Information processing system, information processing method, and information processing program

Abstract

This information processing system: receives an interactive answer request pertaining to a manufacturing industry; sets a search query on the basis of the answer request; acquires the search result by searching for a manufacturing industry data source, which is a data source related to the manufacturing industry, on the basis of the search query; inputs, to a large language model, a prompt obtained by integrating the answer request and all or a portion of the search result; and outputs an answer output from the large language model.

Description

Information Processing System, Information Processing Method, and Information Processing Program , ,

[0004] ,

[0003] ,

[0005]

[0001] The present disclosure relates to an information processing system, an information processing method, and an information processing program.

[0002] Japanese Patent No. 7413445 (hereinafter simply referred to as Document 1) discloses an operation method of a computer for managing a plurality of design data representing a set of parts. The operation method disclosed in Document 1 includes a step of determining whether or not target design data included in the plurality of design data is related to existing design data already stored, and when the result of the determination is affirmative, a step of obtaining performance information when the performance information linked to the existing design data is stored, and a step of transmitting performance display information for displaying the performance information to a terminal. The determination includes a determination as to whether or not the existing design data represents a part having the same attributes as the part represented by the target design data, and the attributes are material, surface treatment, processing method, or a combination of two or more of these.

[0003] By the way, consider a scenario where a user inputs a question regarding manufacturing to a large language model. In such a case, the answer output from the large language model may not be the answer the user desires. Specifically, the answer output from the large language model may be an answer that does not take into account the context regarding manufacturing, or an answer that contains errors.

[0004] The present disclosure has been made in view of the above circumstances, and aims to obtain an accurate answer when attempting to obtain an answer to a question regarding manufacturing from a large language model.

[0005] To achieve the above objective, a first aspect of this disclosure is an information processing system equipped with a processor, wherein the processor receives an interactive response request relating to the manufacturing industry, sets a search query based on the response request, obtains search results by searching a manufacturing data source which is a data source relating to the manufacturing industry based on the search query, inputs a prompt integrating all or part of the response request and the search results into a large-scale language model, and outputs a response output from the large-scale language model.

[0006] Furthermore, a second aspect of this disclosure is an information processing method in which a computer performs processing that includes receiving a dialogue-style response request relating to the manufacturing industry, setting a search query based on the response request, obtaining search results by searching a manufacturing data source which is a data source related to the manufacturing industry based on the search query, inputting a prompt that integrates all or part of the response request and the search results into a large-scale language model, and outputting a response output from the large-scale language model.

[0007] Furthermore, a third aspect of this disclosure is an information processing program for causing a computer to perform the following processes: receiving a dialogue-style response request relating to the manufacturing industry; setting a search query based on the response request; obtaining search results by searching a manufacturing data source, which is a data source related to the manufacturing industry, based on the search query; inputting a prompt integrating all or part of the response request and the search results into a large-scale language model; and outputting the response output from the large-scale language model.

[0008] According to this disclosure, when attempting to obtain answers to questions related to the manufacturing industry from a large-scale language model, it is possible to obtain highly accurate answers.

[0009] This figure shows an example of the schematic configuration of the information processing system of this embodiment. It is a schematic block diagram of the computers that function as each device in the information processing system. This figure illustrates the processing performed by the information processing system of this embodiment.

[0010] The embodiments will be described in detail below with reference to the drawings.

[0011] <System Configuration of the Information Processing System> Figure 1 is a block diagram of the information processing system 10 of this embodiment. As shown in Figure 1, the information processing system 10 of this embodiment includes an information processing device 16 and a plurality of user terminals 18A, 18B, 18C. In the following description, unless a specific terminal is being referred to, one user terminal will be referred to as user terminal 18. The information processing device 16 and the user terminals 18 are connected to each other via a network 19, such as the Internet.

[0012] The information processing system 10 of this embodiment receives an interactive question about the manufacturing industry from a user. Based on the question, the information processing system 10 sets a search query. Based on the search query, the information processing system 10 retrieves search results by searching a manufacturing data source, which is a data source related to the manufacturing industry. The information processing system 10 inputs a prompt that integrates all or part of the question and search results into a large-scale language model and outputs the answer output from the large-scale language model. This enables the output of highly accurate answers to questions about the manufacturing industry. A detailed explanation follows below.

[0013] (Information Processing Device 16) The information processing device 16 is a server that responds to information transmitted from the user terminal 18. As shown in Figure 1, the information processing device 16 functionally comprises a manufacturing data source 20, a large-scale language model storage unit 22, a reception unit 24, a processing unit 26, and an output unit 28.

[0014] The manufacturing data source 20 is a data source related to the manufacturing industry. The manufacturing data source 20 contains product information, which is information about products. Product information includes, for example, product drawings or CAD drawings. Product information may also include document information related to products. For example, document information related to products includes various reports such as product defect reports, instruction documents regarding the product manufacturing process, documents regarding product manufacturing records, documents regarding product raw materials, or documents regarding product orders and deliveries. Furthermore, the individual data stored in the manufacturing data source 20 may be unstructured data such as images, audio, or plain text, or it may be structured data in which attributes and values ​​are linked and can take the form of a table.

[0015] For example, the manufacturing data source 20 may be a data source in which multiple different product information is associated or linked to each other. However, the manufacturing data source 20 does not necessarily have to be associated or linked to multiple different product information.

[0016] For example, the manufacturing data source 20 may be an SQL (Structured Query Language) based data source. If the manufacturing data source 20 is an SQL-based data source, for example, multiple different product information items in the manufacturing data source 20 are linked by a relationship key. For example, if a drawing file, which is product information, and a CAD file, which is product information, have the same drawing number, the drawing file and the CAD file are linked by the drawing number.

[0017] Alternatively, for example, the manufacturing data source 20 may be a NoSQL-based data source. If the manufacturing data source 20 is a NoSQL-based data source, for example, the relationships between multiple different product information in the manufacturing data source 20 are represented by a graph database. When the relationships between multiple different product information are represented by a graph database, for example, each of product information A and product information B is represented as a node on the graph, and the relationship between product information A and product information B is represented by an edge. For example, a score representing the degree of relationship between product information A and product information B is assigned to the edge. Also, for example, attribute information representing the relationship between product information A and product information B is assigned to the edge. For example, if product information A is a drawing of product X and product information B is a defect report of product X, attribute information representing "drawing-defect report" is assigned to the edge connecting the node corresponding to product information A and the node corresponding to product information B.

[0018] Furthermore, for example, the manufacturing data source 20 may be a file server. A file server is a server that outputs files in response to user requests. In this case, for example, the manufacturing data source 20 may be organized into folders, each containing multiple pieces of product information. For example, the product information stored in one folder may be organized to include product information related to a particular customer. Alternatively, the product information stored in one folder may be organized to include different types of product information related to a particular product (e.g., drawings, CAD, and defect reports). In this case, the multiple pieces of product information stored in a single folder are associated or linked to each other.

[0019] The manufacturing data source 20 can take the form of various data sources as described above. Alternatively, the manufacturing data source 20 may be a data source that combines various data sources as described above. A search is performed on the manufacturing data source 20 by the processing unit 26, which will be described later.

[0020] The large-scale language model memory unit 22 stores known large-scale language models. These large-scale language models are pre-trained using known machine learning techniques.

[0021] The reception unit 24 receives dialogue-style requests for answers related to manufacturing. Specifically, the user inputs dialogue-style requests for answers related to manufacturing into the user terminal 18 by operating the user terminal 18. For example, the user inputs a dialogue-style question request to the user terminal 18, such as, "Please tell me the points to be aware of when manufacturing a product called XX made of stainless steel." The user terminal 18 transmits the dialogue-style request for answers related to manufacturing to the information processing device 16. The reception unit 24 then receives the dialogue-style request for answers related to manufacturing transmitted from the user terminal 18.

[0022] Furthermore, users can also send product information related to their response request (e.g., drawing files or CAD) to the information processing device 16 by operating their user terminal 18. For example, a user can upload a drawing file or CAD and request a response to a question such as "Please tell me the manufacturing precautions for this drawing." In this case, the receiving unit 24 further receives the drawing file or CAD input by the user.

[0023] The processing unit 26 sets a search query based on the response request received by the reception unit 24. Furthermore, if the reception unit 24 receives a drawing file or CAD data, the processing unit 26 sets a search query based on both the response request and the drawing file or CAD data.

[0024] For example, when setting a search query, the processing unit 26 uses a large-scale language model to determine what kind of search should be performed. For instance, the processing unit 26 inputs the answer request entered by the user and the inquiry about what kind of search should be performed into the large-scale language model. Let's assume that the large-scale language model outputs a result indicating that it should use one of three search processes, which will be described later: keyword search, vector search, and filtering.

[0025] In this case, the processing unit 26 sets the search keywords by performing morphological analysis on the response request as a search query for keyword search, and also sets the final search query by setting an extended query. The processing unit 26 also sets the input drawing as the search query for vector search. Furthermore, the processing unit 26 limits the information to, for example, information where the product category is XX and the material is stainless steel, for filtering purposes. The above-described setting and limiting of the search query will be used in the search process described later.

[0026] Next, the processing unit 26 obtains search results by searching the manufacturing data source 20 based on the search query (and restrictions). For example, the processing unit 26 obtains search results by performing a known keyword search on the manufacturing data source 20.

[0027] Alternatively, for example, the processing unit 26 obtains search results by performing a known vector search on the manufacturing data source 20. In a vector search, the feature quantities of product information are represented by vectors, and a similarity search is performed by considering the distance between vectors as similarity. For example, if the product information is a drawing or CAD, the shape of the product drawn in the drawing or CAD is used as a feature quantity. It is also possible to vectorize text in a vector search. Therefore, it is possible to perform a vector search even if the product information is document information.

[0028] Alternatively, for example, the processing unit 26 obtains search results by performing filtering (narrowing down) on the manufacturing data source 20 according to the attribute information of the product information. The product information constituting the manufacturing data source 20 is pre-assigned attribute information such as the material of the product or the product category. Therefore, the processing unit 26 obtains search results according to the limitations set above by performing filtering on the manufacturing data source 20. For this reason, the search in this embodiment is a concept that includes the above filtering in addition to general search processing. For this reason, the search query when performing filtering corresponds to the above limitations (for example, limiting to information where the product category is XX and the material is stainless steel).

[0029] The processing unit 26 obtains search results corresponding to the search query by combining keyword search, vector search, and filtering on the manufacturing data source 20. The processing unit 26 may perform the search using at least one of keyword search, vector search, and filtering. The processing unit 26 may also perform a search different from keyword search, vector search, and filtering.

[0030] Furthermore, the search results in this embodiment may include not only search results based on the search query, but also product information associated with the search results. Specifically, if the above-mentioned manufacturing data source 20 is an SQL-based data source, multiple different product information items may be linked together by a relation key. Therefore, if product information A is obtained as a search result based on the search query, other product information B that is linked to product information A will also be obtained as a search result. A threshold for determining whether a certain product information item is linked to other product information items is set in advance. For example, if a certain product information item is linked to other product information items at a primary level, it is determined that the two product information items are linked. On the other hand, if a certain product information item is linked to other product information items at a secondary level or higher, it is determined that the two product information items are not linked.

[0031] Furthermore, the search results of this embodiment may include product information that is related to the search results. Specifically, if the manufacturing data source 20 is a NoSQL-based data source, multiple different product information items may be linked together by graph data. In such cases, if product information A is obtained as a search result based on a search query, other product information B that is associated with product information A will also be obtained as a search result. A threshold for determining whether a certain product information item is related to another product information item is set in advance. For example, if the relationship between a certain product information item and another product information item is primary, it is determined that the two product information items are related. On the other hand, if the relationship between a certain product information item and another product information item is secondary or higher, it is determined that the two product information items are not related.

[0032] Next, the processing unit 26 generates a prompt that integrates the response request and all or part of the search results. For example, the processing unit 26 generates a prompt that includes the following system prompts. Note that it is not mandatory to use all of the following system prompts. For example, system prompts such as "The response should include product information directly linked to the search results" and "The response should include product information directly related to the search results in the graph database" do not need to be used.

[0033] - To answer user-submitted questions; - To use search results as material for the answer; - To include search results in addition to the generated answer; - To include product information directly linked to the search results in the answer; - To include product information directly related to the search results in the answer from the graph database.

[0034] Then, the processing unit 26 inputs the generated prompt to the large-scale language model and obtains the response output from the large-scale language model.

[0035] The output unit 28 outputs the response obtained by the processing unit 26 to the user terminal 18. The output unit 28 may also output all or part of the search results used to generate the response to the user terminal 18 along with the response. The output unit 28 may also output other product information associated with all or part of the search results used to generate the response to the user terminal 18 along with the response. The output unit 28 may also output other product information associated with all or part of the search results used to generate the response to the user terminal 18.

[0036] (User terminal 18) The user terminal 18 is a terminal operated by the user. Specifically, the user exchanges information with the information processing device 16 by operating the user terminal 18.

[0037] Each of the information processing device 16 and user terminal 18 of the information processing system 10 can be implemented, for example, by a computer 70 as shown in Figure 2. The computer 70 includes a CPU 71, a memory 72 as a temporary storage area, and a non-volatile storage unit 73. The computer 70 also includes an input / output interface (I / F) 74 to which input / output devices (not shown) are connected, and a read / write (R / W) unit 75 that controls the reading and writing of data to and from the recording medium. The computer 70 also includes a network I / F 76 that connects to a network such as the Internet. The CPU 71, memory 72, storage unit 73, input / output I / F 74, R / W unit 75, and network I / F 76 are connected to each other via a bus 77.

[0038] The storage unit 73 can be implemented using a Hard Disk Drive (HDD), Solid State Drive (SSD), flash memory, etc. The storage unit 73, as a storage medium, stores a program for making the computer 70 function. The CPU 71 reads the program from the storage unit 73, loads it into memory 72, and sequentially executes the processes contained in the program.

[0039] <Operation of Information Processing System 10> Next, the operation of the information processing system 10 of this embodiment will be described.

[0040] First, the user operates their user terminal 18 to send an answer request to the information processing device 16. When the reception unit 24 of the information processing device 16 receives the answer request, the information processing device 16 executes the information processing routine shown in FIG. 3.

[0041] In step S100, the reception unit 24 receives the answer request.

[0042] In step S102, the processing unit 26 sets a search query based on the answer request received in step S100.

[0043] In step S104, the processing unit 26 obtains a search result by executing a search on the manufacturing data source 20 based on the search query set in step S102.

[0044] In step S106, the processing unit 26 generates a prompt that integrates all or part of the answer request received in step S100 and the search result obtained in step S104. For example, the processing unit 26 may include the top 10 items in the search result in the prompt. Alternatively, the processing unit 26 may include in the prompt content such as "Consider only the top 10 items in the search result".

[0045] In step S108, the processing unit 26 inputs the prompt obtained in step S106 into the large language model stored in the large language model storage unit 22. Then, in step S108, the processing unit 26 obtains the answer output from the large language model.

[0046] In step S110, the output unit 28 outputs the answer obtained in step S108 to the user terminal 18.

[0047] The answer output from the output unit 28 is transmitted to the user terminal 18. The user checks the answer displayed on the display unit (not shown) of their user terminal 18.

[0048] As described above, the information processing apparatus according to the present embodiment receives an interactive response request related to the manufacturing industry. The information processing apparatus sets a search query based on the response request. The information processing apparatus obtains a search result by searching a manufacturing data source, which is a data source related to the manufacturing industry, based on the search query. The information processing apparatus inputs a prompt integrating the response request and all or part of the search result into a large language model, and outputs the response output from the large language model. Thereby, when trying to obtain a response to a question related to the manufacturing industry from the large language model, an accurate response can be obtained.

[0049] In addition, the present embodiment can execute processing for a response request for a question in an interactive format. Conventionally, there has been no system that accurately returns a response to a response request for an interactive question related to the manufacturing industry. The first reason for this is the problem that a keyword search system cannot process a response request for an interactive question. The second reason is that while generative AI such as a large language model can handle a response request for an interactive question, there is a problem that it cannot output an accurate response based on the context of the manufacturing industry or a specific organization.

[0050] In contrast, in the present embodiment, instead of directly inputting the interactive response request related to the manufacturing industry into the large language model, the response request is input into the large language model together with the search result obtained by searching the manufacturing data source. At this time, in the present embodiment, a prompt integrating the response request and the search result is input into the large language model. Thereby, an accurate response based on the context of the manufacturing industry can be obtained.

[0051] Furthermore, in this embodiment, even when the user cannot think of appropriate search terms, they can obtain accurate answers by inputting conversational questions into the system. Users can also upload product information and submit questions together with the product information of the subject of the question, making it possible to input questions that take into account content that is difficult to put into words (especially images) into the system. In addition, the information processing device 16 outputs what data is used as the basis for the answer and other information that may be related to the answer, making it easier to verify the basis of the answer and conduct further investigations, thus realizing collaboration between people and systems.

[0052] This disclosure is not limited to the embodiments described above, and various modifications and applications are possible without departing from the gist of this disclosure.

[0053] Furthermore, in the above embodiment, each process that the CPU reads and executes software (programs) may be executed by various processors other than the CPU. Examples of such processors include PLDs (Programmable Logic Devices) whose circuit configuration can be changed after manufacturing, such as FPGAs (Field-Programmable Gate Arrays), and dedicated electrical circuits that have a circuit configuration specifically designed to execute a particular process, such as ASICs (Application Specific Integrated Circuits). Each process may be executed by one of these various processors, or by a combination of two or more processors of the same or different types (for example, multiple FPGAs, and a combination of a CPU and an FPGA). More specifically, the hardware structure of these various processors is an electrical circuit that combines circuit elements such as semiconductor elements.

[0054] Furthermore, although the above embodiment describes a configuration in which each program is pre-stored (installed) in a storage device, the invention is not limited to this. The program may be provided in a form stored on a storage medium such as a CD-ROM, DVD-ROM, Blu-ray disc, or USB memory. Alternatively, the program may be provided in a form that is downloaded from an external device via a network.

[0055] (Note) The following is a note regarding the nature of this disclosure.

[0056] (Note 1) An information processing system equipped with a processor, wherein the processor receives an interactive response request relating to the manufacturing industry, sets a search query based on the response request, obtains search results by searching a manufacturing data source which is a data source related to the manufacturing industry based on the search query, inputs a prompt integrating all or part of the response request and the search results to a large-scale language model, and outputs a response output from the large-scale language model. (Note 2) The information processing system according to Note 1, wherein the manufacturing data source includes product information which is information relating to a product. (Note 3) The information processing system according to Note 2, wherein the manufacturing data source includes drawings or CAD of the product as the product information. (Note 4) The information processing system according to Note 2, wherein the manufacturing data source is a data source in which multiple different product information is associated or linked to each other. (Note 5) The information processing system according to Note 4, wherein the multiple different product information are linked to each other by a relation key. (Note 6) The information processing system according to Note 4, wherein multiple different product information items are linked to each other by graph data. (Note 7) The information processing system according to Note 3, wherein the processor further accepts a drawing file or CAD input by a user and sets the search query based on the response request and the drawing file or CAD. (Note 8) The information processing system according to Note 4, wherein the search results include at least one of the other product information items linked to the searched product information and the other product information items linked to the searched product information. (Note 9) The information processing system according to any one of Notes 1 to 8, wherein the processor outputs all or part of the search results along with the response. (Note 10) The information processing system according to Note 5, wherein the processor outputs all or part of the search results including at least one of the multiple different product information items along with the response.(Note 11) An information processing method in which a computer performs the following steps: receiving an interactive response request concerning the manufacturing industry; setting a search query based on the response request; obtaining search results by searching a manufacturing data source, which is a data source related to the manufacturing industry, based on the search query; inputting a prompt integrating all or part of the response request and the search results into a large-scale language model; and outputting the response output from the large-scale language model. (Note 12) An information processing program for causing a computer to perform the following steps: receiving an interactive response request concerning the manufacturing industry; setting a search query based on the response request; obtaining search results by searching a manufacturing data source, which is a data source related to the manufacturing industry, based on the search query; inputting a prompt integrating all or part of the response request and the search results into a large-scale language model; and outputting the response output from the large-scale language model.

[0057] The disclosure of Japanese Patent Application No. 2024-229554, filed on 26 December 2024, is incorporated herein by reference in its entirety. All documents, patent applications, and technical standards described herein are incorporated herein by reference to the same extent as if each individual document, patent application, and technical standard were specifically and individually noted to be incorporated by reference.

Claims

1. An information processing system comprising a processor, wherein the processor receives an interactive response request relating to the manufacturing industry, sets a search query based on the response request, obtains search results by searching a manufacturing data source which is a data source relating to the manufacturing industry based on the search query, inputs a prompt integrating all or part of the response request and the search results to a large-scale language model, and outputs a response output from the large-scale language model.

2. The information processing system according to claim 1, wherein the manufacturing data source includes product information, which is information relating to a product.

3. The information processing system according to claim 2, wherein the manufacturing data source includes, as product information, drawings or CAD of the product.

4. The information processing system according to claim 2, wherein the manufacturing data source is a data source in which multiple different product information is associated or linked with each other.

5. The information processing system according to claim 4, wherein multiple different product information items are linked together by a relation key.

6. The information processing system according to claim 4, wherein multiple different product information items are represented by graph data.

7. The information processing system according to claim 3, wherein the processor further receives a drawing file or CAD input by a user, and sets the search query based on the response request and the drawing file or CAD.

8. The information processing system according to claim 4, wherein the search results include at least one of the other product information linked to the searched product information and the other product information associated with the searched product information.

9. The information processing system according to claim 1 or 2, wherein the processor outputs all or part of the search results together with the answer.

10. The information processing system according to claim 5, wherein the processor outputs all or part of the search results, which include at least one of the multiple different product information items, along with the answer.

11. An information processing method in which a computer performs the following steps: receiving an interactive response request concerning the manufacturing industry; setting a search query based on the response request; obtaining search results by searching a manufacturing data source, which is a data source related to the manufacturing industry, based on the search query; inputting a prompt integrating all or part of the response request and the search results into a large-scale language model; and outputting a response output from the large-scale language model.

12. An information processing program for causing a computer to perform the following processes: receiving an interactive response request related to the manufacturing industry; setting a search query based on the response request; obtaining search results by searching a manufacturing data source, which is a data source related to the manufacturing industry, based on the search query; inputting a prompt integrating all or part of the response request and the search results into a large-scale language model; and outputting the response output from the large-scale language model.

Images