System, method, program

Abstract

The prompts are optimized based on information relevant to the function of the generated AI application. [Solution] An application management device 101 comprising: a request acquisition unit 1013 that acquires requests entered by a user into a generation AI application; a reference information extraction unit 1014 that extracts reference information to be referenced according to the function of the generation AI application from the first request acquired by the request acquisition unit 1013 based on set extraction conditions; a reference information storage unit 1015 that stores the extracted reference information; and a prompt generation unit 1016 that generates a prompt to be input to a generation device 201 that executes processing of the second request acquired by the request acquisition unit 1013, based on the reference information stored in the reference information storage unit 1015 and the second request.

Description

Technical Field

[0001] The present invention relates to a system, a method, and a program.

Background Art

[0002] There are applications for utilizing generative AI. Among generative AI applications, there are those that store users' unique information and generate responses based on the users' unique information. Patent Document 1 describes an information processing apparatus that, when receiving input of question data from a terminal device, generates a prompt for requesting generation of response data for the question data based on the information unique to the user and transmits it to a large language model.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] When generating a prompt based on a user's unique information in the use of a generative AI application, the user's unique information not related to the purpose or function of the generative AI application may be included in the prompt. In this case, the generated prompt is not optimized, and it is difficult to improve the processing accuracy of the generative AI application.

[0005] Therefore, an object of the present invention is to optimize a prompt based on information corresponding to the function of a generative AI application.

Means for Solving the Problems

[0006] A system according to one aspect of the present invention comprises: a request acquisition unit that acquires requests entered by a user into a generation AI application; a reference information extraction unit that extracts reference information to be referenced according to the function of the generation AI application from a first request acquired by the request acquisition unit based on set extraction conditions; a reference information storage unit that stores the extracted reference information; and a prompt generation unit that generates a prompt to be input to a generation device that executes processing of a second request acquired by the request acquisition unit, based on the reference information stored in the reference information storage unit and the second request. [Effects of the Invention]

[0007] The prompts can be optimized based on information relevant to the functions of the generation AI application. [Brief explanation of the drawing]

[0008] [Figure 1] This figure shows an overview of the system configuration according to this embodiment. [Figure 2] This is a block diagram of each device in the system according to this embodiment. [Figure 3] This is an example of the information referenced by the meeting scheduling application according to this embodiment. [Figure 4] This is an example of the information referenced by the sales support application according to this embodiment. [Figure 5] This is an example of application information in the system according to this embodiment. [Figure 6] This is an example of the extraction conditions in the system according to this embodiment. [Figure 7] This is a sequence diagram of the processing in the system according to this embodiment. [Figure 8] This is a flowchart of the reference information extraction process in the system according to this embodiment. [Figure 9] This is a flowchart of the prompt generation process in the system according to this embodiment. [Figure 10]This figure shows an example of updating reference information in the system according to this embodiment. [Figure 11] This is an example of a screen for the meeting scheduling application in the system according to this embodiment. [Figure 12] This is an example of a screen for the sales support application in the system according to this embodiment. [Figure 13] This is an example of the screen for setting extraction conditions in the system according to this embodiment. [Modes for carrying out the invention]

[0009] A preferred embodiment of the present invention will be described with reference to the attached drawings. In each drawing, components denoted by the same reference numerals have the same or similar configuration.

[0010] <<Overview of the structure>> Referring to Figure 1, the system outline of the present invention will be described. In the following description, users of the generation AI application will be referred to as "users," and administrators of the generation AI application will be referred to as "administrative users."

[0011] The generation AI application system 10 includes an application management device 101, a generation device 201, user terminals 301a and 301b, and an external system 401. User terminals 301a and 301b are collectively referred to as user terminal 301.

[0012] The application management device 101 is connected to the generation device 201, the user terminal 301, and the external system 401 via the network N so as to be able to communicate with them.

[0013] In the AI ​​generation application system 10, the user inputs a request to the AI ​​generation application from the user terminal 301. The application management device 101 generates a prompt based on the request and sends it to the generation device 201. The generation device 201 generates a response based on the prompt, and the response is displayed to the user terminal 301 via the application management device 101.

[0014] The generation device 201 generates an answer using a generation AI model based on the prompt received from the application management device 101. The generation AI model is, for example, a large language model. At this time, the generation AI model may be selected from a plurality of candidates.

[0015] The user terminal 301 is an information processing terminal operated by the user, and is, for example, a smartphone, a tablet terminal, or a personal computer (for example, a desktop or a laptop).

[0016] The external system 401 is a system outside the management of the application management device 101 and has external information related to the request. The external system 401 is, for example, a server that provides an application such as a schedule management application, or a server that provides a website.

[0017] Referring to FIG. 2, each part of the application management device 101, the generation device 201, and the user terminal 301 in the generation AI application system 10 will be described.

[0018] The application management device 101 includes a communication unit 1011, a storage unit 1012, a request acquisition unit 1013, a reference information extraction unit 1014, a reference information storage unit 1015, a prompt generation unit 1016, an external information acquisition unit 1017, and a response presentation unit 1018.

[0019] The communication unit 1011 controls communication between the application management device 101 and the generation device 201, the user terminal 301, and the external system 401.

[0020] The storage unit 1012 stores various types of information used for processing in the application management device 101. The storage unit 1012 includes a user information DB 10121, a department information DB 10122, a common information DB 10123, a knowledge information DB 10124, and an extraction condition DB 10125.

[0021] Referring to Figures 3 and 4, the user information DB 10121 and department information DB 10122 will be described. User information DB 10121 is a database that manages reference information for each user. Department information DB 10122 is a database that manages reference information at the department level to which the user belongs. Each generation AI application has its own user information DB 10121 and department information DB 10122. Note that department information DB 10122 may be referenced in common by multiple generation AI applications.

[0022] Figure 3 shows an example of the user information DB10121 and department information DB10122 referenced by the meeting scheduling application in the generated AI application. User information DB10121 contains the following items: user identification information, organization identification information, schedulingable meetings, and remarks. User identification information is an identifier that uniquely identifies the user. Organization identification information is information that identifies the department to which the user belongs. A department is, for example, an organizational unit such as a division or section in a company. Alternatively, a department may be a unit such as a project team or working group. Schedulable meetings are information indicating the names of meetings that the user can optionally attend, that is, meetings that they do not have to attend. Remarks indicate supplementary information not included in the above items. In the example in Figure 3, the user with user identification information U001 belongs to department ORG001, and "Progress X" and "Regular Meeting Y" are registered as schedulingable meetings, and "Tuesday / Thursday afternoon priority" is registered as a remark. This information is used as reference information when scheduling meetings.

[0023] Department information DB10122 contains the following items: organizational identification information, meeting rules, and remarks. Meeting rules indicate the rules for conducting meetings within the department. For example, organizational identification information ORG001 has "Meeting room A takes priority" registered as a meeting rule.

[0024] Figure 4 shows an example of user information DB10121 and department information DB10122, which are referenced by the sales support application within the generated AI application. User information DB10121 contains the following items: user identification information, organization identification information, individual focus area, and individual KPI. Individual focus area indicates the business area that the user is focusing on. Individual KPI indicates the performance indicator set for the user. In the example in Figure 4, the user with user identification information U004 belongs to department ORG001, has "Proposals for the financial industry" registered as their individual focus area, and has "10 proposals per month" registered as their individual KPI. Department information DB10122 contains the following items: organization identification information, organization focus area, and remarks. Organization focus area is information that indicates the business area that the department is focusing on.

[0025] Referring to Figure 5, the Common Information DB 10123 and Knowledge Information DB 10124 will be explained. The Common Information DB 10123 is a database that manages company-wide reference information that is commonly referenced in multiple generation AI applications. The Common Information DB 10123 has items for company-wide focus areas. Company-wide focus areas are information that indicates the business areas that the organization as a whole is focusing on. In the example in Figure 5, "cost reduction," "DX promotion support," and "ESG compliance support" are registered.

[0026] Knowledge Information DB10124 is a database that manages knowledge information that shows insights related to the functions of the generation AI application. Knowledge information is knowledge accumulated in the business area targeted by the generation AI application and is distinct from user information and organizational information. Knowledge Information DB10124 contains application identification information and knowledge items.

[0027] In the example in Figure 5, the app identification information for APP001, related to the meeting scheduling application, registers insights such as "Meeting rooms are crowded on Monday mornings...", while the app identification information for APP002, related to the sales support application, registers insights such as "Services effective in the financial industry are...", "Services effective in the manufacturing industry are...", and "Industries that will experience rapid growth within the next five years are...". Thus, the insight information includes insights from different business areas for each application. The insight information may be accumulated through the use of the generating AI application. Furthermore, at least a portion of the insight information stored in the insight information DB10124 may be commonly referenced by multiple generating AI applications.

[0028] Referring to Figure 6, the extraction conditions DB10125 will be explained. The extraction conditions DB10125 is a database that manages the extraction conditions for reference information for each generated AI application. The extraction conditions DB10125 has the following items: application identification information, target category, item name, judgment word, and extraction range. The target category is information that indicates the storage location of the extracted reference information, either in user information or department information. The item name is information that indicates the item of the reference information to be extracted. The judgment word is a keyword used to determine whether or not it is included in the request. The extraction range is information that indicates the range of information to be extracted from the request if it is determined that the judgment word is included.

[0029] In the example in Figure 6, in the row where the application identification information is APP001, the target category is user information, and the item name is adjustable meeting, the judgment words are set to "meeting," "regular meeting," "MTG," etc., and the extraction range is set to "person's name, meeting name." The extraction conditions can be set, for example, by an administrator user who manages the generation AI application through the administrator user terminal (not shown).

[0030] The reference information storage unit 1015 processes the reference information extracted by the reference information extraction unit 1014 and stores it in the user information DB 10121 or the department information DB 10122. The reference information storage unit 1015 stores the extracted reference information, separating it into user-related information and organizational-related information according to the target category in the extraction condition DB 10125. The reference information storage unit 1015 may also update the stored reference information based on newly extracted reference information. The reference information storage unit 1015 may, for example, store and update the extracted reference information cumulatively, or it may store and update it in a way that overwrites previously stored reference information. The contents of the common information DB 10123 and the knowledge information DB 10124 may, for example, be set by the administrator, or they may be updated as appropriate in response to user requests.

[0031] The request acquisition unit 1013 acquires requests entered by the user into the AI ​​generation application. Specifically, the request acquisition unit 1013 acquires requests entered by the user into the AI ​​generation application via the user terminal 301. Here, the first request and the second request may be requests entered by the same user or requests entered by different users.

[0032] The reference information extraction unit 1014 extracts reference information that is referenced according to the function of the generated AI application from the request acquired by the request acquisition unit 1013 based on extraction conditions. The reference information extraction unit 1014 obtains extraction conditions from the extraction condition DB 10125 that correspond to the application identification information of the generated AI application for which the request was entered.

[0033] For example, the reference information extraction unit 1014 determines whether the acquired judgment word of the extraction condition is included in the request, and if the judgment word is included, it extracts reference information from the request based on the extraction range of the extraction condition. The reference information storage unit 1015 stores the extracted reference information in the user information DB 10121 or the department information DB 10122 according to the target category of the extraction condition. Note that the method of extracting reference information by the reference information extraction unit 1014 is not limited to the method based on the judgment word. For example, the reference information extraction unit 1014 may extract reference information from the request using a generation AI model or a machine learning model.

[0034] The prompt generation unit 1016 generates prompts for processing requests based on reference information stored in the storage unit 1012 and requests acquired by the request acquisition unit 1013. Depending on the request, the prompt generation unit 1016 acquires the reference information necessary for prompt generation from the user information DB 10121, department information DB 10122, common information DB 10123, or knowledge information DB 10124, and generates prompts by combining them with the request. In addition, depending on the function of the generation AI application, the prompt generation unit 1016 may generate prompts based on external information described later, in addition to the reference information and requests.

[0035] The external information acquisition unit 1017 acquires external information related to the request from the external system 401. The external information acquisition unit 1017 determines whether or not to acquire external information based on determination conditions set according to the function of the generating AI application. The determination conditions may be set for each generating AI application, for example, when the request contains a specific keyword, or when external information is always acquired depending on the type of generating AI application. For example, in a meeting scheduling application, if the request is related to scheduling, the external information acquired is the schedule management application, which is the external system 401, which contains the schedule information of the members and the availability of the meeting room. In this case, the external information may be managed by multiple information processing devices.

[0036] The response presentation unit 1018 generates a response to be presented to the user terminal 301 based on the generated data received from the generation device 201. The response presentation unit 1018 may present the generated data as a response as is, or it may configure display elements on the response display screen based on the generated data.

[0037] The generation device 201 includes a communication unit 2011, a processing information acquisition unit 2012, an analysis model unit 2013, and a generated data transmission unit 2014.

[0038] The communication unit 2011 controls communication between the generation device 201 and the application management device 101.

[0039] The processing information acquisition unit 2012 acquires processing information, including prompts generated by the prompt generation unit 1016, from the application management device 101. The processing information may include parameters necessary for processing.

[0040] The analysis model unit 2013 includes an analysis unit 20131 and an answer generation unit 20132. The analysis unit 20131 analyzes the processing information acquired by the processing information acquisition unit 2012. The answer generation unit 20132 generates answers corresponding to the processing information as generated data according to the analysis results.

[0041] The generated data transmission unit 2014 transmits the generated data generated by the analysis model unit 2013 to the application management device 101.

[0042] The user terminal 301 includes a communication unit 3011, an operation acquisition unit 3012, a storage unit 3013, and a display unit 3014. The communication unit 3011 controls communication between the user terminal 301 and the application management device 101. The operation acquisition unit 3012 acquires operation input from the user. The operation acquisition unit 3012 is, for example, a keyboard, mouse, or touch panel. The storage unit 3013 stores various types of information used for processing on the user terminal 301. The display unit 3014 is the display of the user terminal 301.

[0043] <<Process Overview>> Referring to Figure 7, the overall sequence of the generation AI application system 10 will be described.

[0044] In step S701, a first request is sent from the user terminal 301a to the application management device 101. The request acquisition unit 1013 acquires the first request.

[0045] In step S702, the application management device 101 performs a reference information extraction process. Details of the reference information extraction process will be described later.

[0046] In step S703, the application management device 101 sends a prompt corresponding to the first request to the generation device 201.

[0047] In step S704, the analysis model unit 2013 of the generation device 201 generates a response based on the acquired prompt.

[0048] In step S705, the generation device 201 transmits the generated data to the application management device 101.

[0049] In step S706, the application management device 101 sends a response based on the acquired generated data to the user terminal 301a.

[0050] In step S707, a second request is sent from the user terminal 301b to the application management device 101. The request acquisition unit 1013 acquires the second request.

[0051] In step S708, the application management device 101 executes a prompt generation process. Details of the prompt generation process will be described later.

[0052] In step S709, the application management device 101 sends the generated prompt to the generation device 201.

[0053] In step S710, the analysis model unit 2013 of the generation device 201 generates a response based on the acquired prompt.

[0054] In step S711, the generation device 201 transmits the generated data to the application management device 101.

[0055] In step S712, the application management device 101 sends a response based on the acquired generated data to the user terminal 301b.

[0056] Refer to Figure 8 for a detailed explanation of the reference information extraction process. The reference information extraction process is performed on requests acquired by the request acquisition unit 1013. In step S801, the reference information extraction unit 1014 acquires extraction conditions from the extraction condition DB 10125 based on the information of the generation AI application for which the request was input, according to the functions of the generation AI application.

[0057] In step S802, the reference information extraction unit 1014 determines whether the request contains information that matches the extraction conditions. For example, the reference information extraction unit 1014 determines whether the acquired extraction condition judgment word is included in the request.

[0058] If it is determined that the request contains information that matches the extraction criteria (step S802: YES), in step S803, the reference information extraction unit 1014 extracts reference information from the request based on the extraction range of the extraction criteria.

[0059] In step S804, the reference information storage unit 1015 stores the extracted reference information in the user information DB 10121 or the department information DB 10122, according to the target category of the extraction conditions.

[0060] If it is determined that no information matching the extraction criteria is included (step S802: NO), the reference information extraction unit 1014 terminates processing without extracting reference information from the request.

[0061] The prompt generation process will be explained with reference to Figure 9. In step S901, the prompt generation unit 1016 determines whether or not to obtain reference information for processing the second request.

[0062] If it is determined that reference information should be obtained (step S901: YES), in step S902, the prompt generation unit 1016 obtains reference information related to the second request from the user information DB 10121, department information DB 10122, common information DB 10123, and knowledge information DB 10124.

[0063] In step S903, the external information acquisition unit 1017 determines whether or not to acquire external information in processing the second request.

[0064] If it is determined that external information should be acquired (step S903: YES), in step S904, the external information acquisition unit 1017 acquires external information from the external system 401.

[0065] In step S905, the prompt generation unit 1016 generates a prompt to execute the processing of the second request based on the acquired reference information, external information (if acquired), and the second request. Even if it is determined that reference information is not to be acquired (step S901: NO), the process proceeds to step S903, where it is determined whether or not external information needs to be acquired, and then a prompt is generated in step S905. Note that the prompt generation process may be performed not only for the second request but also for other requests acquired by the request acquisition unit 1013.

[0066] Referring to Figure 10, an example of updating reference information in a meeting scheduling application will be explained. In the response display screen 1001 shown in Figure 10, the user (U001) inputs "Regular meeting X is now mandatory to attend. Also, meeting room B can now be used as a meeting room." as request 10011. When the request acquisition unit 1013 acquires request 10011, the reference information extraction unit 1014 executes the reference information extraction process.

[0067] The reference information extraction unit 1014 retrieves the extraction conditions corresponding to the meeting coordination application (APP001) from the extraction conditions DB 10125. The word "regular meeting" included in request 10011 matches the judgment word in the extraction conditions DB 10125. Based on the extraction range (person's name, meeting name), the reference information extraction unit 1014 extracts the information "Regular meeting X is mandatory" from request 10011. Since mandatory meetings do not fall under the category of meetings that can be coordinated, the reference information storage unit 1015 removes Regular Meeting X from the "Meetings that can be coordinated" information for U001 in the user information DB 10121.

[0068] Similarly, the word "meeting room" included in request 10011 matches the judgment word in extraction condition DB 10125. The reference information extraction unit 1014 extracts the information "Meeting Room B is also available" based on the extraction range (location name, rule content). The reference information storage unit 1015 adds "Meeting Room B is also available" to the meeting rule for the department whose department information in department information DB 10122 is "ORG001".

[0069] Subsequently, on the response display screen 1001, a response 10012 indicating "Information has been updated" is displayed to the user. In the example above, both the user information DB 10121 and the department information DB 10122 are updated from a single request, but it is also acceptable for only one of the DBs to be updated from a single request. Furthermore, if the request does not contain information that matches the extraction criteria, the extraction and updating of reference information will not be performed.

[0070] Refer to Figure 11 to explain an example screen of the meeting scheduling application. Figure 11 is an example screen showing the update results from Figure 10. In the response display screen 1101 shown in Figure 11, the user enters "I would like to have a one-hour meeting with A, B, and C at Office A" as request 11011. When the request acquisition unit 1013 acquires request 11011, the prompt generation unit 1016 executes the prompt generation process.

[0071] The prompt generation unit 1016 determines that it is time to acquire reference information and retrieves information on available meetings for Person A, Person B, and Person C from User Information DB 10121, and retrieves "Meeting Room A is prioritized, Meeting Room B is also available" as a meeting rule for ORG001 from Department Information DB 10122. Here, the meeting rule in Department Information DB 10122 includes the information "Meeting Room B is also available" which was added in the update process in Figure 10. In addition, related information from Common Information DB 10123, such as "Cost Reduction" or "Meetings should be within 1 hour," may be referenced. In addition, knowledge information related to the meeting scheduling application, such as "Meeting rooms are crowded on Monday mornings," may be referenced from Knowledge Information DB 10124.

[0072] The external information acquisition unit 1017 determines that it is time to acquire external information and obtains the schedule information of each member and the availability of meeting rooms as external information from the schedule management application, which is the external system 401. The prompt generation unit 1016 generates a prompt based on the acquired reference information, external information, and request 11011.

[0073] The generation device 201 generates a response based on the generated prompt. On the response display screen 1101, the user is presented with candidate dates and times and available meeting rooms as response 11012. The inclusion of meeting room B as a candidate reflects the fact that the information "Meeting room B is also available" was added to the department information DB 10122 during the update process in Figure 10. Furthermore, the candidate dates and times and meeting room selection fields included in response 11012 are information configured on the response display screen 1101 by the response presentation unit 1018 based on the generated data generated by the generation device 201. Here, response 11012 may also reflect the knowledge included in the knowledge information, such as "Meeting rooms are crowded on Monday mornings...".

[0074] Refer to Figure 12 to explain an example screen of the sales support application. In the response display screen 1201 shown in Figure 12, user A (U004) inputs request 12011 as "Our department's focus area is the financial industry, so what products can we propose there?" When the request acquisition unit 1013 acquires request 12011, the reference information extraction unit 1014 executes the reference information extraction process. The word "department" included in request 12011 matches the judgment word in the extraction condition DB 10125. The reference information extraction unit 1014 extracts "financial industry," and the reference information storage unit 1015 stores "financial industry" in the organizational focus area ORG001 of the department information DB 10122. The extracted reference information may also be used to generate prompts based on request 12011.

[0075] When generating a prompt for request 12011, the prompt generation unit 1016 retrieves the individual KPI (10 proposals per month) U004 from the user information DB 10121 shown in Figure 4, the organizational focus area (financial industry) ORG001 from the department information DB 10122, and the company-wide focus area (DX promotion support) from the common information DB 10123. The prompt generation unit 1016 may also retrieve information on services effective for the financial industry as knowledge from the knowledge information DB 10124. This reference information is reflected in response 12012. Specifically, "Prioritize projects that can be proposed in the short term to match the target of 10 proposals per month" reflects the individual KPI from user information DB 10121, and "Select mainly DX-related services for the financial industry from the perspective of DX promotion support" reflects the information "DX promotion support" included in the company-wide focus area in common information DB 10123. The generated response 12012 may include insights into "services effective for the financial industry."

[0076] Furthermore, on the response display screen 1202, user B (U005) inputs "I would like to propose a data analysis service" as request 12021. User B (U005) belongs to the same department (ORG001) as user A (U004). When generating a prompt for request 12021, the prompt generation unit 1016 retrieves U005's personal KPI (monthly order value of 5 million) from user information DB 10121, the organization's focus area (financial industry) for ORG001 from department information DB 10122, and the company-wide focus area (ESG support) from common information DB 10123. The prompt generation unit 1016 may also retrieve information on services effective for the financial industry as knowledge from knowledge information DB 10124. Here, the organization's focus area "financial industry" in department information DB 10122 is reference information extracted from user A's request 12011. In this way, reference information extracted from the first request entered by the first user (User A) is referenced when generating a prompt for a second user (User B), who is different from the first user.

[0077] Response 12022 reflects this reference information. Specifically, "Prioritizing high-value projects considering order value targets" reflects the individual KPIs in User Information DB10121, and "ESG support is a company-wide theme, including ESG analysis services for the financial industry" reflects the information "ESG support" included in the company-wide focus areas of Common Information DB10123 and the information "Financial Industry" included in the organizational focus areas of Departmental Information DB10122. The generated response 12022 may also include insights into "services effective for the financial industry."

[0078] Refer to Figure 13 to illustrate an example of a screen for setting extraction conditions for reference information. Figure 13 shows a settings screen 1301 for the administrator to set extraction conditions for each generated AI application. In the settings screen 1301, "Meeting Coordination Application (APP001)" is selected as the target generated AI application. The settings screen 1301 includes a user information extraction condition setting area 1302 and a department information extraction condition setting area 1303.

[0079] The user information extraction condition setting area 1302 includes an item name input field 13021, a first judgment word input field 13022, a second judgment word input field 13023, and an extraction range input field 13024. In the example in Figure 13, "Adjustable Meeting" is set in the item name input field 13021, "Optional Participation" etc. is set in the first judgment word input field 13022, "MTG", "Regular Meeting" etc. is set in the second judgment word input field 13023, and "Person's Name, Meeting Name" etc. is set in the extraction range input field 13024.

[0080] The department information extraction condition setting area 1303 includes an item name input field 13031, a first judgment word input field 13032, a second judgment word input field 13033, and an extraction range input field 13034. In the example in Figure 13, "Meeting Rules" is set in the item name input field 13031, "Meeting Room," "Priority," etc. are set in the first judgment word input field 13032, "Possible," "Prohibited," etc. are set in the second judgment word input field 13033, and "Location Name, Rule Content," etc. are set in the extraction range input field 13034.

[0081] The extraction conditions set by the administrator via the settings screen are stored in the extraction conditions DB 10125. The settings in the user information extraction condition setting area 1302 in Figure 13 correspond to the rows in the extraction conditions DB 10125 in Figure 6 where the application identification information is APP001 and the target category is user information. Similarly, the settings in the department information extraction condition setting area 1303 correspond to the rows in the extraction conditions DB 10125 in Figure 6 where the application identification information is APP001 and the target category is department information.

[0082] This embodiment has been described above. The generation AI application system 10 according to this embodiment includes a request acquisition unit 1013 that acquires requests entered by a user into the generation AI application, a reference information extraction unit 1014 that extracts reference information to be referenced according to the function of the generation AI application from the first request acquired by the request acquisition unit 1013 based on set extraction conditions, a reference information storage unit 1015 that stores the extracted reference information, and a prompt generation unit 1016 that generates a prompt to be input to the generation device 201 that executes processing of the second request acquired by the request acquisition unit 1013, based on the reference information stored in the reference information storage unit 1015 and the second request.

[0083] In this embodiment, reference information is extracted based on set extraction conditions, so that it is narrowed down to information corresponding to the function of the generation AI application and used for prompt generation. Therefore, it is possible to prevent unnecessary information from being included in the prompt according to the function of the generation AI application, optimize the prompt, and improve the processing accuracy of the generation AI application.

[0084] In the above embodiment, the extraction conditions include conditions for extracting information about the user and conditions for extracting information about the organization to which the user belongs, and the reference information storage unit 1015 may store the extracted reference information separately into information about the user and information about the organization.

[0085] According to this embodiment, since the extraction conditions include conditions for extracting information about the user and conditions for extracting information about the organization, information about the user and information about the organization can be extracted separately from the request and stored separately. Therefore, the prompt generation unit can selectively refer only to the reference information of the categories necessary for prompt generation, thereby improving the processing efficiency of the generation AI application.

[0086] In the above embodiment, at least a portion of the information relating to the organization may be commonly referenced in multiple generative AI applications.

[0087] In this configuration, some information about the organization is used in common among the various generative AI applications, thus maintaining information consistency. For example, information about the organization's overall policies can be referenced in common across multiple generative AI applications. Therefore, the accuracy of prompts based on reference information can be improved, and the processing accuracy of the generative AI applications can be enhanced.

[0088] In the above embodiment, the extraction conditions include conditions for extracting knowledge information that indicates insights regarding the functionality of the generated AI application, and the reference information storage unit 1015 may store the extracted knowledge information.

[0089] In this embodiment, insight information is extracted and stored from requests based on extraction conditions corresponding to the functions of the generation AI application. As a result, insight information is accumulated through the use of the generation AI application, and the accuracy of prompts based on the accumulated insight information improves, thereby increasing the processing accuracy of the generation AI application.

[0090] In the above embodiment, at least a portion of the knowledge information may be commonly referenced in multiple generative AI applications.

[0091] In this configuration, some of the knowledge information is used in common among each generating AI application, thus maintaining information consistency. Therefore, the accuracy of prompts based on reference information is improved, and the processing accuracy of the generating AI application can be enhanced.

[0092] In the above embodiment, the first request may be a request entered by the first user, and the second request may be a request entered by a second user who is different from the first user.

[0093] According to this embodiment, reference information extracted from the first user's request can be used to generate prompts in response to the second user's request, thereby improving the accuracy of the prompts and enhancing the processing accuracy of the generation AI application.

[0094] In the above embodiment, the reference information extraction unit 1014 may extract reference information from the second request based on the extraction conditions, and the reference information storage unit 1015 may update the stored reference information based on the extracted reference information.

[0095] According to this embodiment, reference information can be updated as needed based on various requests, including the second request. For example, as the generation AI application is used, the reference information is updated and accumulated. Therefore, the inclusion of outdated information in prompts is suppressed, and the accuracy of prompts based on reference information is improved, thereby increasing the processing accuracy of the generation AI application.

[0096] In the above embodiment, the system further includes an external information acquisition unit 1017 that acquires external information related to the second request from an external system 401, and the prompt generation unit 1016 may generate a prompt based on the reference information, the external information, and the second request.

[0097] According to this embodiment, external information related to the second request is obtained from the external system 401, and a prompt is generated based on the reference information and the external information. Therefore, in addition to the reference information, information related to the request can be supplemented from the external system 401, thereby improving the processing accuracy of the generation AI application.

[0098] The method according to this embodiment includes a computer acquiring a request entered by a user into a generation AI application, extracting reference information to be referenced according to the function of the generation AI application from the acquired first request based on set extraction conditions, storing the extracted reference information, and generating a prompt to be input to a generation device that executes processing of the acquired second request, based on the stored reference information and the second request.

[0099] The program according to this embodiment causes the computer to perform the following actions: to obtain a request entered by the user into the generation AI application; to extract reference information that is referenced according to the function of the generation AI application from the obtained first request based on set extraction conditions; to store the extracted reference information; and to generate a prompt to be input to the generation device that executes the processing of the obtained second request, based on the stored reference information and the second request.

[0100] The embodiments described above are provided to facilitate understanding of the present invention and are not intended to limit its interpretation. The elements, their arrangement, conditions, etc., of the embodiments are not limited to those exemplified and can be modified as appropriate. Furthermore, it is possible to partially substitute or combine the configurations shown in different embodiments. [Explanation of symbols]

[0101] 10...Generating AI application system, 101...Application management device, 1013...Request acquisition unit, 1014...Reference information extraction unit, 1015...Reference information storage unit, 1016...Prompt generation unit, 201...Generating device, 301a,301b...User terminals

Claims

1. A request acquisition unit that acquires requests entered by the user into the AI ​​application, A reference information extraction unit extracts reference information, which is referenced according to the function of the generation AI application, from the first request acquired by the request acquisition unit based on set extraction conditions. A reference information storage unit for storing the extracted reference information, A system comprising: a prompt generation unit that generates a prompt to be input to a generation device that executes processing of a second request acquired by the request acquisition unit, based on reference information stored in the reference information storage unit and the second request.

2. The system according to claim 1, The extraction conditions include conditions for extracting information about the user and conditions for extracting information about the organization to which the user belongs. The aforementioned reference information storage unit is a system that stores the extracted reference information, separating it into information about the user and information about the organization.

3. The system according to claim 2, A system in which at least a portion of the information concerning the aforementioned organization is commonly referenced in multiple generative AI applications.

4. The system according to claim 1, The extraction conditions include conditions for extracting insight information that indicates insights regarding the functionality of the generated AI application, The aforementioned reference information storage unit is a system that stores the extracted knowledge information.

5. The system according to claim 4, A system in which at least a portion of the aforementioned knowledge information is commonly referenced in multiple generative AI applications.

6. The system according to claim 1, A system in which the first request is a request entered by a first user, and the second request is a request entered by a second user different from the first user.

7. The system according to claim 1, The system comprises a reference information extraction unit that extracts reference information from the second request based on the extraction conditions, and a reference information storage unit that updates the stored reference information based on the extracted reference information.

8. The system according to claim 1, The system further includes an external information acquisition unit that acquires external information related to the second request from an external system, The prompt generation unit is a system that generates the prompt based on the reference information, the external information, and the second request.

9. Computers To obtain the requests entered by the user into the AI ​​application, The reference information to be referenced according to the function of the aforementioned generation AI application is extracted from the acquired first request based on the set extraction conditions. The extracted reference information is to be stored, A method comprising generating a prompt to be input to a generating device that performs processing of a second acquired request, based on stored reference information and the second request.

10. On the computer, To obtain the requests entered by the user into the AI ​​application, The reference information to be referenced according to the function of the aforementioned generation AI application is extracted from the acquired first request based on the set extraction conditions. The extracted reference information is to be stored, A program that generates a prompt to be input to a generating device that processes the acquired second request, based on stored reference information and the second request, and causes the device to execute this.

Images