." "Water supply mechanism P001 → Pump P002B" may be hint information associated with the edge connecting the asset node indicating the "water supply mechanism P001" and the asset node indicating the "pump P002B". (Alternatively, this portion may be hint information indicating the asset configuration associated with the asset node indicating the "water supply mechanism P001".) The portion "The asset has the following relationships." "Pump P002B → Impeller P003B" may be hint information associated with the edge connecting the asset node indicating the "pump P002B" and the asset node indicating the "impeller P003B". (Alternatively, this portion may be hint information indicating the asset configuration associated with the asset node indicating the "pump P002B".) When the hint set information shown in FIG. 9 is the hint set information 701 received as the knowledge graph information 167 from the knowledge graph system 180, the hint set information 701 may be held by the knowledge graph 181 as being related to, for example, the asset node indicating the "water supply mechanism P001" (in a relatively higher-level hierarchy). The prompt template 901 may be recorded in a recording medium or a recording device within the system 101 (for example, the non-volatile recording medium (recording device) 203 in FIG. 2). Alternatively, the prompt template 901 may be recorded in a recording medium or a recording device within the knowledge graph system 180. In that case, upon request from the system 101, the knowledge graph system 180 may provide the prompt template 901 to the system 101.

[0062] For each set of nodes and edges representing the knowledge of facilities and infrastructure in the knowledge graph 181, there may be a prompt template for that set. In that case, for example, the prompt template may be registered (recorded) in association with the asset nodes at a relatively higher level in the set of nodes and edges. Alternatively, there may be separate prompt templates for each type of document 161 that triggers a change to the knowledge graph 181. For example, there may be a maintenance prompt template corresponding to a maintenance document created in the maintenance activities for facilities and infrastructure, a design prompt template corresponding to a design document created in the design activities for facilities and infrastructure, and an operation prompt template corresponding to an operation document created in the operation activities for facilities and infrastructure, each existing separately.

[0063] In step 814 (prompt presentation step) of FIG. 8, the prompt presentation unit 109 presents the prompt 168 to the language model system 190. The language model system 190 inputs the presented prompt 168 into the language model 191. The language model 191 generates (outputs) response information 170, which is a response to the prompt 168. The language model system 190 presents the response information 170 to the system 101. The prompt 168 includes not only the document information 162 but also the knowledge graph information 167 (for example, hint set information). Therefore, the language model 191 can generate (output) the response information 170 while fully considering the domain knowledge regarding the object of the knowledge represented by the knowledge graph 181 (for example, facilities and infrastructure) updated based on the document information 162. Also, even when there are omissions or the like in the content of the document 161 indicated by the document information 162, the response information 170 can be generated (output) based on the content of the document 161 supplemented or corrected. Below the lower part of FIG. 9, an example of the response information 170 is shown. In the example of FIG. 9, since the prompt template 901 and the prompt 168 are for obtaining the response information 170 in a manner similar to a class diagram in format X (for example, the PlantUML format), the example of the response information 170 shown in the lower part of FIG. 9 is a class diagram 970 (or something similar to a class diagram). In the example of the response information 170 in FIG. 9, "ID: P001" is shown for the class of "water supply mechanism", "ID: P002B" is shown for the class of "pump", "ID: P003B" and "Replacement date: 2022 / 10 / 10" are shown for the class of "impeller", and "water supply mechanism → pump: has_a" and "pump → impeller: has_a" are shown for the edges between asset nodes. Note that the format of the response information 170 is not limited to the above. For example, it may be the format of the response information 170 as shown in FIGS. 11, 13, and 15. In step 815 (response information reception step) of FIG. 8, the response information reception unit 110 receives the response information 170 from the language model system 190.

[0064] In step 816 (update query construction step) of FIG. 8, the update query construction unit 111 constructs an update query 171 based on the response information 170. The process of the update query construction unit 111 and the form of the update query 171 have already been described with reference to FIG. 1.

[0065] In step 817 of FIG. 8, a verification unit (which the system 101 may have) not shown in FIGS. 6 and 7 may verify the validity of the update query 171. The verification unit may, for example, automatically check whether the update query 171 violates the constraints and regulations in the knowledge graph 181. Alternatively, the verification unit may, for example, control the display or output device 207 to display or output the update query 171. In this case, the verification unit may receive, via the input device 206, an input of a judgment result on the suitability of the update query 171 from a person who has viewed the display result or output result. When the verification unit obtains a negative verification result regarding the validity of the update query, the control may return to any of the processing steps relatively upstream on the flowchart of FIG. 8. When the verification unit obtains an affirmative verification result regarding the validity of the update query, the control transitions to step 818. Incidentally, when it is desired not to verify the validity of the update query 171, or when the validity of the update query 171 is expected to be generally affirmative, it is also allowed for the system 101 not to have a verification unit or for the system 101 not to execute step 817.

[0066] In step 818 (update query presentation step) of FIG. 8, the update query presentation unit 112 presents the update query 171 to the knowledge graph system 180. The processing performed by the update query presentation unit 112 and the processing performed by the knowledge graph system 180 presented with the update query 171 have already been described with reference to FIG. 1.

[0067] In step 819 of FIG. 8, a document and a differential registration instruction unit (which may be included in system 101) not shown in FIGS. 6 and 7 may present a history registration instruction to the knowledge graph system 180. The history registration instruction is for instructing the knowledge graph system 180 to register (record) the document information 162 received in step 801 and the content of the update result of the knowledge graph 181 performed by the knowledge graph system 180 in step 818 (the differential information (differential sub-graph information) of the knowledge graph 181 before and after the update). The document information 162 registered (recorded) by the history registration instruction becomes the "past document information" shown in FIG. 3. Also, the differential information (differential sub-graph information) registered (recorded) by the history registration instruction becomes the "past differential information" shown in FIG. 3. The "past document information" and the "past differential information" have already been described with reference to FIG. 3. When the "past document information" and the "past differential information" are registered (recorded), they may be associated with any of the nodes and edges involved in the update of the knowledge graph 181. For example, they may be associated with an asset node at a relatively higher level (for example, among the asset nodes indicating "water supply mechanism", "pump", and "impeller", the asset node indicating "water supply mechanism") and the "past document information" and the "past differential information" may be registered (recorded). In addition, if the optimization process of hint set information and the like shown in the flowchart of FIG. 17 described later is not performed, the necessity of registering (recording) the "past document information" and the "past differential information" becomes low, so the process of this step 819 may not be executed.

[0068] As described above, in the embodiments of the present disclosure, regarding the specific asset 182 described in the name in the document 161 indicated by the document information 162 and the related asset 183 related to the specific asset 182, the prompt 168 can be flexibly constructed according to the aspect of the knowledge graph information 167 received by the system 101 after being extracted from the knowledge graph 181. Specifically, when the knowledge graph information 167 received by the system 101 does not include hint set information 701 comprehensively showing knowledge about a plurality of nodes and edges in the knowledge graph 181, hint set information is newly constructed from the individual hint information included in the received knowledge graph information 167. On the other hand, when the knowledge graph information 167 received by the system 101 includes the hint set information 701, the processing step of constructing the hint set information may be omitted. Then, based on the document information 162 and the knowledge graph information 167 (for example, hint set information), the prompt 168 is constructed. Also, even when the knowledge graph information 167 received by the system 101 does not include the hint set information 701 comprehensively showing knowledge about a plurality of nodes and edges in the knowledge graph 181, the hint set information newly constructed by the system 101 may be registered (recorded) in the knowledge graph 181. Therefore, when the newly registered (recorded) hint set information is utilized later, the construction of the prompt 168 becomes efficient.

[0069] 2. Examples of Knowledge Graph, Document Information, Knowledge Graph Information, and Answer Information (Update Query) in 4 Hereinafter, a first example, a second example, and a third example will be described. For each of the examples, a combination of a knowledge graph, document information showing the content of the document that triggers the update of the knowledge graph, knowledge graph information (hint set information and hint information) that can be used for constructing a prompt, and answer information (update query) of a language model for the prompt will be described. The first case is about a water supply mechanism with the identifier "P001", which includes pumps with the identifiers "P002A" and "P002B", and the pump with the identifier "P002A" includes an impeller with the identifier "P003A", and the pump with the identifier "P002B" includes an impeller with the identifier "P003B". On the other hand, both the second case and the third case are about facilities where the water supply mechanism with the identifier "P001" includes the pump with the identifier "P002B", and the pump with the identifier "P002B" includes the impeller with the identifier "P003B". As shown in FIGS. 10, 12, and 14, in any case, the knowledge graph may include a schema and an instance group. The schema and the instance group have already been described in the section "3. Examples of the Structure of the Knowledge Graph". In FIGS. 10, 12, and 14, in order to distinguish the nodes included in the schema from the nodes included in the instance group, the nodes included in the schema are given the name "schema node". In any of FIGS. 10, 12, and 14, within the schema or the instance group, asset nodes in an inclusion relationship (hierarchical relationship) are connected by an edge of the type "has_a (ownership)". On the other hand, between the asset schema nodes in the schema and the asset nodes in the instance group, those corresponding to the same type of asset may be connected by an edge of the type "is_a (instantiation)". For example, the water supply mechanism schema node in the schema and the water supply mechanism node in the instance group may be connected by an edge of the type "is_a (instantiation)". The same applies to pumps and impellers.

[0070] 4 of 2 1. The First Case (Figs. 10 - 11) Figure 10 shows the knowledge graph 1000 in the first case. In Figure 10, among the nodes included in the knowledge graph, only the asset nodes are shown. Although not shown in Figure 10, there may be symptom nodes or failure mode nodes. Among the instance group 1002 shown in Figure 10, the sub-graph 1004 updated based on the document information shown in Figure 11 is shown by a double-line enclosure in Figure 10. That is, the update of the knowledge graph in the first case is of the type that updates the information (node information or edge information) related to the set of nodes (and the set of edges) already existing in the knowledge graph.

[0071] Figure 11 shows the combination 1100 of document information, knowledge graph information (hint set information and hint information), and answer information (update query) in the first case. In Figure 11, the case of updating the knowledge graph based on the document information 162A (preservation document A) and the case of updating the knowledge graph based on the document information 162B (preservation document B) are shown together.

[0072] 4 of 2 of 1 of 1. Case of document information 162A (preservation document A) First, the case of updating the knowledge graph based on the document information 162A (preservation document A) will be described. As shown in the left part of Figure 11, the document information 162A (preservation document A) has the content (as a maintenance report) of "Replaced the impeller of the second pump during the inspection of the water supply mechanism P001 on October 1, 2022." The document information 162A (preservation document A) includes the asset names of "water supply mechanism P001", "second pump", and "impeller". In the document information 162A (preserved document A), it is not clear which pump among those included in the "water supply mechanism P001" the "second pump" is (i.e., what the identifier assigned to the "second pump" is). Also, in the document information 162A (preserved document A), it is not clear which impeller among those included in the "second pump" the "impeller" is (i.e., what the identifier assigned to the "impeller" is). Furthermore, in the document information 162A (preserved document A), it is not clear what information in the knowledge graph should be updated along with "replacing the impeller". Due to the above circumstances, even if a prompt is constructed based on the document information 162A (preserved document A) without using the knowledge graph information and the prompt is presented to the language model, a language model that has not been trained with information from the knowledge domain related to the knowledge graph is likely to generate (output) inappropriate response information for the prompt (causing hallucinations (errors)).

[0073] Therefore, the system 101 acquires knowledge graph information regarding the "water supply mechanism P001", "second pump", and "impeller", which are assets (specific assets) whose names appear in the document information 162A (preserved document A), from the knowledge graph system 180. The acquired knowledge graph information may be the one shown as the "hint set information 1101" in FIG. 11, or may be the one shown as the "hint information 1111", "hint information 1112", "hint information 1113", and "hint information 1114" in FIG. 11. When individual hint information such as "hint information 1111", "hint information 1112", "hint information 1113", and "hint information 1114" is acquired, the hint set information 1101 may be constructed again from these hint information. Hint information 1111 is "In P001, the second pump is P002B." Hint information 1111 may be node information related to the water supply mechanism P001 node 1021, or may be edge information related to the edge (an edge of the "has_a (ownership)" type) connecting the water supply mechanism P001 node 1021 and the pump P002B node 1022. Hint information 1112 is "The impeller of P002B is P003B." Hint information 1112 may be node information related to the pump P002B node 1022, or may be edge information related to the edge (an edge of the "has_a (ownership)" type) connecting the pump P002B node 1022 and the impeller P003B node 1023. Hint information 1113 is "Replacement means updating the installation date to the replacement date." Hint information 1113 may be node information related to the water supply mechanism P001 node 1021, or may be node information related to the pump P002B node 1022, or may be node information related to the impeller P003B node 1023. Hint information 1114 is "When an asset is replaced, its subordinate assets are also replaced." Hint information 1114 may be node information related to the water supply mechanism P001 node 1021, or may be node information related to the pump P002B node 1022. Incidentally, when the combined information of hint information 1111, hint information 1112, hint information 1113, and hint information 1114 is registered (recorded) as hint set information 1101, the hint set information 1101 may be, for example, node information related to the water supply mechanism P001 node 1021, which is a relatively upper-level asset node (it may also be node information related to other nodes).

[0074] The knowledge graph information (hint set information and hint information) as described above provides knowledge for resolving unclear points in the document information 162A (preserved document A). Specifically, the hint information 1111 specifies that the identifier of the "second pump" included in the water supply mechanism with the identifier "P001" is "P002B". Also, the hint information 1112 specifies that the identifier of the impeller included in the pump (second pump) with the identifier "P002B" is "P003B". Furthermore, the hint information 1113 specifies that when an asset (e.g., an impeller) is replaced, the information of "installation date", which is the node information regarding the asset, should be updated to match the "replacement date" of the asset. In this way, it can be expected that the knowledge graph information (hint set information and hint information) constructs a prompt along the knowledge target (e.g., facilities and infrastructure) represented by the knowledge graph, and constructs a prompt after supplementing or correcting matters omitted etc. in the document information. When a prompt with appropriate content is input into the language model, it can be expected that the language model generates (outputs) response information with appropriate content. In the example on the left side of FIG. 11, "Update the installation date of P003B to 2022 / 10 / 01." is obtained as the response information 170A. Based on the response information 170A, an update query A is constructed. The update query A is applied to the knowledge graph.

[0075] Case of Document Information 162B (Preserved Document B) of 4 of 2 of 1 of 2. Next, the case of updating the knowledge graph based on the document information 162B (preserved document B) is described. As shown in the right part of FIG. 11, the document information 162B (preserved document B) has the content (as a maintenance report) of "During the inspection of the water supply mechanism P001 on 2022 / 10 / 01, the second pump was replaced." The document information 162B (preserved document B) includes the asset names of "water supply mechanism P001" and "second pump". The document information 162B (preserved document B) does not include the asset name of "impeller P003B". In the document information 162B (preserved document B), it is not clear which pump among those included in the "water supply mechanism P001" the "second pump" is (what the identifier assigned to the "second pump" is). Also, since the pump includes an impeller, when a certain pump is replaced, the impeller included in that certain pump will also be implicitly replaced. However, there is no mention of the implicit replacement of the impeller in the document information 162B (preserved document B). Furthermore, in the document information 162B (preserved document B), it is not clear what information in the knowledge graph should be updated along with "replacing the pump". Due to the above circumstances, even if a prompt is constructed based on the document information 162B (preserved document B) without using the knowledge graph information and the prompt is presented to the language model, a language model that has not been trained with information from the knowledge domain related to the knowledge graph is likely to generate (output) inappropriate response information for the prompt (causing hallucinations (errors)).

[0076] Therefore, in the same manner as in the case of the document information 162A (preserved document A), the system 101 obtains from the knowledge graph system 180 the knowledge graph information regarding the "water supply mechanism P001" and the "second pump", which are assets (specific assets) whose names appear in the document information 162B (preserved document B). Also, when the "impeller P003B" is defined as an associated asset related to the specific asset, the system 101 may also obtain from the knowledge graph system 180 the knowledge graph information regarding the "impeller P003B". The obtained knowledge graph information (hint set information and hint information) may be the same as that obtained in the case of the document information 162A (preserved document A).

[0077] The above knowledge graph information (hint set information and hint information) provides knowledge to resolve unclear points in the document information 162B (preserved document B). Specifically, the hint information 1111 specifies that the identifier of the "second pump" included in the water supply mechanism with the identifier "P001" is "P002B". Also, the hint information 1112 specifies that the identifier of the impeller included in the pump (second pump) with the identifier "P002B" is "P003B". Furthermore, the hint information 1113 specifies that when an asset (e.g., a pump or an impeller) is replaced, the information of "installation date", which is the node information regarding the asset, should be updated to match the "replacement date" of the asset. In addition, the hint information 1114 specifies that when an asset (e.g., pump P002B) in a relatively higher layer is replaced, implicitly, the asset (e.g., impeller P003B) in a relatively lower layer in an inclusion relationship is also replaced. In this way, it can be expected that the knowledge graph information (hint set information and hint information) constructs a prompt along with the object of the knowledge represented by the knowledge graph (e.g., a facility or an infrastructure), and constructs a prompt after supplementing or correcting matters omitted etc. in the document information. When a prompt with appropriate content is input into the language model, it can be expected that the language model generates (outputs) response information with appropriate content. In the example on the right side of Fig. 11, as the response information 170B, "Update the installation date of P002B to 2022 / 10 / 01." and "Update the installation date of P003B to 2022 / 10 / 01." are obtained. Based on the response information 170B, an update query B is constructed. The update query B is applied to the knowledge graph.

[0078] 4 of 2 of 2. The second case (Figs. 12 - 13) FIG. 12 shows the knowledge graph 1200 in the second case. Among the instance group 1202 shown in FIG. 12, the sub-graph 1204 added based on the document information shown in FIG. 13 is indicated by a double-line enclosure in FIG. 12. That is, the update of the knowledge graph in the second case is of the type that adds a sub-graph, which is a set of nodes (and a set of edges) that did not exist in the knowledge graph before the update.

[0079] FIG. 13 shows a combination 1300 of document information, knowledge graph information (hint set information and hint information), and answer information (update query) in the second case. FIG. 13 shows a case where the knowledge graph is updated based on the document information 162C (preserved document C). As shown in FIG. 13, the document information 162C (preserved document C) has the content of "The water supply function of P001 has stopped. The cause is impeller damage." (as a defect report). The document information 162C (preserved document C) includes the identifier of the asset "P001" and the asset name "impeller". The document information 162C (preserved document C) does not include the asset name "pump P002B". In the document information 162C (preserved document C), the type of the asset with the identifier "P001" is not clear. Also, in the document information 162C (preserved document C), it is not clear what inclusion relationship (hierarchical relationship) exists between the asset with the identifier "P001" and the asset "impeller", and it is not clear what the identifier assigned to the "impeller" is. Due to the above circumstances, even if a prompt is constructed based on the document information 162C (preserved document C) without using the knowledge graph information and the prompt is presented to the language model, a language model that has not been trained with information on the knowledge domain related to the knowledge graph is likely to generate (output) inappropriate answer information for the prompt (causing hallucination (error)).

[0080] Therefore, in the same manner as in the first case, the system 101 obtains, from the knowledge graph system 180, knowledge graph information regarding "P001" and "impeller", which are assets (specific assets) in which names and identifiers appear in the document information 162C (preserved document C). Also, when "pump P002B" is defined as a related asset related to the specific asset, the system 101 of the embodiment of the present disclosure may also obtain knowledge graph information regarding "pump P002B" from the knowledge graph system 180. The obtained knowledge graph information may be the one shown as "hint set information 1301" in FIG. 13, or may be the one shown as "hint information 1311", "hint information 1312", "hint information 1313", and "hint information 1314" in FIG. 13. When individual hint information such as "hint information 1311", "hint information 1312", "hint information 1313", and "hint information 1314" is obtained, hint set information 1301 may be reconstructed from these hint information. The hint information 1311 is "P001 is a water supply mechanism." The hint information 1311 may be node information related to the water supply mechanism P001 node 1221. The hint information 1312 is "P001 has P002B." The hint information 1312 may be node information related to the water supply mechanism P001 node 1221, or may be edge information related to the edge (an edge of the "has_a (ownership)" type) connecting the water supply mechanism P001 node 1221 and the pump P002B node 1222. The hint information 1313 is "P002B has P003B." The hint information 1313 may be node information related to the pump P002B node 1222, or may be edge information related to the edge (an edge of the "has_a (ownership)" type) connecting the pump P002B node 1222 and the impeller P003B node 1223. The hint information 1314 is "P003 is an impeller." The hint information 1314 may be node information related to the impeller P003B node 1223. Incidentally, when the information combining the hint information 1311, the hint information 1312, the hint information 1313, and the hint information 1314 is registered (recorded) as the hint set information 1301, the hint set information 1301 may be, for example, node information related to the water supply mechanism P001 node 1221, which is an asset node in a relatively higher layer (it may also be node information related to other nodes).

[0081] The knowledge graph information (hint set information and hint information) as described above provides knowledge for resolving unclear points in the document information 162C (preserved document C). Specifically, the hint information 1311 clarifies that the type of the asset with the identifier "P001" is "water supply mechanism." Also, the combined information of the hint information 1312, the hint information 1313, and the hint information 1314 clarifies that the asset with the identifier "P001" has the asset with the identifier "P002B," and the asset with the identifier "P002B" has the impeller with the identifier "P003B." In this way, it can be expected that the prompt is constructed along the target of the knowledge represented by the knowledge graph (hint set information and hint information) (for example, facilities and infrastructure), or the prompt is constructed after supplementing or correcting the matters omitted in the document information. When a prompt with appropriate content is input to the language model, it can be expected that the language model generates (outputs) response information with appropriate content. In the example of FIG. 13, as the response information 170C, "Add the following relationship: " and "P001 → water supply function stop → impeller damage → P003B" are obtained. Based on the response information 170C, an update query C is constructed. The update query C is applied to the knowledge graph. As a result of applying the update query C to the knowledge graph, as shown in FIG. 12, a water supply function stop node 1241 and an impeller damage mode node 1243 are added to the instance group 1202, and the water supply mechanism P001 node 1221 and the water supply function stop node 1241 are connected by an edge of the "symptom" type, the water supply function stop node 1241 and the impeller damage mode node 1243 are connected by an edge of the "reason" type, and the impeller damage mode node 1243 and the impeller P003B node 1223 are connected by an edge of the "at" type.

[0082] The third case (FIGS. 14 to 15) of 4 of 2 FIG. 14 shows a knowledge graph 1400 in the third case. Among the instance group 1402 shown in FIG. 14, the sub-graph 1404 added based on the document information shown in FIG. 15 is indicated by a double-line enclosure in FIG. 14. That is, the update of the knowledge graph in the third case is of the type that adds a sub-graph, which is a set of nodes (and a set of edges) that did not exist in the knowledge graph before the update. The instance group 1402 in FIG. 14 and the instance group 1202 in FIG. 12 have the same nodes and edges included in the instance group (except for the upper two digits of the four-digit reference numbers assigned to the nodes). Therefore, in FIG. 14, the inside of the instance group 1402 is abbreviated. In FIG. 14, schema 1401 has not only the water supply mechanism schema node 1411, the pump schema node 1412, and the impeller schema node 1413, which are schema nodes related to the asset, but also the water supply function stop schema node 1451, the pump stop schema node 1452, the pipe leakage schema node 1462, the impeller damage schema node 1453, and the motor damage schema node 1463, which are schema nodes related to symptoms or failure modes. Schema 1401 may comprehensively represent possible relationships regarding the asset, symptoms, and failure modes as much as possible. On the other hand, the instance group 1402 may represent the relationships regarding the asset, symptoms, and failure modes that are assumed to actually exist or have a high possibility of existing in the objects of knowledge (such as facilities and infrastructure) represented by the knowledge graph.

[0083] FIG. 15 shows a combination 1500 of document information, knowledge graph information (hint set information and hint information), and answer information (update query) in the third case. FIG. 15 shows a case where the knowledge graph is updated based on the document information 162D (preserved document D). As shown in FIG. 15, the document information 162D (preserved document D) has the content (as a maintenance report) of "There was a notice of the water supply function stop of P001 on 2022 / 10 / 01." and "The impeller of the second pump of P001 was replaced on 2022 / 10 / 02." The document information 162D (preserved document D) includes the identifier of the asset "P001" and the asset names "second pump" and "impeller". In the document information 162D (conservation document D), the type of the asset with the identifier "P001" is not clear. Also, in the document information 162D (conservation document D), it is not clear which pump among those included in the asset with the identifier "P001" the "second pump" is (what the identifier assigned to the "second pump" is). Furthermore, in the document information 162D (conservation document D), it is not clear which impeller among those included in the "second pump" the "impeller" is (what the identifier assigned to the "impeller" is). Moreover, in the document information 162D (conservation document D), the relationship between the occurrence of "water supply function stop" and "impeller replacement" is not clear. Due to the above circumstances, even if a prompt is constructed based on the document information 162D (conservation document D) without using knowledge graph information and the prompt is presented to the language model, a language model that has not been trained with information on the knowledge domain related to the knowledge graph is likely to generate (output) inappropriate response information for the prompt (causing hallucination (error)).

[0084] Therefore, in the same manner as in the first case, the system 101 obtains, from the knowledge graph system 180, knowledge graph information regarding "P001", "No. 2 pump", and "impeller", which are assets (specific assets) in which names and identifiers appear in the document information 162D (preserved document D). The system 101 may further obtain, from the knowledge graph system 180, knowledge graph information from each of the schema nodes in the schema 1401 regarding "water supply mechanism" as the asset type corresponding to "P001", "pump" as the asset type corresponding to "No. 2 pump", and "impeller" as the asset type. At this time, not only the schema nodes regarding assets (asset schema nodes) in the schema 1401 but also the schema nodes regarding symptoms and failure modes that are directly or indirectly connected by the asset schema nodes and edges may be targeted for obtaining the knowledge graph information. The obtained knowledge graph information may be the one shown as "hint set information 1501" in FIG. 15, or may be the one shown as "hint information 1511", "hint information 1512", "hint information 1513", "hint information 1514", and "hint information 1515" in FIG. 15. When individual hint information such as "hint information 1511", "hint information 1512", "hint information 1513", "hint information 1514", and "hint information 1515" is obtained, the hint set information 1501 may be reconstructed from these hint information. The hint information 1511 is "In P001, the second pump is P002B." The hint information 1511 may be node information related to the water supply mechanism P001 node 1421, or may be edge information related to the edge (an edge of the "has_a (ownership)" type) connecting the water supply mechanism P001 node 1421 and the pump P002B node 1422. The hint information 1512 is "The impeller of P002B is P003B." The hint information 1512 may be node information related to the pump P002B node 1422, or may be edge information related to the edge (an edge of the "has_a (ownership)" type) connecting the pump P002B node 1422 and the impeller P003B node 1423. The hint information 1513 is "The causes of the water supply function stop are pump stop and pipe leakage." The hint information 1513 may be node information related to the water supply function stop schema node 1451, or may be edge information related to the edge (an edge of the "has_a (ownership)" type) connecting the water supply function stop schema node 1451 and the pump stop schema node 1452, or may be edge information related to the edge (an edge of the "reason (reason)" type) connecting the water supply function stop schema node 1451 and the pipe leakage schema node 1462. The hint information 1514 is "The causes of the pump stop are impeller damage and motor damage." The hint information 1514 may be node information related to the pump stop schema node 1452, or may be edge information related to the edge (an edge of the "reason (reason)" type) connecting the pump stop schema node 1452 and the impeller damage schema node 1453, or may be edge information related to the edge (an edge of the "reason (reason)" type) connecting the pump stop schema node 1452 and the motor damage schema node 1463. The hint information 1515 is "The countermeasure for impeller damage is impeller replacement." The hint information 1515 may be node information related to the impeller damage schema node 1453.When the combined information of hint information 1511, hint information 1512, hint information 1513, hint information 1514, and hint information 1515 is registered (recorded) as hint set information 1501, the hint set information 1501 may be, for example, node information related to the water supply mechanism P001 node 1421, which is an asset node at a relatively higher level (it may also be node information related to other nodes).

[0085] The knowledge graph information (hint set information and hint information) as described above provides knowledge to resolve unclear points in the document information 162D (preserved document D). Specifically, the hint information 1511 specifies that the type of the asset with the identifier "P001" is "water supply mechanism", and also specifies that the identifier of the "second pump" included in the water supply mechanism with the identifier "P001" is "P002B". Also, the hint information 1512 specifies that the identifier of the impeller included in the pump (second pump) with the identifier "P002B" is "P003B". Furthermore, the combined information of hint information 1513, hint information 1514, and hint information 1515 specifies that pump stop is one of the assumed causes of the water supply function stop, impeller damage is one of the assumed causes of the pump stop, and impeller replacement is one of the measures (solutions) for the impeller damage. That is, the combined information of hint information 1513, hint information 1514, and hint information 1515 shows the relationship between the symptom of water supply function stop in the water supply mechanism and the measure (solution) of impeller replacement. In this way, it is expected that the knowledge graph information (hint set information and hint information) can construct a prompt along the object of the knowledge represented by the knowledge graph (for example, facilities or infrastructure), and can construct a prompt after supplementing or correcting the matters omitted in the document information. When a prompt with appropriate content is input to the language model, it is expected that the language model will generate (output) response information with appropriate content. In the example of FIG. 15, as response information 170D, "Add the following relationship:" and "P001 → water supply function stop → impeller damage → P003B" are obtained. Based on the response information 170D, an update query D is constructed. The update query D is applied to the knowledge graph. As a result of applying the update query D to the knowledge graph, as shown in FIG. 14 (since it is abbreviated in FIG. 14, refer to FIG. 12 as well), after adding a water supply function stop node 1441 and an impeller damage mode node 1443 to the instance group 1402, the water supply mechanism P001 node 1421 and the water supply function stop node 1441 are connected by an edge of the "symptom" type, the water supply function stop node 1441 and the impeller damage mode node 1443 are connected by an edge of the "reason" type, and the impeller damage mode node 1443 and the impeller P003B node 1423 are connected by an edge of the "at" type.

[0086] 3. Optimization processing of hint set information, etc. (FIGS. 16 to 17) of 4 As shown by the hint set information registration instruction unit 113 in FIG. 6 or FIG. 7, the hint set information registration instruction 173 in FIG. 7, step 811 in FIG. 8, or step 812 in FIG. 8, in the embodiment of the present disclosure, when new hint set information is constructed for the construction of the prompt 168, it is possible to register (record) the hint set information in the knowledge graph 181. Also, it is possible for those handling the system 101 or the knowledge graph system 180 to manually register (record) the hint set information constructed through trial and error in the knowledge graph 181. However, not all of the hint information included in the hint set information 701 registered (recorded) in the knowledge graph 181 by the above method contributes to the construction of the update query 171 with appropriate content. That is, there may be useless or low-benefit hint information among the hint information included in the hint set information 701. If the hint set information 701 contains useless or low-benefit hint information, it is considered that the data volume, bandwidth, communication time, required computing resources, computing time, etc. will become extremely large for the hint set information 701 and the information constructed using the hint set information 701 (for example, the prompt 168). Therefore, hereinafter, a process for optimizing the hint set information 701 and the like registered (recorded) in the knowledge graph 181, which can be executed by the embodiments of the present disclosure, will be described. By optimizing the hint set information 701 and the like, it is expected that the data volume, bandwidth, communication time, required computing resources, computing time, etc. will become reasonable for the hint set information 701 and the information constructed using the hint set information 701 (for example, the prompt 168).

[0087] FIG. 16 shows a functional configuration 1600 related to the process of optimizing hint set information and the like. Although the boundary line of the system 101 is not explicitly shown in FIG. 16, in FIG. 16, each of the functional units (shown by the solid rectangular frames), excluding the knowledge graph system 180 and the language model 191 (the illustration of the language model system 190 is omitted), may be possessed by the system 101. As shown in FIG. 16, in addition to the functional units already described, the system 101 may have a knowledge graph update history reception unit 121, a hint information group reception unit 122, a temporary hint set formation unit 123, a temporary difference information identification unit 124, a similarity calculation unit 125, a compatible hint set information identification unit 126, and a compatible information registration instruction unit 130. Further, as functional units included in the compatible hint set information identification unit 126, an evaluation value processing unit 128 (including a function 127) and another function control unit 129 may exist. The functions and processes performed by each of the functional units will be described together with the processing steps in the flowchart of FIG. 17. In FIG. 16, the dotted rectangular frame indicates some information (data). In addition to the information (data) already described, the system 101 may handle some or all of the past document information 1601, the past difference information 1602, the hint information group 1603, the temporary hint set information 1623, the temporary prompt 1608, the temporary answer information 1610, the temporary difference information 1624, the similarity 1625, the hint information number information 1641, the temporary search range information 1644, the evaluation value 1627, the compatible hint set information 1633, the compatible search range information 1634, and the compatible information registration instruction 1630. The significance of these information (data) and the handling of these information (data) will be described together with the processing steps in the flowchart of FIG. 17.

[0088] FIG. 17 shows a flowchart 1700 of a process for optimizing hint set information and the like. Hereinafter, the processing steps shown in FIG. 17 will be sequentially described while referring to FIG. 16. In step 1701 of FIG. 17, the knowledge graph update history receiving unit 121 receives the past document information 1601 and the past difference information 1602 from the knowledge graph system 180. In the description using FIG. 3 and the description regarding step 819 of FIG. 8, the past document information and the past difference information have already been described. In step 1701, the knowledge graph update history receiving unit 121 may receive one pair of the past document information 1601 and the past difference information 1602 that are associated with each other. Note that the past difference information 1602 in the pair of the past document information 1601 and the past difference information 1602 may indicate the content of the past update of the knowledge graph 181 according to the update query 171 constructed by the cooperation of the system 101 and the language model 191 based on the past document information 1601. Alternatively, the past difference information 1602 may indicate the content of the past update of the knowledge graph 181 by an update query created manually based on the past document information 1601. In step 1702 of FIG. 17, the hint information group reception unit 122 receives a hint information group 1603 composed of one or more pieces of hint information from the knowledge graph system 180. The one or more pieces of hint information included in the hint information group 1603 may be information or knowledge (other than hint set information) included in the range of the node group or edge group indicated by the past document information 1601 or the past difference information 1602 as an update or addition target. Alternatively, the range of the node group or edge group for which hint information is to be collected in step 1702 may be wider than the range of the node group or edge group directly indicated by the past document information 1601 or the past difference information 1602 as an update or addition target. In addition, after individual hint information is extracted from the hint set information registered (recorded) in the knowledge graph system 180, each of the extracted hint information may be included in the hint information group 1603.

[0089] The loop of the processing steps from step 1703 to step 1710 of FIG. 17 may be controlled by the other function control unit 129 included in the matching hint set information specifying unit 126. While the processing included in the loop is being executed, the other function control unit 129 may control the repeated execution of the processing in the temporary hint set forming unit 123, the prompt construction unit 108, the prompt presentation unit 109, the response information reception unit 110, the temporary difference information specifying unit 124, the similarity calculation unit 125, and the evaluation value processing unit 128.

[0090] In step 1703 of FIG. 17, the temporary hint set forming unit 123 selects one or more pieces of hint information from the hint information group 1603 and forms (constructs) temporary hint set information 1623 based on the selected hint information. As the loop of the processing steps from step 1703 to step 1710 is repeated, step 1703 is usually repeatedly executed. Each time step 1703 is executed, the combination of the hint information included in the temporary hint set information 1623 changes. When determining the combination of the hint sets included in the temporary hint set information 1623 by the temporary hint set forming unit 123, for example, it may be based on reinforcement learning that reflects the results of the iterations already executed for the loop of the processing steps from step 1703 to step 1710. Alternatively, when determining the combination of the hint sets included in the temporary hint set information 1623 by the temporary hint set forming unit 123, for example, it may be based on a genetic algorithm that reflects the results of the iterations already executed for the loop. Or, the temporary hint set forming unit 123 may determine the combination of the hint sets included in the temporary hint set information 1623 each time step 1703 is executed by using any optimization method other than reinforcement learning and genetic algorithms.

[0091] In step 1704 of FIG. 17, the prompt construction unit 108 constructs a temporary prompt 1608 based on the past document information 1601 and the temporary hint set information 1623. The details of the processing performed by the prompt construction unit 108 have already been described. In the parts that have already been described about the prompt construction unit 108, replacing "document information" with "past document information", "knowledge graph information" or "hint set information" with "temporary hint set information", and "prompt" with "temporary prompt" is the description of the processing performed by the prompt construction unit 108 in step 1704. As the loop of processing steps from step 1703 to step 1710 is repeated, step 1704 is usually repeatedly executed. Each time step 1704 is executed, since the combination of hint information included in the temporary hint set information 1623 is different, the temporary prompt 1608 constructed in step 1704 also becomes different each time step 1704 is executed. In step 1705 of FIG. 17, the prompt presentation unit 109 presents the temporary prompt 1608 to the language model system 190. The language model system 190 inputs the presented temporary prompt 1608 into the language model 191. The language model 191 generates (outputs) temporary answer information 1610 which is an answer to the temporary prompt 1608. The language model system 190 presents the temporary answer information 1610 to the system 101. Note that the temporary answer information 1610 may be in the same form as the answer information 170 already described. In step 1706 of FIG. 17, the answer information reception unit 110 receives the temporary answer information 1610 from the language model system 190.

[0092] In step 1707 of FIG. 17, the temporary difference information specifying unit 124 specifies (generates) temporary difference information 1624. The temporary difference information 1624 indicates the difference that occurs in the knowledge graph 181 when it is assumed that a temporary update query is constructed based on the temporary answer information 1610 and the temporary update query is applied to the knowledge graph 181. Here, the difference may mean an update of information (node information or edge information) regarding nodes and edges that already exist in the knowledge graph 181 before the temporary update query is applied, or may mean the addition of nodes and edges that do not exist in the knowledge graph 181 before the temporary update query is applied. Such a difference may be called a difference sub-graph. The temporary difference information specifying unit 124 may specify (generate) the temporary difference information 1624 by analyzing the content of the temporary answer information 1610. At this time, the temporary difference information specifying unit 124 may receive information and knowledge regarding the range (node group or edge group) to be updated in the knowledge graph 181 when it is assumed that a temporary update query that can be constructed based on the temporary difference information 1624 is applied to the knowledge graph 181, from the knowledge graph system 180. Alternatively, the temporary difference information specifying unit 124 may request the update query construction unit 111 to construct a temporary update query, and request the update query presentation unit 112 to present the temporary update query to the knowledge graph system 180. However, instead of actually updating the knowledge graph 181 with the execution of step 1707, the temporary difference information specifying unit 124 may receive the temporary difference information 1624 indicating the assumed update result when it is assumed that the knowledge graph 181 is updated by the temporary update query, from the knowledge graph system 180.

[0093] In step 1708 of FIG. 17, the similarity calculation unit 125 calculates a similarity 1625. The similarity 1625 is the similarity between the difference (such as a difference sub-graph) indicated by the temporary difference information 1624 and the difference (such as a difference sub-graph) indicated by the past difference information 1602 in the knowledge graph 181. For example, in a case where the difference in the knowledge graph 181 means addition of nodes or edges that did not exist in the knowledge graph 181 before the update (addition of a difference subgraph), the similarity 1625 may be the topological similarity between the difference subgraph indicated by the temporary difference information 1624 and the difference subgraph indicated by the past difference information 1602. Also, for example, in a case where the difference in the knowledge graph 181 means update of information (node information or edge information) regarding nodes or edges that existed in the knowledge graph 181 before the update, the similarity 1625 may indicate the ratio of matching of the information to be updated, as a result of comparing the list of information to be updated indicated by the temporary difference information 1624 with the list of information to be updated indicated by the past difference information 1602.

[0094] In step 1709 of FIG. 17, the evaluation value processing unit 128 calculates an evaluation value 1627. The evaluation value 1627 may be calculated based at least on the similarity 1625. As already mentioned, for optimization of hint set information and the like, there is an issue that it is desirable to minimize the inclusion of useless hint information or hint information with low usefulness in the hint set information. Among the temporary hint set information 1623 that can realize the temporary difference information 1624 whose similarity to the past difference information 1602 is acceptable, if a "slim" temporary hint set information 1623 with a small number of hint information included in the temporary hint set information 1623 is selected as the optimized hint set information (compatible hint set information 1633), the above issue can be solved. Therefore, the evaluation value processing unit 128 may calculate an evaluation value 1627 by means of a function 127 based on both the similarity 1625 and the attribute value regarding the temporary hint set information 1623. Here, the attribute value regarding the temporary hint set information 1623 may be, for example, one or both of the hint information number information 1641 indicating the number of hint information included in the temporary hint set information 1623 and the temporary search range information 1644 (temporary search hop number information) indicating the search range (for example, the range included in a predetermined number of node hops starting from a certain node) when collecting the hint information included in the temporary hint set information 1623 in the knowledge graph 181. Further, the function 127 may increase the evaluation value 1627 as the similarity 1625 is higher, and decrease the evaluation value 1627 as the number of hint information indicated by the hint information number information 1641 is larger, or decrease the evaluation value 1627 as the search range indicated by the temporary search range information 1644 is wider (the higher the search node hop number).

[0095] In step 1710 of FIG. 17, the other function control unit 129 determines whether the evaluation value 1627 satisfies a predetermined condition. For example, the other function control unit 129 determines whether the evaluation value 1627 is greater than (or equal to or greater than) a predetermined threshold. If the determination result in step 1710 is affirmative, then the temporary hint set information 1623 at this time is regarded as the matching hint set information 1633, and the control transitions to step 1711. If the determination result in step 1710 is negative, the control returns to step 1703, a new temporary hint set information 1623 is formed, and then the processing steps from step 1703 to step 1709 are newly executed. Note that FIG. 17 shows an algorithm for exploratorily discovering the temporary hint set information 1623 such that the evaluation value 1627 satisfies a predetermined condition. Instead of this, it is assumed that a predetermined number of temporary hint set information 1623 are formed, and for each of the formed temporary hint set information 1623, after steps 1703 to 1709 are executed and the respective evaluation values 1627 are calculated, the temporary hint set information 1623 that brings a relatively high (for example, the highest value) evaluation value 1627 may be regarded as the matching hint set information 1633.

[0096] In step 1711 of FIG. 17, the compliance information registration instruction unit 130 presents a compliance information registration instruction 1630 to the knowledge graph system 180. The compliance information registration instruction 1630 is for requesting to register (record) the compliance hint set information 1633 as the hint set information 701, or to register (record) the compliance search range information 1634 as the search range information. The compliance hint set information 1633 is the temporary hint set information 1623 that yields an evaluation value 1627 satisfying a predetermined condition in step 1710 (or the temporary hint set information 1623 that yields a relatively high (e.g., the highest value) evaluation value 1627). The compliance search range information 1634 is the temporary search range information 1644 (temporary search hop count information) indicating the search range (e.g., the range included in a predetermined number of node hops starting from a certain node) when collecting the hint set information included in the temporary hint set information 1623 regarded as the compliance hint set information 1633 in the knowledge graph 181. Note that the information registered (recorded) in the knowledge graph system 180 by the compliance information registration instruction 1630 is not limited to the above-mentioned compliance hint set information 1633 and compliance search range information 1634. Generally, the information registered (recorded) in the knowledge graph system 180 by the compliance information registration instruction 1630 may be information indicating a method for collecting hint information for forming optimized hint set information.

[0097] 4. Processing of Supplementary Document Display (FIGS. 18 to 19) In an embodiment of the present disclosure, when the system 101 receives the knowledge graph information 167 from the knowledge graph system 180, it is expected that the prompt 168 can be constructed along with the object of the knowledge represented by the knowledge graph 181 (e.g., facilities and infrastructure), or the prompt 168 can be constructed after supplementing or correcting the matters omitted in the document information 162. Here, the embodiment of the present disclosure not only utilizes the knowledge graph information 167 for constructing the prompt 168, but also presents (displays) the supplementary document information 1841 (supplementary document), which is the result of supplementing or correcting the document information 162 by the knowledge graph information 167, to the person handling the object of the knowledge represented by the knowledge graph 181 (e.g., facilities and infrastructure), and may confirm the validity of the content of the supplementary document information 1841 (supplementary document). FIG. 18 shows a functional configuration 1800 related to the display of the supplementary document information 1841 (supplementary document) and the confirmation of the validity of the content of the supplementary document information 1841 (supplementary document). As shown in FIG. 18, in addition to the functional units already described, the system 101 may have a supplementary document information generation unit 141, a supplementary document display control unit 142, and a supplementary document validity information reception unit 143. Also, as shown in FIG. 18, in addition to the information (data) already described, the system 101 may handle the supplementary document information 1841 and the supplementary document validity information 1843. FIG. 19 shows a display screen (display of the supplementary document) 1900 displayed by the display or output device 207 based on the control performed by the supplementary document display control unit 142. FIG. 19 also shows the information input by the person handling the object of the knowledge represented by the knowledge graph 181 (e.g., facilities and infrastructure) using the input device 206.

[0098] Hereinafter, with reference to FIGS. 18 and 19, the process related to the display of the supplementary document formed by supplementing or correcting the document information 162 based on the knowledge graph information 167 and the confirmation of the validity of the content of the supplementary document will be described. Up to the point where system 101 receives document information 162 and knowledge graph information 167 and constructs prompt 168 (or its candidates) based on this information, the processing already described (for example, the processing of steps 801 to 813 in FIG. 8) may be the same. However, before presenting the constructed prompt 168 to the language model system 190, the following processing regarding the display of supplementary documents and the confirmation of the validity of the content of the supplementary documents is performed. First, the supplementary document information generation unit 141 generates supplementary document information 1841 based on the document information 162 and the knowledge graph information 167. The supplementary document information 1841 reflects the context of the objects of knowledge (for example, facilities and infrastructure) represented by the knowledge graph 181 with respect to the document information 162, or is a supplement or correction regarding matters omitted, etc. in the document information 162 with respect to the document information 162. Note that the prompt construction unit 108 may also serve as the supplementary document information generation unit 141. Next, the supplementary document display control unit 142 controls the display or output device 207 to perform display based on the supplementary document information 1841. In the example of the display screen (display of supplementary document) shown in FIG. 19, in window 1901, corresponding to document 161 (document information 162) which is a document input by a person handling the subject of knowledge represented by knowledge graph 181 (for example, a facility or infrastructure), i.e., "During the regular maintenance on October 1, 2022, at facility X, replace parts numbered 100 and 101.", a chart showing the relationship between the text "Propose update information.", "During the regular maintenance on October 1, 2022, perform the following:", "(1) Based on the replacement regulations of TBM, replace part A (part number 100) at part Y of facility X.", "(2) Based on the replacement regulations of TBM, replace part B (part number 101) at part Y of facility X." of supplementary document information 1841 (supplementary document) and the solid-line frames each having labels "facility X", "part Y", "part A part number 100", "(TBM for part A) 10000h", "part B part number 101", "(TBM for part B) 10000h" is displayed. Note that "TBM" is an abbreviation of "Time Based Maintenance" and means the time interval for performing maintenance (for example, part replacement) regardless of the presence or absence of a failure. A person who views the text or chart which is supplementary document information 1841 (supplementary document) as shown in FIG. 19 determines the validity of the content of supplementary document information 1841 (supplementary document). When the person determines that the content of supplementary document information 1841 (supplementary document) is valid, as shown in FIG. 19, "Approve the proposal." is input using input device 206. This information "Approve the proposal." becomes supplementary document validity information 1843. When supplementary document validity information reception unit 143 receives supplementary document validity information 1843 from input device 206, it transmits supplementary document validity information 1843 to prompt construction unit 108. After confirming that supplementary document validity information 1843 affirms the validity of supplementary document information 1841, prompt construction unit 108 transmits the constructed prompt 168 to prompt presentation unit 109. The processing after prompt 168 is transmitted to prompt presentation unit 109 may be the same as that already described (for example, steps 814 and subsequent steps in FIG. 8). If the supplementary document validity information 1843 negates the validity of the supplementary document information 1841, the prompt construction unit 108 may reconstruct the prompt 168, and the supplementary document information generation unit 141 may regenerate the supplementary document information 1841.

[0099] As described above, if the display of the supplementary document information 1841 (supplementary document) and the confirmation of the validity of the content of the supplementary document information 1841 (supplementary document) are performed, for example, at a stage prior to verifying the validity of the update query 171 such as step 817 in FIG. 8, a person handling the object of the knowledge represented by the knowledge graph 181 (for example, a facility or infrastructure) can confirm whether the content of the update of the knowledge graph 181 that the system 101 is about to perform is valid.

[0100] 4.5. Processing for displaying uniqueness inquiries (FIGS. 20-21) In the embodiment of the present disclosure, the degree of omission or the like in the document information 162 received by the system 101 may be too large. Therefore, even if the system 101 attempts to supplement or correct the matters omitted or the like in the document information 162 based on the knowledge graph information 167, it may not be possible to uniquely identify the intention of the person who input the document information 162. Therefore, in the embodiment of the present disclosure, processing such as displaying a uniqueness inquiry for uniquely identifying the intention of the person who input the document information 162 may be performed. FIG. 20 shows a functional configuration 2000 related to the display of option information 2044 for uniqueness inquiries and the confirmation of the intention of the person who input the document information 162. As shown in FIG. 20, in addition to the functional units already described, the system 101 may include a uniqueness determination unit 144, an option display control unit 145, and a selection information reception unit 146. Also, as shown in FIG. 20, in addition to the information (data) already described, the system 101 may handle option information 2044 and selection information 2046. Fig. 21 shows a display screen (display of uniqueness query) 2100 displayed by the display or output device 207 based on the control performed by the option display control unit 145. Fig. 21 also shows information input using the input device 206 by a person (such as a person who input the document information 162) who handles the subject of knowledge represented by the knowledge graph 181 (e.g., a facility or infrastructure).

[0101] The process of displaying the uniqueness query and verifying the intent of the person who entered the document information 162 will be described below with reference to FIGS. 20 and 21. FIG. The process up to the point where the system 101 receives the document information 162 and the knowledge graph information 167 and constructs (a candidate for) the prompt 168 based on this information may be the same as the process already described (for example, the process of steps 801 to 813 in FIG. 8). However, before presenting the constructed prompt 168 to the language model system 190, (in a situation where the intended uniqueness of the document information 162 is in doubt) a process for displaying a uniqueness inquiry and confirming the intention of the person who input the document information 162 is performed as shown below. First, the uniqueness determination unit 144 determines whether the content of the update of the knowledge graph 181 based on the document information 162 can be uniquely identified, based on the document information 162 and the knowledge graph information 167. If it is determined that the content of the update of the knowledge graph 181 based on the document information 162 cannot be uniquely identified, the uniqueness determination unit 144 generates option information 2044 indicating options for the content of the update of the knowledge graph 181, based on the document information 162. Note that the uniqueness determination unit 144 may utilize the language model 191 to make this determination. Also, the prompt construction unit 108 may also serve as the uniqueness determination unit 144. Next, the option display control unit 145 controls the display or output device 207 to perform display based on the option information 2044 . In the example of the display screen in FIG. 21 (display of uniqueness inquiry), in window 2101, corresponding to document 161 (document information 162) which is a document input by a person handling the object of knowledge (e.g., facility or infrastructure) represented by knowledge graph 181, i.e., "In the regular maintenance on October 1, 2022, at facility X, part number 100 was replaced.", a chart is displayed showing the relationship between the text of option information 2044, which is "The update cannot be uniquely identified.", "(1) Based on the replacement regulation of the TBM, replace part A (part number 100) at part Y of facility X.", "(2) Due to a sudden failure, replace part A (part number 100) at part Z of facility X.", and the solid-line frames each having labels of "facility X", "part Y", "part A (part number 100) of (part Y)", "TBM 10000h for (part A of part Y)", "operation time: 9000h for (part A of part Y)", "part Z", "part A (part number 100) of (part Z)", "TBM 10000h for (part A of part Z)", "operation time: 1000h for (part A of part Z)". The example in FIG. 21 shows that even considering knowledge graph information 167, it cannot be uniquely determined whether part A indicated by "replace part number 100" in document information 162 is the part A included in part Y or the part A included in part Z. A person who views the text and chart showing option information 2044 in FIG. 21 determines which of (1) and (2) shown as options was intended by document information 162. The person inputs selection information 2046 indicating the determined option using input device 206. In the example shown in FIG. 21, the person inputs "Approve proposal (1)." using input device 206. That is, among the options of (1) and (2), information indicating that option (1) was selected becomes selection information 2046. When selection information receiving unit 146 receives selection information 2046 from input device 206, it transmits selection information 2046 to prompt construction unit 108. The prompt construction unit 108 constructs a prompt 168 so as to reflect the selection information 2046. The prompt construction unit 108 transmits the constructed prompt 168 to the prompt presentation unit 109. The processing after the prompt 168 is transmitted to the prompt presentation unit 109 may be the same as that already described (for example, steps 814 and later in FIG. 8).

[0102] As described above, even when the system 101 attempts to supplement or correct matters omitted or the like in the document information 162 based on the knowledge graph information 167, and even when the intention of the person who input the document information 162 cannot be uniquely identified, the system 101 can appropriately update the knowledge graph 181 after confirming the intention of the person who input the document information 162 and the like.

[0103] 4 of 6. Process of editing knowledge graph information (hint information) (FIGS. 22 to 23) In the embodiment of the present disclosure, the knowledge graph 181 has node information related to each node and edge information related to each edge as knowledge graph information. It has already been explained that these node information and edge information (other than the hint set information) may be called hint information. Here, the embodiment of the present disclosure may have a user interface that enables manual editing of the knowledge graph information (hint information) related to the nodes and edges of the knowledge graph 181. FIG. 22 shows a functional configuration 2200 for realizing a user interface that enables manual editing of knowledge graph information (hint information). As shown in FIG. 22, in the embodiment of the present disclosure, the system 101 may have a knowledge graph information editing screen display control unit 147, a knowledge graph editing information reception unit 148, and a knowledge graph information editing instruction unit 149 in addition to the functional units already described. Also, as shown in FIG. 22, in the embodiment of the present disclosure, the system 101 may handle knowledge graph editing information 2248 and a knowledge graph information editing instruction 2249 in addition to the information (data) already described. FIG. 23 shows a display screen (display of editing of knowledge graph information (hint information)) 2300 as a user interface that enables manual editing of a knowledge graph (hint information).

[0104] Hereinafter, with reference to FIGS. 22 and 23, processing related to realizing a user interface that enables manual editing of a knowledge graph (hint information) will be described. First, a person handling the system 101 or the knowledge graph system 180 may input via the input device 206 an indication of the intention to edit the knowledge graph information (hint information). In FIG. 23, as an example where an input indicating the intention to edit the knowledge graph information (hint information) has been made, the input content "Add hint information for equipment X." is displayed in the window 2301. The input indicating the intention to edit the knowledge graph information (hint information) is received by the knowledge graph edit information reception unit 148. The knowledge graph edit information reception unit 148 requests the knowledge graph information edit screen display control unit 147 to realize a knowledge graph information (hint information) edit screen suitable for the input indicating the intention to edit the knowledge graph information (hint information). In the above example, the knowledge graph edit information reception unit 148 requests the knowledge graph information edit screen display control unit 147 to realize a knowledge graph information edit screen that enables editing of the knowledge graph information (hint information) related to equipment X. The knowledge graph information editing screen display control unit 147 controls the display or output device 207 to realize a knowledge graph information (hint information) editing screen. In FIG. 23, in order to realize a knowledge graph information (hint information) editing screen that enables editing of the knowledge graph information (hint information) related to Facility X, the text "The following figure shows the asset nodes included in Facility X and the edges between the assets." and "Please click on the asset node or edge for which you want to edit the hint information." is displayed in window 2301, and a chart related to Facility X showing the relationship between the solid-line frames each having the labels "Facility X", "Part Y", "Component A Part Number 100", and "Component B Part Number 101" is displayed. Incidentally, in order to realize an editing screen such as that in FIG. 23, the knowledge graph information editing screen display control unit 147 may receive node information and edge information in the knowledge graph corresponding to Facility X from the knowledge graph system 180.

[0105] The person who edits the knowledge graph information (hint information) selects either a node (for example, an asset node) or an edge (for example, an edge connecting asset nodes) in the chart related to Facility X displayed in window 2301 for which they want to edit the knowledge graph information (hint information). For example, the person who edits the knowledge graph information (hint information) uses the input device 206 (for example, a mouse) to perform a click input on either a node or an edge in the chart related to Facility X. The selection input information of the node or edge input via the input device is received by the knowledge graph editing information reception unit 148. The knowledge graph editing information reception unit 148 requests the knowledge graph information editing screen display control unit 147 to realize an editing screen for the knowledge graph information (hint information) related to the node or edge indicated by the selection input information. In the example of FIG. 23, the knowledge graph editing information reception unit 148 requests the knowledge graph information editing screen display control unit 147 to realize a pop-up window 2302 which is an editing screen for the knowledge graph information (hint information) related to "Component A Part Number 100" for which the click input was made. The knowledge graph information editing screen display control unit 147 controls the display or output device 207 to realize a pop-up window 2302 which is an editing screen for knowledge graph information (hint information). In FIG. 23, a pop-up window 2302 which is an editing screen for knowledge graph information (hint information) related to "Component A, Part Number 100" is displayed.

[0106] The person who edits the knowledge graph information (hint information) edits the hint information within the pop-up window 2302. In FIG. 23, within the pop-up window 2302, it is possible to execute the editing of the knowledge graph information (hint information) (here, asset node information other than hint set information) related to "Component A, Part Number 100". When the pop-up window 2302 pops up, the knowledge graph information (hint information) before editing may be displayed in the pop-up window 2302. That is, the editing of the knowledge graph information (hint information) using the pop-up window 2302 may include not only the addition of knowledge graph information (hint information) but also the correction of knowledge graph information (hint information) and the deletion of hint information. When determining the editing of the knowledge graph information (hint information) performed using the pop-up window 2302, a click input is made on the "Execute" icon. When discarding (not finalizing) the editing of the knowledge graph information (hint information) performed using the pop-up window 2302, a click input is made on the "Cancel" icon. The knowledge graph information (hint information) after the finalized editing is referred to as knowledge graph editing information 2248. The knowledge graph editing information 2248 is received by the knowledge graph editing information reception unit 148. The knowledge graph editing information reception unit 148 transmits the knowledge graph editing information 2248 to the knowledge graph information editing instruction unit 149. The knowledge graph information editing instruction unit 149 presents a knowledge graph information editing instruction 2249 to the knowledge graph system 180. The knowledge graph information editing instruction 2249 is for requesting the knowledge graph 181 to reflect the editing (addition, correction, deletion) of the knowledge graph information (hint information) included in the knowledge graph editing information 2248. The knowledge graph system 180 presented with the knowledge graph information editing instruction 2249 reflects the editing (addition, correction, deletion) of the knowledge graph information (hint information) included in the knowledge graph editing information 2248 in the knowledge graph 181.

[0107] In addition, in FIG. 23, an example of editing the knowledge graph information (hint information) related to the asset node is shown, but the target of editing the knowledge graph information (hint information) may be a symptom node, a failure mode node, or an edge. Further, the target of editing the knowledge graph information (hint information) is not limited to the nodes and edges included in the instance group, and may be a schema node or an edge included in the schema. Also, in the above description, an example of editing the node information associated with a node or the edge information associated with an edge is shown, but the hint set information that can be associated with a plurality of nodes or a plurality of edges may also be editable in the same way. For example, along the lines of the few shot learning method, to an asset node at a relatively higher level (for example, an asset node indicating a water supply mechanism), as hint set information, information showing sentences such as "Here is an example.", "When replacing an asset, update the installation date of the asset.", and "When adding an asset, connect the added asset and one of the existing assets with an edge." may be added in the system 101.

[0108] As described above, it is possible to edit various knowledge graph information (for example, hint information) possessed by the knowledge graph 181 in accordance with the intention of the person handling the system 101 or the knowledge graph system 180.

[0109] 5. Others (Modification Examples) The present disclosure is not limited to the above-described embodiments and includes various modifications. Some of the configurations and processes of the embodiments may be replaced with those of other conceivable embodiments. The configurations and processes of the embodiments may be added with those of other conceivable embodiments. For example, in the present disclosure, there may be the following modifications of the embodiments.

[0110] (A) Modifications to the construction of the asset identification request prompt (Figs. 24 to 25) In the above-described embodiment, the asset identification request prompt construction unit 103 constructed the asset identification request prompt 163 based on the document information 162. However, when the language model 191 has not been trained by machine learning specialized in domain knowledge regarding the objects (e.g., facilities and infrastructure) represented in the knowledge graph 181, the language model 191 may not be able to appropriately extract the names of elements (e.g., assets) included in the object (e.g., facilities and infrastructure) from the document information 162. Therefore, in the modification, the asset identification request prompt 163 may include the information and knowledge included in the knowledge graph 181 so that the language model 191 can appropriately extract the names of the elements (e.g., assets) from the document information 162. Fig. 24 shows the functional configuration related to the construction of the asset identification request prompt 163 in the modification. In the modification, the system 101 may have an asset name list information request unit 114 and an asset name list information reception unit 115 in addition to the functional units already described. In the modification, the system 101 may handle an asset name list information request 2414 and an asset name list information 2415 in addition to the information (data) already described. Fig. 25 shows an asset identification request prompt template 2500 that can be used in the modification. Fig. 25 has already been described. Some additional explanations are given below. The asset identification request prompt template 2500 may include text indicating the outline of the request, such as "Please identify the assets from the following text and output a specific asset list that is a list enumerating the identified assets."; text indicating a general description of the assets, such as "Assets are nouns. Assets are the names of devices."; and text specifying the response format when the language model 191 answers, such as "Please format the specific asset list in format Y." Also, the asset identification request prompt template 2500 may include text prompting the language model 191 to use the asset name list information 2415 as a hint, such as "The candidates for assets are as follows. However, other things outside the following candidates may also be assets.", and the asset name list information 2415 may be insertable at the {hint} location. Furthermore, the asset identification request prompt template 2500 may include the text "#Article:" and the document information 162 may be insertable at the {text} location. The asset identification request prompt template 2500 may include the text "#Specific asset list in format Y:". Later, the specific asset response information 164 (specific asset list), which is the response from the language model 191, may be inserted immediately after "#Specific asset list in format Y:". Note that an example of the above format Y is the CSV format.

[0111] With reference to FIGS. 24 and 25, the processing up to the construction of the asset identification request prompt in the modification example is described. First, the asset name list information request unit 114 presents an asset name list information request 2414 to the knowledge graph system 180. The asset name list information request 2414 is for requesting asset name list information 2415, which is information indicating a list of names of elements (such as assets) included in the objects of knowledge (such as facilities and infrastructure) represented in the knowledge graph 181. The range for collecting the names of elements (such as assets) may be the entire knowledge graph 181, or a part of the knowledge graph 181 determined based on the content of the document information 162. Also, the elements for which names are to be collected may be other than assets. In response to the presentation of the asset name list information request 2414, the knowledge graph system 180 may collect the names of elements (such as assets) from the whole or a part of the knowledge graph 181. Or, the knowledge graph system 180 may read out the list information of the names of elements (such as assets) prepared in advance. The knowledge graph system 180 presents to the system 101 the information indicating the list of the collected or read-out names of elements (such as assets) as the asset name list information 2415. The asset name list information reception unit 115 receives the asset name list information 2415 from the knowledge graph system 180. The asset name list information reception unit 115 transmits the asset name list information 2415 to the asset identification request prompt construction unit 103. Based on the document information 162 and the asset name list information 2415, the asset identification request prompt construction unit 103 constructs an asset identification request prompt 163. For example, if the asset identification request prompt template 2500 shown in FIG. 25 is used, the asset identification request prompt construction unit 103 may insert the asset name list information 2415 at the {hint} position and the document information 162 at the {text} position in the asset identification request prompt template 2500 to construct the asset identification request prompt 163. The processing after the construction of the asset identification request prompt 163 may be the same as that already described (for example, steps 803 and later in FIG. 8).

[0112] Such a modification can increase the probability that the language model 191 generates (outputs) appropriate specific asset response information 164 (specific asset list).

[0113] (B) Variation of knowledge graph information request and knowledge graph information reception (Fig. 26) In the above embodiment, as shown in steps 807 and 808 of Fig. 8, when the knowledge graph information request 166 is presented to the knowledge graph system 180 from the system 101 once, in response, the knowledge graph system 180 seems to present the knowledge graph information 167 (for example, hint set information and hint information) related to the specific asset 182 and the related asset 183 to the system 101 all at once. In the variation, the interaction between the system 101 and the knowledge graph system 180 may be more interactive than that shown in Fig. 8.

[0114] Fig. 26 shows a flowchart 2600 of the process in this variation. In this variation, among the flowcharts shown in Fig. 8, the parts of steps 807 and 808 are replaced by those shown in Fig. 26. In this variation, after step 804 of Fig. 8, step 2605 of Fig. 26 may be executed. In step 2605 of Fig. 26, the knowledge graph information request unit 106 may present a preliminary knowledge graph information request to the knowledge graph system 180. The preliminary knowledge graph information request is for requesting (inquiring) the knowledge graph system 180 for the information of the asset identifier for those assets (specific assets) whose asset identifiers are unknown among the assets named in the document information 162. For example, in the example of the document information 162A (preservation document A) in Fig. 11, the asset identifier P001 is explicitly shown for the "water supply mechanism", but the asset identifiers are not shown for the "second pump" and the "impeller". The information of the asset identifier of the "second pump" included in the "water supply mechanism P001" and the information of the asset identifier of the "impeller" included in the "second pump" can be obtained by the preliminary knowledge graph information request. In response to a preliminary knowledge graph information request, the knowledge graph system 180 collects preliminary knowledge graph information consisting of information on the queried asset identifiers. The knowledge graph system 180 presents the preliminary knowledge graph information to the system 101. In step 2606 of FIG. 26, the knowledge graph information reception unit 107 receives preliminary knowledge graph information consisting of information on asset identifiers from the knowledge graph system 180. By steps 2605 and 2606, the system 101 can know the asset identifier of the asset (specific asset) whose name is indicated in the document information 162. After step 2606, the control transitions to step 2607.

[0115] The system 101 and the knowledge graph system 180 may interactively transmit and receive knowledge graph information (for example, node information and edge information) related to each of the specific asset 182 and the related asset 183 related to the document information 162, in sequence, by the processing step group indicated by the loop of steps 2607, 2608, and 2609. Specifically, in step 2607 of FIG. 26, the knowledge graph information request unit 106 presents a knowledge graph information request to the knowledge graph system 180. In this modification example, the knowledge graph information request presented in step 2607 includes either any of the specific asset information that is known to the system 101 at this time, or any of the related asset information. For example, when step 2607 is first executed and the system 101 is known about the specific asset 182 related to the document information 162 but is unknown about the related asset 183, the knowledge graph information request when step 2607 is first executed may include any of the specific asset information of the known specific asset 182. The knowledge graph system 180 collects knowledge graph information regarding the specific asset information or related asset information included in the presented knowledge graph information request. Here, the collected knowledge graph information may include related asset information that identifies a related asset 183 that is unknown to the system 101 at this point in time. The knowledge graph system 180 presents the collected knowledge graph information to the system 101. In step 2608 of FIG. 26, the knowledge graph information receiving unit 107 receives knowledge graph information from the knowledge graph system 180. The knowledge graph information received here is knowledge graph information regarding the specific asset information or related asset information included in the knowledge graph information request in the immediately preceding step 2607. As already pointed out, this knowledge graph information may include related asset information that identifies a related asset 183 that was unknown to the system 101. In step 2609 of FIG. 26, the knowledge graph information request unit 106 determines whether there remains any information regarding an asset (specific asset) whose name is included in the document information 162 (specific asset information) and information regarding a related asset 183 included in the knowledge graph information received in step 2608 (related asset information), which was not included in the knowledge graph information request in step 2607 and thus not presented to the knowledge graph system 180. If the determination result in step 2609 is affirmative (i.e., if there is any information that was not presented in step 2607 among the specific asset information or related asset information), the control returns to step 2607. If the determination result in step 2609 is negative, the control transitions to step 810.

[0116] In the above-described modification example, compared to the case where the processing steps of the flowchart in FIG. 8 are realized, the degree of batch collection of knowledge graph information in the knowledge graph system 180 can be relaxed. Therefore, in the above-described modification example, the function of the knowledge graph system 180 for controlling the knowledge graph 181 can be simplified.

[0117] (C) Modification Example of Hint Set Information (FIG. 27) In the above-described embodiment, it is assumed that one set of hint information is associated with an asset node at a relatively higher level (for example, an asset node indicating a water supply mechanism). In a modification, there may be a plurality of sets of hint information associated with one asset node. FIG. 27 shows an aspect 2700 of the hint set information in this modification. In FIG. 27, hint set information 2771, hint set information 2772, and hint set information 2773 are associated with an asset node 2721 indicating a "water supply mechanism P001", which is an asset node at a relatively higher level. For example, from the system 101, as a combination of specific asset information or related asset information, depending on which combination of asset nodes is presented, the knowledge graph system 180 may select which hint set information to present to the system 101. For example, when "water supply mechanism P001", "pump P002B", and "impeller P003B" are presented simultaneously from the system 101, the knowledge graph system 180 may present the hint set information 2771 to the system 101. When "water supply mechanism P001" and "pump P002B" are presented simultaneously from the system 101, the knowledge graph system 180 may present the hint set information 2772 to the system 101. When only "water supply mechanism P001" is presented from the system 101, the knowledge graph system 180 may present the hint set information 2773 to the system 101. In the above modification, according to the content of the knowledge graph information request 166 presented by the system 101 to the knowledge graph system 180, the knowledge graph system 180 can present more appropriate hint set information to the system 101.

[0118] (D) Modification regarding the display of supplementary documents (FIG. 28) As shown in FIG. 18, in the above-described embodiment, the supplementary document validity information 1843 is used for confirming the validity of the prompt 168 constructed by the prompt construction unit 108. In a modified example, the supplementary document validity information 1843 may be used to confirm the validity of the update query 171 constructed by the update query construction unit 111. FIG. 28 shows the functional configuration in the modified example. Comparing FIG. 18 with FIG. 28, the information used to generate the supplementary document information 1841 is different between the supplementary document information generation unit 141 in FIG. 18 and the supplementary document information generation unit 2841 in FIG. 28. Also, the input destination of the supplementary document validity information 1843, which was the prompt construction unit 108 in FIG. 18, is the update query construction unit 111 in FIG. 28. Hereinafter, the processing in FIG. 28 (the parts different from FIG. 18) will be described. As shown in FIG. 28, in the modified example, the supplementary document information 1841 (supplementary document) is generated after the system 101 (answer information reception unit 110) receives the answer information 170, which is the output from the language model 191. The supplementary document information generation unit 2841 generates the supplementary document information 1841 (supplementary document) based on the answer information 170 (or based on the answer information 170 and the document information 162). The supplementary document validity information 1843 input by a person handling the system 101 or the knowledge graph system 180 via the input device 206 is input to the update query construction unit 111 (or a verification unit not shown). When the supplementary document validity information 1843 indicates that the validity of the supplementary document information 1841 (supplementary document) is affirmed, the update query construction unit 111 may construct the update query 171. When the supplementary document validity information 1843 indicates that the validity of the supplementary document information 1841 (supplementary document) is denied, the update query construction unit 111 may abort the construction of the update query 171. In this case, the system 101 may, for example, start over from the processing of the prompt construction unit 108. Note that when the supplementary document validity information 1843 is input to a verification unit (not shown), the supplementary document validity information 1843 may be reflected in the verification result of the update query in the verification unit. The above-described modification example can reflect the judgment result of the person handling the system 101 or the knowledge graph system 180 on both the validity of supplementing and correcting the omitted parts in the document information 162 based on the knowledge graph information 167 and the validity of the answer information 170 generated (output) by the language model 191 in a series of processes performed in the system 101.

[0119] (E) Modification example regarding the display of uniqueness inquiries (Fig. 29) As shown in Fig. 20, in the above-described embodiment, the selection information 2046 is used by the prompt construction unit 108 to construct the prompt 168. In the modification example, when the system 101 cannot uniquely identify the content of the update of the knowledge graph 181 based on the document information 162 even using the knowledge graph information 167, the system 101 may construct separate prompts for each option that could be the content of the update of the knowledge graph 181 and obtain separate answer information 170 from the language model 191. Then, the system 101 may select one of the answer information 170 based on the selection information 2046. Hereinafter, the process in Fig. 29 (the part different from Fig. 20) will be described. In the modification example, when the prompt construction unit 108 determines that it cannot uniquely identify the content of the update of the knowledge graph 181 based on the document information 162 even using the knowledge graph information 167, the prompt construction unit 108 constructs separate prompts for each option that could be the content of the update of the knowledge graph 181. Incidentally, for this determination, the prompt construction unit 108 may utilize the language model 191. When separate prompts are constructed, separate answer information 170 for each prompt is provided from the language model system 190 to the system 101 (answer information reception unit 110). The uniqueness determination unit 2944 determines whether there are multiple pieces of response information 170 for the document information 162. If there is one piece of response information 170 for one piece of document information 162, the uniqueness of the document information 162 is ensured, and subsequent processing may be the same as the processing described in FIG. 8. If there are multiple pieces of response information 170 for one piece of document information 162, the uniqueness determination unit 2944 generates option information 2044 that shows the content of each piece of response information 170 as an option. The selection information 2046 input by a person handling the system 101 or the knowledge graph system 180 via the input device 206 is input to the update query construction unit 111. The update query construction unit 111 selects the response information 170 corresponding to the option indicated by the selection information 2046 and constructs an update query 171 using the selected response information 170. According to the above modification example, when the uniqueness of the document information 162 is not sufficient, a person handling the system 101 or the knowledge graph system 180 can select the content of the update of the knowledge graph 181 using the response information 170 which is direct information for constructing the update query 171.

[0120] (F) Modification Example without Using Hint Set Information In the above embodiment, in the knowledge graph 181, as knowledge graph information, it was possible that hint set information existed or hint set information could be additionally registered (recorded). In the modification example, in the knowledge graph 181, as knowledge graph information, there is node information related to the nodes included in the knowledge graph 181 (other than hint set information) and edge information related to the edges included in the knowledge graph 181, but hint set information may not exist in the knowledge graph 181 (even if it is additionally registered (recorded)). The above modification example has the effect that although the processing load of the system 101 may increase in that the system 101 constructs a hint set information aspect each time when constructing a prompt, since there is no hint set information in the knowledge graph 181, the management of the information in the knowledge graph 181 is simplified.

[0121] (G) Variation using only hint set information In the above embodiment, in the knowledge graph 181, as knowledge graph information, node information (other than hint set information) related to nodes included in the knowledge graph 181 and edge information related to edges included in the knowledge graph 181 may exist. In the variation, although there is hint set information associated with, for example, relatively higher-level asset nodes in the knowledge graph 181, node information (other than hint set information) related to nodes included in the knowledge graph 181 and edge information related to edges included in the knowledge graph 181 do not necessarily have to exist in the knowledge graph 181. Although the above variation may require effort to initially aggregate and register (record) knowledge graph information related to a plurality of nodes and edges as hint set information associated with, for example, relatively higher-level asset nodes, it has the effect of maintaining a state where the knowledge graph information is aggregated without being dispersed within the knowledge graph 181.

[0122] (H) Variation of generating an update query for the language model In the above embodiment, based on the answer information 170 generated (output) by the language model 191, the update query construction unit 111 within the system 101 constructed the update query 171. In the variation, the content of the request to the language model 191 based on the prompt 168 may be used as a generation request for the update query 171. In the variation, the language model 191 generates (outputs) the update query 171, and the system 101 receives the update query 171. The system 101 presents the received update query 171 to the knowledge graph system 180. In the variation, the system 101 does not necessarily have to have the update query construction unit 111. Although the above variation may reduce the degree of freedom in constructing the update query 171, on the other hand, it can achieve a simplification of the functional configuration of the system 101.

[0123] The technical matters shown in each of the embodiments of the present disclosure and the modified examples of the embodiments described above can be combined as appropriate as long as no technical contradiction occurs.

Claims

1. A system comprising: a document information receiving unit that receives document information which is information included in a document that triggers an update to a knowledge graph; a knowledge graph information receiving unit that receives, from a knowledge graph system that manages the knowledge graph, knowledge graph information which is information included in the knowledge graph; a prompt construction unit that constructs a prompt based on the document information and the knowledge graph information, wherein the prompt is for requesting a presentation of information used for constructing an update query for updating the knowledge graph; a prompt presentation unit that presents the prompt to a language model system that manages a language model; an answer information receiving unit that receives answer information indicating an answer from the language model system in response to the prompt; an update query construction unit that constructs the update query based on the answer information; and an update query presentation unit that presents the update query to the knowledge graph system.

2. The system according to claim 1, wherein the knowledge graph represents knowledge about one or more assets, and the system further comprises a knowledge graph information request unit that presents a knowledge graph information request to the knowledge graph system, the knowledge graph information request including specific asset information that specifies one or more of the assets described in the document as specific assets; the knowledge graph information includes hint information; the hint information is about one or more of the specific assets or one or more related assets related to one or more of the specific assets; and the prompt construction unit constructs the prompt based on the document information and the hint information.

3. The system according to claim 2, wherein the system further comprises a hint set information registration instruction unit that presents a hint set information registration instruction for updating the knowledge graph to the knowledge graph system so that a set of the hint information composed of one or more of the hint information is used as hint set information about any asset.

4. The system according to claim 2, wherein the system further comprises An asset identification request prompt construction unit that constructs an asset identification request prompt for requesting the language model to identify one or more of the specific assets described in the document based on the document information; An asset identification request prompt presentation unit that presents the asset identification request prompt to the language model system; A specific asset response information reception unit that receives specific asset response information indicating a response from the language model system to the asset identification request prompt; The knowledge graph information request unit determines the specific asset information to be presented to the knowledge graph system based on the specific asset response information.

5. The system according to claim 4, The system further includes An asset name list information reception unit that receives asset name list information indicating a list of asset names in the knowledge graph from the knowledge graph system; The asset identification request prompt construction unit constructs the asset identification request prompt based on the document information and the asset name list information.

6. The system according to claim 1, The knowledge graph represents knowledge about one or more assets, The system further includes A knowledge graph information request unit that presents a knowledge graph information request to the knowledge graph system, where the knowledge graph information request includes specific asset information that identifies one or more of the assets described in the document as specific assets. The knowledge graph information includes hint information or hint set information. The hint information relates to one or more of the specific assets or one or more related assets related to one or more of the specific assets. The hint set information relates to any of the specific assets or the related assets. When the received knowledge graph information includes the hint set information, the prompt construction unit constructs the prompt based on the document information and the hint set information. When the received knowledge graph information does not include the hint set information, the prompt construction unit constructs the prompt based on the document information and the hint information. System.

7. The system according to claim 1, wherein the knowledge graph represents knowledge about one or more assets, the system further comprises a knowledge graph information request unit that presents a knowledge graph information request to the knowledge graph system, the knowledge graph information request including specific asset information that specifies one or more of the assets described in the document as specific assets, having the knowledge graph information request unit, the knowledge graph information includes hint set information, the hint set information is related to any one of one or more of the specific assets or one or more related assets related to one or more of the specific assets, the prompt construction unit constructs the prompt based on the document information and the hint set information. System.

8. The system according to claim 1, wherein the knowledge graph represents knowledge about one or more assets, the system further comprises a knowledge graph update history reception unit that receives, from the knowledge graph system, past document information that is information included in a document that triggered the update when the knowledge graph was updated in the past, and past difference information indicating the changes to the knowledge graph due to the update, a hint information group reception unit that receives, from the knowledge graph system, a hint information group consisting of hint information about one or more of the assets, having a temporary hint set formation unit that selects one or more of the hint information from the hint information group to form temporary hint set information consisting of the selected hint information, the prompt construction unit constructs a temporary prompt based on the past document information and the temporary hint set information, the prompt presentation unit presents the temporary prompt to the language model system, the answer information reception unit receives temporary answer information indicating an answer from the language model system to the temporary prompt, the system further comprises A provisional difference information specifying unit that constructs a provisional update query based on the provisional answer information and specifies provisional difference information indicating changes to the knowledge graph when the provisional update query is applied to the knowledge graph; A similarity calculation unit that calculates the similarity between the changes to the knowledge graph indicated by the past difference information and the changes to the knowledge graph indicated by the provisional difference information; By controlling the provisional hint set formation unit, the prompt construction unit, the prompt presentation unit, the answer information reception unit, the provisional difference information specifying unit, and the similarity calculation unit, each similarity for each of the provisional hint set information is obtained, and based on the similarity, a compatible hint set information identifying unit that identifies compatible hint set information that is the appropriate provisional hint set information; A compatible information registration instruction unit that presents a compatible information registration instruction for updating the knowledge graph so that the compatible hint set information or the compatible search range information indicating the range to search within the knowledge graph to collect the hint information included in the compatible hint set information is used as hint set information or search range information regarding any asset.

9. The system according to claim 8, wherein the compatible hint set information identifying unit identifies the compatible hint set information based on, in addition to the similarity, the number of hint information included in the provisional hint set information or provisional search range information indicating the range to search within the knowledge graph to collect the hint information included in the provisional hint set information.

10. The system according to claim 2, wherein the knowledge graph is for managing the asset, the symptoms that the asset may have, the failure modes that the asset may assume, managing the relationships between the assets, and managing the relationships between the asset, the symptoms, and the failure modes, wherein the hint information is information indicating an attribute that the asset has, a relationship between the assets, a relationship between the asset and the symptoms, or a relationship between the symptoms and the failure modes, wherein the document that triggers the update of the knowledge graph is a preservation document.

11. The system according to claim 1, wherein the system further includes a supplementary document information generation unit that generates supplementary document information indicating a supplementary document that supplements the content of the document based on the document information and the knowledge graph information. A supplementary document display control unit that controls to display the supplementary document based on the supplementary document information; A system having a supplementary document validity information reception unit that receives supplementary document validity information which is input information indicating the validity of the supplementary document.

12. The system according to claim 1, wherein the system further includes a uniqueness determination unit that determines whether or not the content of the update of the knowledge graph indicated by the document is uniquely specified based on the document information and the knowledge graph information; an option display control unit that controls to display options that could be the content of the update of the knowledge graph indicated by the document when the content of the update of the knowledge graph indicated by the document is not uniquely specified; A system having a selection information reception unit that receives selection information indicating which of the options is selected.

13. The system according to claim 1, wherein the system further includes a knowledge graph editing information reception unit that receives knowledge graph editing information based on an input for editing the knowledge graph information; A system having a knowledge graph information editing instruction unit that presents a knowledge graph information editing instruction for reflecting the content of the editing of the knowledge graph information indicated by the knowledge graph editing information in the knowledge graph to the knowledge graph system.

14. A method executed by a system, comprising: a document information reception step of receiving document information which is information included in a document that triggers an update of a knowledge graph; a knowledge graph information reception step of receiving knowledge graph information which is information included in the knowledge graph from a knowledge graph system that manages the knowledge graph; a prompt construction step of constructing a prompt based on the document information and the knowledge graph information, wherein the prompt is for requesting the presentation of information used for constructing an update query for updating the knowledge graph; a prompt presentation step of presenting the prompt to a language model system that manages a language model; an answer information reception step of receiving answer information indicating an answer from the language model system to the prompt; an update query construction step of constructing the update query based on the answer information; A method having an update query presentation step of presenting the update query to the knowledge graph system.

15. A program, which causes a system to ​ A document information receiving step of receiving document information, which is information included in a document that triggers an update of the knowledge graph; A knowledge graph information receiving step of receiving knowledge graph information, which is information included in the knowledge graph, from the knowledge graph system that manages the knowledge graph; A prompt construction step of constructing a prompt based on the document information and the knowledge graph information, wherein the prompt is for requesting the presentation of information used for constructing an update query for updating the knowledge graph; A prompt presentation step of presenting the prompt to a language model system that manages a language model; A response information receiving step of receiving response information indicating a response from the language model system to the prompt; An update query construction step of constructing the update query based on the response information; A program for causing an update query presentation step of presenting the update query to the knowledge graph system to be executed.

Images