Information processing system

Abstract

In information processing systems that utilize search extension generation, the aim is to improve search accuracy and obtain useful answers. [Solution] The information processing system (10) creates a response by inputting the search results from the search extension database (11) into the LLM (20). The search extension database comprises an aggregate database (12) and multiple individual databases (13) each storing a portion of the information stored in the aggregate database. When the information processing system creates a response based on a search using the aggregate database, it adds the search results obtained from the aggregate database to the individual database corresponding to the attributes of the user who searched the aggregate database.

Description

Technical Field

[0001] The present invention relates to an information processing system including retrieval augmented generation.

Background Art

[0002] Information processing using large language models (LLMs) has become widespread. However, it can be difficult to generate text that reflects in-house data by simply using an LLM. For this reason, retrieval augmented generation (RAG) has attracted attention. For example, an information processing apparatus has been proposed that selects a corresponding database based on the content of a question, generates a prompt based on the selected database and the question, and generates an answer by inputting the prompt into a language model (for example, Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In the conventional technology, it is difficult to construct a database system so that a database for retrieval augmented generation is appropriately selected. For this reason, accurate retrieval may not be performed, and useful answers may not be obtained for questions.

[0005] One aspect of the present invention aims to improve retrieval accuracy and obtain useful answers in an information processing system using retrieval augmented generation.

Means for Solving the Problems

[0006] An information processing system according to one aspect of the present invention comprises: an aggregate database; a plurality of individual databases, each storing a portion of the information stored in the aggregate database; a user management unit for managing user attributes; a search history database for recording search information representing the results of a search performed in association with the attributes of the search user who searched the aggregate database when the aggregate database is searched; a database update unit for updating at least one of the plurality of individual databases based on the search information recorded in the search history database; and a search control unit for inputting a question sentence from a search user and search results obtained from the aggregate database or an individual database selected by the search user in response to the question sentence into a predetermined language model to obtain an answer sentence. [Effects of the Invention]

[0007] According to the above-described embodiment, in an information processing system that utilizes search extension generation, search accuracy can be improved and useful answers can be obtained. [Brief explanation of the drawing]

[0008] [Figure 1] This figure shows an example of a search extension generation system. [Figure 2] This figure shows an example of an information processing system according to an embodiment of the present invention. [Figure 3] This diagram illustrates an example of how to construct aggregated and individual databases. [Figure 4] This figure shows an example of a user management database. [Figure 5] This figure shows an example of a search history database. [Figure 6] A flowchart illustrating an example of the preparation procedure for an information processing system. [Figure 7] This flowchart shows an example of the processing performed by the information processing system during the operation of the search extension generation system. [Figure 8] This diagram shows an example of a search screen used by a search user. [Figure 9] This figure shows an example of an updated search history database. [Figure 10] This figure shows an example of the process for updating individual databases. [Figure 11] This flowchart shows the variations in processing of the information processing system during the operation of the search extension generation system. [Figure 12] This figure shows an example of the hardware configuration of an information processing system. [Modes for carrying out the invention]

[0009] Figure 1 shows an example of a Search Augmentation Generator (RAG) system. Search Augmentation Generator is a framework that obtains highly accurate answers by combining a search phase and a generation phase. The generation phase is implemented, for example, using a Large-Scale Language Model (LLM). However, since LLMs depend on pre-trained datasets, they may not be able to generate answers based on internal company data or the latest information. Therefore, in Search Augmentation Generator, a search phase is provided before the generation phase in which the LLM generates answers, and a database where internal company data etc. is stored is searched.

[0010] In the example shown in Figure 1, the search application, upon receiving a question from a user, obtains search results by searching the company's internal database. The search application then inputs the user's question and the search results obtained from the internal database into the LLM (Low-Level Memory). This generates an answer to the question in the LLM. The search application then provides the generated answer to the user. Therefore, the user can obtain answers based not only on publicly available information but also on internal company data.

[0011] Figure 2 shows an example of an information processing system according to an embodiment of the present invention. The information processing system 10 according to the embodiment of the present invention constitutes a search expansion generation system by cooperating with the LLM 20. The LLM 20 is a natural language processing system constructed using a vast amount of text data and advanced deep learning techniques, and has functions to understand, predict, and create sentences in various natural languages (e.g., Japanese, English, etc.). Note that the information processing system 10 mainly provides the search phase of search expansion generation, and the LLM 20 mainly provides the generation phase of search expansion generation.

[0012] In this embodiment, the information processing system 10 and the LLM 20 are connected to each other via the network 30. The network 30 is realized, for example, by a public network, but may include a private network. Also, the network 30 may be a wireless network, a wired network, or may include a wireless network and a wired network.

[0013] The search user can input a question sentence into the search expansion generation system using the terminal 40. The question sentence may be a sentence or one or more words. The question sentence input in the terminal 40 is processed by the information processing system 10. At this time, the information processing system 10 can generate an answer sentence using the LLM 20. Then, the information processing system 10 transmits the generated answer sentence to the terminal 40. Thereby, the search user can obtain an answer sentence corresponding to the question sentence.

[0014] As shown in Figure 2, the information processing system 10 includes a search expansion database 11, a user management database 14, a search history database 15, a search control unit 16, and a database update unit 17. The search expansion database 11 includes an aggregated database 12 and a plurality of individual databases 13 (13a to 13n). Note that the information processing system 10 may further include other functions, devices, and circuits not shown in Figure 2.

[0015] The database 11 for search expansion (i.e., the aggregated database 12 and the individual databases 13) is, for example, an in-house database of a company that owns the information processing system 10. In this case, the search user is an employee of the company. However, the database 11 for search expansion is not limited to an in-house database and may be a database constructed outside the company.

[0016] In the aggregated database 12, various documents used within the company and various documents stored within the company are registered. Assume that the various documents are stored in files as electronic data. In the example shown in FIG. 3, documents A to F are registered in the aggregated database 12. Note that the content of each document is preferably converted into vector data and stored in the aggregated database 12. However, in the embodiment shown in FIG. 3, the content of each document is represented by text.

[0017] Each record in the aggregated database 12 is composed of the name of the document, the page number within the document, and the text describing the content of the document. For example, in record 1, the content of page 1 of document A, "The basic technology of ■■ is 〇〇.", is registered, and in record 2, the content of page 1 of document B, "The advanced technology of ●● is ▲▲.", is registered.

[0018] In this embodiment, the individual databases 13 (13a to 13n) are constructed by classifying and registering the information registered in the aggregated database 12 for each predetermined category. That is, a part of the information stored in the aggregated database 12 is stored in each individual database 13. As an example, individual databases 13 are constructed for each job type within the company that owns the information processing system 10. Therefore, different information will be stored in the individual databases 13a to 13n. However, a part of the information stored in each of the individual databases 13a to 13n may be stored redundantly in one or more other individual databases 13a to 13n.

[0019] In the example shown in Figure 3, for example, the information registered in records 1 to 3 of the aggregate database 12 is registered in individual database 13a created for the research and development department, and the information registered in records 4 to 6 of the aggregate database 12 is registered in individual database 13b created for the product planning and marketing department. The administrator of the information processing system 10 may decide, for example, which individual database 13 each document registered in the aggregate database 12 should be registered in. In this case, the administrator can register each document registered in the aggregate database 12 in any one or more individual databases 13. Alternatively, each document registered in the aggregate database 12 may be registered in one or more individual databases 13 by a computer based on the content of the document.

[0020] The user management database 14 manages users who are permitted to access the information processing system 10 and perform information retrieval. In this embodiment, the users to be managed are employees of the company that owns the information processing system 10. The user management database 14 registers the attributes of each user. For example, the job title of each employee (i.e., user) within the company that owns the information processing system 10 is registered in the user management database 14 as an attribute of each user. Note that the user management database 14 is an example of a user management unit that manages the attributes of each user.

[0021] Figure 4 shows an example of a user management database 14. In this embodiment, each record in the user management database 14 consists of a name, ID, password, and job title. The name represents the name of each employee (i.e., user) of the company. The ID is information that uniquely identifies each employee within the company, and may be, for example, an employee number. The password is authentication information that each employee uses to access the information processing system 10. The job title is one of the pieces of information that represent the attributes of each employee, and represents the department to which each employee belongs within the company. For example, record 1 contains an employee whose name is "Fuji Taro", ID is "100", password is "100a", and job title is "Research and Development".

[0022] Furthermore, the information representing the attributes of each employee is not limited to job title; it can be any information that allows the company's employees to be grouped into two or more groups, such as industry or age. However, it is preferable that the attributes used to group the company's employees are the same as the attributes related to each individual database 13 (13a to 13n). In this embodiment, an individual database 13 is constructed for each job title within the company, and employees are grouped according to their job titles within the company.

[0023] The search history database 15 records search information associated with the attributes of the search user who searched the aggregate database 12 when the aggregate database 12 is searched. The search information represents the content of the search process performed and may include, for example, the question entered at the time of the search, the answer created in response to that question, and information identifying the documents referenced during the search. When a search user performs a search using the information processing system 10, they can either select the aggregate database 12 or select their desired individual databases 13a to 13n.

[0024] Figure 5 shows an example of the search history database 15. Each record in the search history database 15 consists of an ID, job title, question, answer, referenced document, and referenced page. The ID and job title represent the ID and job title of the search user who performed the search using the aggregated database 12, respectively. The question and answer represent the question entered by the search user during the search and the answer to that question, respectively. For example, the answer may be a document created by LLM20. The referenced document and referenced page represent the document referenced when creating the answer from the question and the page within that document, respectively.

[0025] For example, in record 1 of the search history database 15 shown in Figure 5, a search user with ID "101" and job title "Research and Development" entered "What are the cutting-edge technologies of ●●?" as the question, and received the answer "The cutting-edge technologies of ●● are ▲▲ and △△." In addition, it is also recorded that pages 1 and 2 of document B were referenced in this search.

[0026] In the example shown in Figure 5, five search histories are registered in the search history database 15, but there is no upper limit to the number of search histories that can be registered. Also, in the example shown in Figure 5, the search information includes the question, answer, reference document, and reference page, but the embodiments of the present invention are not limited to this. For example, the search information recorded in the search history database 15 does not have to include the question, and instead of the answer, it may include search results obtained from the aggregation database 12. However, it is preferable that the search information includes the reference document and reference page that represent the source of the search results.

[0027] The search control unit 16 processes the query entered by the search user. At this time, the search control unit 16 obtains search results using the aggregate database 12 or individual databases 13 as needed. Specifically, when the search user instructs the search user to search the aggregate database 12, the search control unit 16 searches the aggregate database 12 and obtains the search results. Also, when the search user selects a desired individual database 13, the search control unit 16 searches the selected individual database 13 and obtains the search results. Then, the search control unit 16 obtains the answer by inputting the query into the LLM 20. Also, when search results have been obtained from the aggregate database 12 or individual databases 13, the search control unit 16 obtains the answer by inputting the query and search results into the LLM 20. After this, the search control unit 16 provides the answer created by the LLM 20 to the search user.

[0028] The database update unit 17 updates the individual databases 13 (13a to 13n) by referring to the search history database 15. The operation of the database update unit 17 will be described later.

[0029] The configuration of the information processing system 10 shown in Figure 2 is just one example, and the embodiments of the present invention are not limited to this configuration. For example, the information processing system 10 may include an LLM20. Alternatively, the information processing system 10 may not include an LLM20, but may have an interface with an LLM20. In this case, the interface with the LLM20 may be implemented by the search control unit 16.

[0030] Figure 6 is a flowchart showing an example of the preparation procedure for the information processing system 10. This preparation procedure corresponds to the initial setup of the information processing system 10 and is performed before the actual operation of the information processing system 10 begins. However, the processing in S4 may be performed while the information processing system 10 is in operation.

[0031] In S1, the aggregate database 12 is created. The aggregate database 12 is created by registering various documents used within the company and various documents stored within the company, as explained with reference to Figure 3. These documents may be registered manually by the administrator of the information processing system 10, or they may be registered by an application program. The name of each document and the page number within the document are also registered.

[0032] In S2, individual databases 13 are created. As explained with reference to Figure 3, individual databases 13 are created by classifying and registering the information registered in the aggregate database 12 into predetermined categories. In this embodiment, an individual database 13 is created for each job type within the company that owns the information processing system 10. However, it is not necessary to create individual databases 13 that are uniquely associated with attributes for grouping users (i.e., job type). For example, one individual database 13 may be created for multiple job types (research and development and product design). In other words, each individual database 13 only needs to be created in association with the attributes of one or more users. Furthermore, individual databases 13 may be created manually by the administrator of the information processing system 10, or they may be created by an application program.

[0033] In this embodiment, the aggregate database 12 is created first, and then multiple individual databases 13 are created from the aggregate database 12. However, embodiments of the present invention are not limited to this procedure. For example, multiple individual databases 13 may be created first, and then the aggregate database 12 may be created by integrating the contents of the multiple individual databases 13.

[0034] In S3, a user management database 14 is created. As explained with reference to Figure 4, the user management database 14 is created by registering the ID, password, job title, etc., for each employee of the company that owns the information processing system 10. The user management database 14 may be created manually by the administrator of the information processing system 10, or it may be created by an application program.

[0035] In S4, a search history database 15 is created. As explained with reference to Figure 5, the search history database 15 is created by registering information related to searches when a search user accesses the aggregate database 12. Specifically, it registers the search user's attributes (in this case, the search user's occupation), the search query entered by the search user, the answer to that query, and information identifying the documents referenced when creating the answer from the query. The search history database 15 is updated each time a search user accesses the aggregate database 12 during the actual operation of the information processing system 10. That is, each time a search user accesses the aggregate database 12, search-related information is added to the search history database 15.

[0036] Figure 7 is a flowchart showing an example of the processing performed by the information processing system 10 during the operation of the search extension generation system. It is assumed that when the processing in this flowchart is executed, the aggregate database 12, individual databases 13, user management database 14, and search history database 15 have been created according to the procedure shown in Figure 6.

[0037] In S11, the information processing system 10 awaits a search request from a search user. In this embodiment, the search user can input a search request to the information processing system 10 using a search screen provided by the information processing system 10 or the search control unit 16. In this case, the search control unit 16 receives instructions or information entered by the search user on the search screen. However, the search screen may be provided by an application program implemented on each terminal.

[0038] Figure 8 shows an example of a search screen used by a search user. The search screen is displayed on the display device of terminal 40, for example, when a search user launches a search application on terminal 40 as shown in Figure 2. The search screen also includes a user information input area, a database selection area, a question input area, and an answer display area.

[0039] The user information input area is provided for entering information related to the search user. Specifically, when a search user starts a search, they enter their name, ID, and occupation in the user information input area. Although omitted in Figure 8, the search user also enters their password using the user information input area.

[0040] The database selection area is used to select either the aggregate database 12 or individual databases 13a to 13n. That is, the search user can select either the aggregate database 12 or an individual database 13 corresponding to the desired job category. In this embodiment, individual databases 13(13a) related to research and development, individual databases 13(13b) related to product planning and marketing, and individual databases 13(13c) related to product design have been created.

[0041] Here, the aggregate database 12 stores all documents designated by the administrator of the information processing system 10. Therefore, in cases where the possibility of missing documents in a search is to be minimized, it is preferable for the search user to select the aggregate database 12. In contrast, the individual databases 13a to 13n each store documents related to specific job types. Therefore, when accessing an individual database 13 corresponding to a desired job type, the search scope is narrowed (i.e., the number of documents to search is small), resulting in higher search accuracy. Therefore, in cases where high search accuracy is required, it is preferable for the search user to select an appropriate individual database 13. Note that in cases where the search extension generation system is not used, the search user does not need to select either the aggregate database 12 or an individual database 13.

[0042] The question input area is used to enter a question. That is, when a search user performs a search, they enter a question in the question input area. The information processing system 10 then generates an answer to the question. The answer generated by the information processing system 10 is then displayed in the answer display area of ​​the search screen.

[0043] In the example shown in Figure 8, a search user whose name is "Fujisaburo," ID is "103," and occupation is "product design" selects the aggregate database 12 and enters the question "Where are the design documents for ■■?". The information processing system 10 then generates the answer "The design documents for ■■ are ○○."

[0044] Let's return to the explanation of the flowchart in Figure 7. When the search control unit 16 receives a search request from the search user, it executes the requested search process in S12. For example, when the search user selects an aggregate database, the search control unit 16 accesses the aggregate database 12 according to the query entered by the search user and obtains the search results. Next, the search control unit 16 inputs the query entered by the search user and the search results obtained from the aggregate database 12 into the LLM 20 to obtain an answer. The search control unit 16 then sends this answer to the terminal 40. This answer is then displayed on the search screen shown in Figure 8.

[0045] When a user selects an individual database related to research and development, the search control unit 16 accesses the individual database 13a in accordance with the user's input and retrieves search results. Subsequently, the search control unit 16 inputs the user's input and the search results obtained from the individual database 13a into the LLM 20 to obtain an answer. The search control unit 16 then transmits this answer to the terminal 40. This answer is then displayed on the search screen shown in Figure 8.

[0046] After this, the information processing system 10 executes a process to update the search extension database 11. That is, processes S13 to S16 are executed.

[0047] In S13, the database update unit 17 determines whether the aggregate database 12 has been selected by the search user in the search request received by the information processing system 10 in S11. If the aggregate database 12 has been selected by the search user, the information processing system 10 proceeds to S14 to execute the database update process. On the other hand, if any of the individual databases 13 have been selected by the search user, or if neither the search extension database 11 (i.e., the aggregate database 12, nor the individual databases 13) has been selected, the database update process is not executed, and the information processing system 10 returns to S11.

[0048] In S14, the database update unit 17 updates the search history database 15. Specifically, the database update unit 17 adds the contents of the search process in S12 to the search history database 15.

[0049] For example, suppose a search user requests the search shown in Figure 8. That is, a search user whose name is "Fuji Saburo", ID is "103", and occupation is "product design", selects the aggregate database 12 and enters the question "Where are the design documents for ■■?". At this time, assume that the search history database 15 has the five records shown in Figure 5 registered.

[0050] The search control unit 16 searches the aggregate database 12 using the question entered by the search user. Here, it is assumed that the search result is obtained from the first page of document E stored in the aggregate database 12. Then, the search control unit 16 inputs the entered question and the search result obtained from the aggregate database 12 into the LLM 20. This yields the answer "The design document for ■■ is ○○." In this case, as shown in Figure 9, the database update unit 17 adds a record representing the content of the search process to the search history database 15. The added record indicates that when a user in the product design department entered the question "Where is the design document for ■■?", the search result was obtained from the first page of document E, and the answer "The design document for ■■ is ○○." was created by the LLM 20.

[0051] In S15-S16, the database update unit 17 updates the individual databases 13 based on the search history database 15. That is, the database update unit 17 updates the individual databases 13 based on the search results using the aggregate database 12. At this time, at least one of the multiple individual databases 13 is updated.

[0052] Specifically, when the aggregate database 12 is searched in response to a search request from a search user with certain attributes, the database update unit 17 saves the search results obtained from the aggregate database 12 to the individual database 13 corresponding to the attributes of that search user. Furthermore, if the search results obtained from the aggregate database 12 are stored in other individual databases 13, it is preferable for the database update unit 17 to delete those search results from those individual databases 13. If the search results obtained from the aggregate database 12 are stored in multiple other individual databases 13, the database update unit 17 may delete the search results from each of those multiple individual databases 13, or it may delete the search results from any one of those multiple individual databases 13.

[0053] For example, as shown in Figure 10, suppose that three records are registered in the individual product planning and marketing database 13b, and three records are registered in the individual product design database 13c. Also, the search shown in Figure 8 is performed, and the search results are obtained from the aggregate database 12. Then, as shown in Figure 9, search information representing the results of the search shown in Figure 8 is added to the sixth record of the search history database 15.

[0054] In this case, the database update unit 17 updates the individual database 13 based on the contents of the sixth record in the search history database 15. Here, this record contains search information indicating that a response "The design document for ■■ is ○○." was created in response to the question "Where is the design document for ■■?" from a search user belonging to "product design". Therefore, in S15, the database update unit 17 adds the contents of the sixth record in the search history database 15 to the individual database 13 corresponding to "product design" (i.e., the individual product design database 13c). In Figure 10, the third record has been added to the individual product design database 13c.

[0055] Furthermore, the database update unit 17 checks for the existence of an individual database 13 that contains the same information as that added to the individual product design database 13c. In this embodiment, the third record of the individual product planning and marketing database 13b contains the same information as that added to the individual product design database 13c. Therefore, in S16, the database update unit 17 deletes the third record of the individual product planning and marketing database 13b. In other words, the database update unit 17 deletes the search results obtained by the search control unit 16 from the search extension database 11 from individual databases 13 other than the individual database 13 (in this case, the individual product planning and marketing database 13b) to which the search results were added (in this case, the individual product design database 13c). In other words, the database update unit 17 deletes the search results obtained by the search control unit 16 from the search extension database 11 from individual databases 13 (in this case, the individual product planning and marketing database 13b) that do not correspond to the attributes of the search user.

[0056] As described above, updating the individual database 13 is expected to improve search accuracy. Specifically, the individual product design database 13c will contain the search results of search users belonging to "product design." Here, search users belonging to "product design" are highly likely to select the individual product design database 13c when performing a search. Therefore, in subsequent searches, search users belonging to "product design" can access databases containing useful information (i.e., the individual product design database 13c). As a result, search accuracy will improve.

[0057] Furthermore, search users belonging to "product planning and marketing" are highly likely to select the individual product planning and marketing database 13b when performing a search. In this individual product planning and marketing database 13b, search results from search users belonging to "product design" have been removed. In other words, information related to searches by users belonging to other departments has been removed from the individual product planning and marketing database 13b. Therefore, search users belonging to "product planning and marketing" can access a database (i.e., the individual product planning and marketing database 13b) from which information not useful in the product planning and marketing field has been removed in subsequent searches. As a result, search accuracy is improved.

[0058] Furthermore, according to the embodiment of the present invention, each time a search user selects the aggregated database 12 and performs a search, the individual database 13 is automatically updated so that highly relevant information is added and less relevant information is deleted. This improves the efficiency of maintenance for the search extension database 11 and reduces the burden on the administrator of the information processing system 10.

[0059] The database update unit 17 may execute processes S15-S16 immediately after the search process in S12 is completed, or immediately after the update process in S14. Alternatively, the database update unit 17 may execute processes S15-S16 at a timing specified by a predetermined rule. For example, the database update unit 17 may periodically refer to the search history database 15 (once a day, once a week, etc.) and execute processes S15-S16. Furthermore, the database update unit 17 does not have to execute process S16.

[0060] In the above-described embodiment, the search results obtained from the aggregated database 12 are stored in the individual database 13 corresponding to the job title of the search user. However, embodiments of the present invention are not limited to this procedure. For example, if a search user selects the aggregated database 12 and requests a search, and the response text obtained from the information processing system 10 is not useful to the search user, it is undesirable to store the search results obtained from the aggregated database 12 in the individual database 13 corresponding to the job title of that search user.

[0061] Therefore, the information processing system 10 may ask the search user whether the response text it created in response to the search request was useful or not. In this case, the information processing system 10 updates the individual database 13 only if the response text was useful to the search user.

[0062] In the example shown in Figure 11, the search control unit 16 inquires with the search user in S21 whether the response text generated in response to the search request was useful or not. This inquiry is made, for example, using the search screen shown in Figure 8. The search user then inputs information on the search screen indicating whether or not the response text was useful.

[0063] When the database update unit 17 receives information indicating that the response was useful, it executes the processes S15-S16. On the other hand, when it receives information indicating that the response was not useful, the database update unit 17 skips the processes S15-S16. According to this procedure, only useful information is registered in the individual database 13, so an improvement in search accuracy can be expected.

[0064] <Hardware Configuration> Figure 12 shows an example of the hardware configuration of the information processing system 10. The information processing system 10 is implemented by a computer 100 which includes a processor 101, memory 102, storage device 103, input / output device 104, recording medium reader 105, and communication interface 106.

[0065] The processor 101 can execute various programs stored in the storage device 103. The processor 101 executes an information processing program according to an embodiment of the present invention, thereby providing the functions of the search control unit 16 and the database update unit 17 shown in Figure 1. The information processing program according to an embodiment of the present invention includes, for example, program code describing the procedure shown in Figure 7 or Figure 11. Memory 102 is used as a workspace for the processor 101. The storage device 103 stores the information processing program and other programs. The search extension database 11 (aggregate database 12 and individual databases 13), the user management database 14, and the search history database 15 are, for example, provided within the storage device 103.

[0066] The input / output device 104 may include input devices such as a keyboard, mouse, touch panel, and microphone. The input / output device 104 may also include output devices such as a display device and speaker. The recording medium reader 105 can acquire data and information recorded on the recording medium 110. The recording medium 110 is a removable recording medium that can be attached to and detached from the computer 100. The recording medium 110 can be implemented, for example, by semiconductor memory, a medium from which signals can be read by optical means, or a medium from which signals can be read by magnetic means. The information processing program may be provided from the recording medium 110 to the computer 100. The communication interface 106 provides the function of connecting to a network. When the information processing program is stored on the program server 120, the computer 100 may acquire the information processing program from the program server 120. [Explanation of symbols]

[0067] 10 Information Processing Systems 11 Database for Search Extension 12 Aggregation Database 13 (13a~13n) Individual Databases 14. User Management Database 15 Search History Database 16 Search Control Unit 17 Database Update Section 20 LLM 30 Networks 40 devices

Claims

1. Aggregated database and Multiple individual databases, each storing a portion of the information stored in the aforementioned aggregate database, The user management department manages user attributes, A search history database records search information representing the results of the search performed when the aggregate database is searched, associated with the attributes of the search user who searched the aggregate database. A database update unit updates at least one of the plurality of individual databases based on the search information recorded in the search history database, A search control unit inputs a question sentence from a search user, and search results obtained from the aggregated database or individual databases selected by the search user in response to the question sentence, into a predetermined language model to obtain an answer sentence. An information processing system equipped with the following features.

2. When a first search user having the first attribute searches the aggregated database and obtains a first search result, the database update unit saves the first search result in the first individual database among the plurality of individual databases that corresponds to the first attribute. The information processing system according to feature 1.

3. The database update unit deletes the first search result from the individual databases other than the first individual database among the plurality of individual databases. The information processing system according to feature 2.

4. The search control unit transmits the response text to the terminal of the first search user. When the database update unit receives information from the terminal of the first search user indicating that the answer is useful, it saves the first search result in the first individual database. The information processing system according to feature 2.

5. The user's attributes are their job title within the company that owns the information processing system. The information processing system according to feature 1.

6. The search control unit provides a search screen to the terminal used by the search user, which includes an area for inputting the user's attributes, an area for selecting either the aggregated database or one of the multiple individual databases as the search extension database to be used, and an area for inputting a question. The information processing system according to feature 1.

Images