Information processing device, relevance estimation system, relevance estimation method, and program

Abstract

The aim is to provide technology that improves customer satisfaction when obtaining information related to inquiries. [Solution] An information processing device comprising: an inquiry acquisition unit that acquires inquiry information relating to an inquiry; a relevance estimation unit that causes a generating AI to estimate the degree of relevance between each of the one or more target information related to the inquiry and the inquiry; and a relevance correction unit that corrects the degree of relevance according to the estimation unit of the target information for which the degree of relevance has been estimated by the generating AI.

Description

Technical Field

[0001] The present invention relates to an information processing apparatus, a relevance estimation system, a relevance estimation method, and a program.

Background Art

[0002] In fields such as user support, when receiving a user's question, there is a technology for extracting cases related to the question.

[0003] For example, in the related case reference system disclosed in Patent Document 1, a user transmits a question email describing an error message or error log of a trouble related to an information processing system to a responsible terminal. When the responsible terminal transfers the question email to a case accumulation server, the case accumulation server extracts keywords from the question email and searches for cases using the keywords.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Conventionally, in order to obtain information desired by a user, it has been common to perform a process of searching for a character string included in the information to be searched using keywords, as in the technology according to Patent Document 1. However, a certain skill is required to represent a desire by keywords, and in some cases, there is a possibility that the desired information cannot be obtained. It is difficult to say that the relationship between the keywords and the obtained information is visualized, and it is not easy to verify whether the information along with the desire has been acquired.

[0006] The present invention has been made in view of the above points, and an object thereof is to provide a technology for improving the satisfaction when obtaining information related to an inquiry. [Means for solving the problem]

[0007] This application includes several means for solving the above-mentioned problems, some examples of which are as follows.

[0008] To solve the above problems, an information processing device according to one aspect of the present invention is characterized by comprising: an inquiry acquisition unit that acquires inquiry information relating to an inquiry; a relevance estimation unit that causes a generating AI to estimate the degree of relevance of each of the one or more target information related to the inquiry with respect to the inquiry; and a relevance correction unit that corrects the degree of relevance according to the estimation unit of the target information for which the degree of relevance has been estimated by the generating AI.

[0009] The correlation correction unit may be characterized by correcting the correlation using a plurality of groups, each composed of a plurality of the aforementioned target information whose correlation has been estimated, as the estimation unit.

[0010] The information processing device may be characterized by comprising a target information extraction unit that extracts a plurality of target information for each of the plurality of group information, the relevance estimation unit having the generating AI re-estimate the relevance for each of the plurality of target information extracted by the target information extraction unit to obtain a re-estimated relevance, and the relevance correction unit performing a correction using the re-estimated relevance.

[0011] The correlation correction unit may be characterized by performing the correction by normalizing the correlation of other subject information belonging to the same group using the re-estimated correlation of the subject information.

[0012] The information processing device may include a target information acquisition unit that acquires document information relating to past cases as target information, and the relevance estimation unit may be characterized by causing a generating AI to estimate the relevance of a plurality of document pieces of information related to the inquiry as target information.

[0013] The correlation estimation unit obtains a limit value, which is the maximum amount of information per processing, and determines the number of extractable data using the limit value and the number of groups. The target information extraction unit may be characterized by extracting the target information for each of the multiple groups, up to the number of extractable data points.

[0014] The correlation estimation unit may be characterized in that, if the total number of target information extracted by the target information extraction unit exceeds the limit value, it divides the extracted target information into groups, the target information extraction unit extracts multiple target information for each group, the correlation estimation unit repeats the division until the total number of target information extracted by the target information extraction unit is less than or equal to the limit value, and for each of the target information included in the group whose total number of target information is less than or equal to the limit value, it has the generating AI re-estimate the correlation to obtain a re-estimated correlation.

[0015] The information processing device may include a target information acquisition unit that acquires document information relating to past cases, and the relevance estimation unit may be characterized by causing a generating AI to select the document information, generating a reason for selection for each selected document information, estimating the degree of relevance, and causing the generating AI to estimate the re-estimated degree of relevance using the reason for selection and the degree of relevance.

[0016] Furthermore, in order to solve the above problems, another aspect of the present invention relates to a relevance estimation system, which is a computer-based relevance estimation system, characterized in that the computer performs: an inquiry acquisition step of acquiring inquiry information relating to an inquiry; a relevance estimation step of causing a generating AI to estimate the degree of relevance between each of a plurality of object information related to the inquiry and the inquiry; and a relevance correction step of correcting the degree of relevance according to the estimation unit of the object information for which the degree of relevance has been estimated by the generating AI.

[0017] Also, in order to solve the above problems, a relevance estimation method according to another aspect of the present invention is a relevance estimation method executed by an information processing apparatus, and includes an inquiry acquisition procedure for acquiring inquiry information regarding an inquiry, a relevance estimation procedure for causing a generation AI to estimate the relevance with the inquiry for each of a plurality of target information related to the inquiry, and a relevance correction procedure for correcting the relevance according to the estimation unit of the target information for which the relevance has been estimated by the generation AI.

[0018] Also, in order to solve the above problems, a program according to another aspect of the present invention is a program for causing a processing unit of a computer to execute a relevance estimation method, and includes an inquiry acquisition procedure for acquiring inquiry information regarding an inquiry, a relevance estimation procedure for causing a generation AI to estimate the relevance with the inquiry for each of a plurality of target information related to the inquiry, and a relevance correction procedure for correcting the relevance according to the estimation unit of the target information for which the relevance has been estimated by the generation AI.

Advantages of the Invention

[0019] According to the present invention, it is possible to provide a technique for improving the satisfaction when obtaining information related to an inquiry.

[0020] Problems, configurations, and effects other than those described above will be clarified by the description of the following embodiments.

Brief Description of the Drawings

[0021] [Figure 1] It is a diagram showing an example of a functional block of a relevance estimation system. [Figure 2] It is a diagram showing an example of a data structure of target information. [Figure 3] It is a diagram showing an example of a hardware configuration of a relevance estimation device. [Figure 4] It is a flowchart showing an example of target information selection processing. [Figure 5] It is a diagram showing an example of an overview of relevance estimation processing. [Figure 6] It is a diagram showing an example of target information selection processing in a modification of this embodiment.

Best Mode for Carrying Out the Invention

[0022] Hereinafter, an example of an embodiment of the present invention will be described based on the drawings. FIG. 1 is a diagram showing an example of a functional block of a relevance estimation system 1 in the first embodiment. The relevance estimation system 1 in this embodiment includes a relevance estimation device 10, a target information request device 20, and a generation AI system 30. The relevance estimation device 10, the target information request device 20, and the generation AI system 30 are communicably connected via a network N.

[0023] As an example of this embodiment, the target information request device 20 is a device operated by a person in charge of providing user support for solving problems that the user has. The relevance estimation device 10 selects target information related to the inquiry according to the inquiry information from the target information request device 20 and responds to the target information request device 20. As an example, the target information is document information related to past solution cases that have solved the problems related to the inquiry. The response includes, in addition to the selected target information, a relevance indicating the degree of relevance to the inquiry, and a reason for selecting the target information. The relevance and the reason for selection are output by the target information request device 20 together with the selected target information.

[0024] The relevance estimation device 10 is an information processing device such as a server computer, PC (Personal Computer), or workstation, and is, for example, a device owned by a business operator that provides relevance estimation services. The relevance estimation device 10 comprises a processing unit 110, a storage unit 120, an input unit 130, an output unit 140, and a communication unit 150. The processing unit 110 comprehensively controls the entire relevance estimation device 10. The storage unit 120 stores information necessary for processing by the processing unit 110. The input unit 130 receives information input to the relevance estimation device 10 from input devices connected via the input interface of the relevance estimation device 10. The output unit 140 outputs the information stored in the relevance estimation device 10 to output devices connected via the output interface of the relevance estimation device 10. The communication unit 150 mediates the transmission and reception of information with other information processing devices connected via the network N.

[0025] The processing unit 110 comprises a query acquisition unit 111, a relevance estimation unit 112, a response unit 113, a relevance correction unit 114, a target information extraction unit 115, and a target information acquisition unit 116. The query acquisition unit 111 acquires query information, which is information related to the query, from the target information request device 20. The query information includes search information used to search for target information. For example, the search information may be keywords indicating business challenges faced by users handled by the person in charge of the target information request device 20, or sentences containing those keywords.

[0026] The relevance estimation unit 112 instructs the generating AI to select one or more target information related to the search information. The relevance estimation unit 112 also instructs the generating AI to estimate the degree of relevance of each target information to the inquiry and to generate the reason for the selection.

[0027] Due to processing load limitations, there may be restrictions on the amount of data that the generation AI system 30 (described later) can process simultaneously. If the total amount of data for the target information 121 records (described later) exceeds a preset limit, the relevance estimation unit 112 divides the target information 121 records into groups according to predetermined estimation units, thereby causing the generation AI system 30 to perform distributed processing. The relevance estimation unit 112 generates execution instructions, i.e., prompts, for the generation AI system 30 to select records of the target information 121 for each group, estimate their relevance, and generate reasons for selection. In this embodiment, the records of the target information 121 may be simply referred to as "target information."

[0028] The relevance estimation unit 112, upon transmitting an execution instruction to the generation AI system 30, obtains the selected target information, the degree of relevance between the target information and the search information, and the reason for selecting the target information. As an example, the relevance estimation unit 112 causes the generation AI to select a solution case related to the search information from past solution cases. The selection of target information includes selecting a predetermined number of target information with a high degree of relevance from the target information to be searched. Furthermore, the selection of target information includes assigning a degree of relevance to all target information to be searched. This is because, when the response unit 113 (described later) outputs the selected target information in order of decreasing relevance, the output of target information with a low degree of relevance will have a lower priority, effectively making it equivalent to selecting target information with a high priority.

[0029] Furthermore, the relevance estimation unit 112 generates a prompt for the generating AI to re-estimate the relevance for each of the multiple target information extracted by the target information extraction unit 115, which will be described later. As a result, the relevance estimation unit 112 obtains a re-estimated relevance for each of the extracted target information.

[0030] The response unit 113 responds to the target information request device 20, which is the source of the inquiry, with the degree of relevance to the search information and the reason for selection for each target information selected in response to the inquiry. The response unit 113 may include all of the target information selected by the degree of relevance estimation unit 112 in the response, or it may include a predetermined number of resolution target information with a high degree of relevance from among the selected target information.

[0031] The relevance correction unit 114 corrects the relevance estimated by the generating AI according to the estimation unit of the target information for which the relevance has been estimated by the generating AI. That is, the relevance correction unit 114 corrects the relevance using multiple groups, each composed of multiple target information for which the relevance has been estimated, as estimation units. More specifically, the relevance correction unit 114 performs the correction by normalizing the relevance of other target information belonging to the same group using the re-estimated relevance obtained for the target information.

[0032] The target information extraction unit 115 extracts multiple target information for each group composed of multiple target information.

[0033] The target information acquisition unit 116 retrieves records of the target information 121 to be searched from among the target information 121 stored in the storage unit 120. For example, the target information 121 is document information related to past cases.

[0034] The memory unit 120 stores the target information 121. The target information 121 will be described later.

[0035] The target information request device 20 is an information processing device such as a PC, workstation, smartphone, PDA (Personal Data Assistant), or tablet device. The target information request device 20 comprises a processing unit 210, a storage unit 220, an input unit 230, an output unit 240, and a communication unit 250. The processing unit 210 comprehensively controls the entire target information request device 20. The storage unit 220 stores information necessary for processing by the processing unit 210. The input unit 230 receives information input to the target information request device 20 from input devices connected via the input interface of the target information request device 20. The output unit 240 outputs information stored in the target information request device 20 to output devices connected via the output interface of the target information request device 20. The communication unit 250 mediates the transmission and reception of information with other information processing devices connected via the network N.

[0036] The processing unit 210 includes an inquiry management unit 211. The inquiry management unit 211 transmits inquiry information, including search information entered by the person in charge, to the relevance estimation device 10. The inquiry management unit 211 outputs the selected target information, the degree of relevance, and the reason for selection, which are included in the response transmitted from the relevance estimation device 10, to the output device of the target information request device 20.

[0037] The generation AI system 30 utilizes machine learning and natural language processing to generate and output various documents, images, videos, audio, program code, etc., in response to prompts. If an API (Application Programming Interface) is provided, the relevance estimation device 10 can also access the API function to input prompts. The generation AI system 30 can utilize LLMs (Large Language Models) such as OpenAI's CHATGPT® or Google's Gemini. The generation AI system 30 may be built locally, and the relevance estimation device 10 may access it via a LAN (Local Area Network).

[0038] Hereinafter, the generation AI system 30 will be described as a device that generates text data in response to instructions using an LLM. However, the model used by the generation AI system 30 is not limited to an LLM. For example, the generation AI system 30 may select images related to an image input as search information as target information, and generate a document for each piece of target information that includes the degree of relevance and the reason for selection.

[0039] Network N is a communication network such as a LAN (Local Area Network), WAN (Wide Area Network), the Internet, or a mobile phone network. Network N may also be a VPN (Virtual Private Network) on a wireless communication network such as a mobile phone network.

[0040] Figure 2 shows an example of the data structure of target information 121. For example, target information 121 includes target ID, case name, case summary, issue, solution, value provided, customer industry, customer size, service name, characteristics, target business, delivery method, and keywords.

[0041] The Information ID is an identifier that identifies the record of Target Information 121. The Case Name is a string indicating the name of the Target Information (i.e., the record of Target Information 121). The Case Summary is a string indicating a brief description of the Target Information. The Issues are information indicating the specific problems or difficulties faced in the case related to the Target Information. The Resolutions are information indicating the issues resolved by the case. The Value Provided is information indicating the value provided to the customer in the case.

[0042] Customer Industry is information indicating the customer's industry. Customer Size is information indicating the customer's size, such as annual sales or number of employees. Service Name is information indicating the name of the customer's service or product. Features are information indicating noteworthy points in the case study. Target Business is information indicating the content of the business being targeted. For example, target business may be "production planning" or "equipment maintenance." Delivery Method is information indicating the specific method used to solve the problem. For example, the delivery method may include the technology implemented or specific procedures. Keywords are information indicating the words used to allow case study searches.

[0043] Note that the target information 121 does not necessarily have to include all of the items shown in Figure 2, and may include other items. For example, the target information 121 may include information such as the customer name, project phase indicating the progress of the project, project overview diagram, project scale, information identifying the internal website, introductory materials, interview sheet, collaborating or partner company, registration date for the target information 121, design person in charge, sales person in charge, etc. Also, certain information such as the project overview may be generated by the generation AI when generating the selection reason, for example.

[0044] Figure 3 shows an example of the hardware configuration of the relevance estimation device 10. The relevance estimation device 10 comprises a processor 101, memory 102, storage 103, communication device 104, and a bus 105 connecting the devices. In addition, the relevance estimation device 10 may also include an input device and an output device.

[0045] The processor 101 is a computing device such as a CPU (Central Processing Unit) or GPU (Graphics Processing Unit), and it performs processing according to a program recorded in memory 102 or storage 103. In the relevance estimation device 10, processing is performed by the processor 101, which operates according to a program read from memory 102 or storage 103. The processing unit 110, query acquisition unit 111, relevance estimation unit 112, response unit 113, relevance correction unit 114, target information extraction unit 115, and target information acquisition unit 116 each realize their respective functions by having the processor 101 execute a program.

[0046] Memory 102 is a storage device such as RAM (Random Access Memory) or flash memory, and functions as a storage area where programs and data are temporarily read. Storage 103 is a writable and readable storage device. The storage unit 120's functions are realized by either memory 102 or storage 103. Alternatively, the storage unit 120 may be realized by a storage device connected via communication device 104.

[0047] The communication device 104 is an interface for connecting the relevance estimation device 10 to an external device. For example, the communication device 104 uses an antenna that can utilize predetermined radio waves (e.g., 5GHz band, 2.4GHz band, etc.) to establish a connection with other information processing devices using the Wi-Fi standard for wireless communication. The communication unit 150's functions are realized by the communication device 104.

[0048] The processor 101, memory 102, storage 103, and communication device 104 are connected to each other by connecting wires such as a bus 105.

[0049] Each processing unit (inquiry acquisition unit 111, relevance estimation unit 112, response unit 113, relevance correction unit 114, target information extraction unit 115, and target information acquisition unit 116) may be constructed using dedicated hardware (ASIC, GPU, etc.) to implement its respective function. Furthermore, the processing of each processing unit may be executed on a single piece of hardware or on multiple pieces of hardware.

[0050] Since the target information request device 20 basically has the same hardware configuration as the relevance estimation device 10, its explanation will be omitted.

[0051] Figure 4 is a flowchart showing an example of the target information selection process. In this embodiment, the relevance estimation unit 112 of the relevance estimation device 10 accepts the setting of a limit value, which is the upper limit of the amount of data that the generating AI system 30 can process simultaneously, before the start of the process shown in this flowchart. The type of limit value is not limited and may be, for example, an upper limit of the data size, an upper limit of the number of data points, or an upper limit of the number of characters contained in the data.

[0052] The processing in this flowchart begins, for example, when the inquiry management unit 211 of the target information request device 20 sends inquiry information to the relevance estimation device 10, and the relevance estimation device 10 receives the inquiry information via the communication unit 150.

[0053] First, the inquiry acquisition unit 111 acquires the inquiry information (step S11). Specifically, the inquiry acquisition unit 111 acquires the search information contained in the inquiry information.

[0054] Next, the target information acquisition unit 116 acquires the target information (step S12). For example, the target information acquisition unit 116 acquires past resolution cases from the records contained in the target information 121. The acquired records of the target information 121 are treated as search targets. As an example, the target information acquisition unit 116 acquires all the records of the target information contained in the target information 121. As another example, the target information acquisition unit 116 may acquire multiple records contained in the target information 121 by using the information contained in the query information acquired before the start of this process to extract target information related to the query information from the target information 121. For example, the target information acquisition unit 116 may extract information that identifies a client from the query information and acquire cases related to that client from the target information 121.

[0055] Next, the relevance estimation unit 112 divides the target information into groups so that it fits within the limit value (step S13). Specifically, the relevance estimation unit 112 uses the limit value as the estimation unit and groups the multiple target information acquired in step S12 so that the target information included in each group fits within the limit value. For example, if the limit value is the maximum number of data per processing and the limit value is "20", and the number of target information acquired in step S12 is "100", the relevance estimation unit 112 divides the target information into 5 groups.

[0056] Next, the relevance estimation unit 112 generates commands for each group to select target information, generate a reason for selection for each piece of target information, and estimate the degree of relevance (step S14). Specifically, the relevance estimation unit 112 generates a prompt command to select target information related to the search information from the target information to be searched obtained in step S12. The prompt may include the target information to be searched obtained in step S12, or it may include information specifying the storage location of the target information. The relevance estimation unit 112 includes a command in the prompt to output the degree of relevance between the target information and the search information, and the reason for selecting the target information. As an example, the relevance estimation unit 112 generates the following prompt.

[0057] (Example of a prompt) Select relevant case studies from the attached list of target information based on the entered search information and output them as a list of relevant case studies. The search should include not only term matches but also cases with similar meanings. When outputting, please include the selection reason explaining why each case study is relevant to the search information, and a 10-point relevance rating. The output should be presented in descending order of relevance based on "Information ID," "Case Name," "Selection Reason," and "Relevance Level." "Input Search Information" Please give me some examples related to temperature. "Input Target Information 1" Information ID:SC-1001 Project Name: Remote Maintenance using Temperature and Humidity Sensors and Monitoring Center BPO Project Overview: Combining temperature and humidity sensors with a monitoring center BPO to provide low-cost remote monitoring. Problem: Conventional temperature and humidity sensors had issues such as requiring on-site verification. Solution: By utilizing a monitoring center BPO, real-time remote monitoring and health checks became possible at a low cost. It also became possible to manage multiple systems simultaneously. Keywords: Visualization temperature and humidity sensor "Input Target Information 2" Information ID:SC-1002 Project Name: Weather-Based Traffic Demand Forecasting and Transportation Optimization Project Overview: Predicting traffic volume that changes due to weather conditions and optimizing transportation routes. Problem: Conventional weather-based traffic volume forecasts had large discrepancies, requiring a large margin of error, which led to losses. Solution: Traffic volume was predicted using AI based on weather forecasts, reducing the margin rate to less than 20% of the conventional rate. Furthermore, transportation costs were reduced by optimizing transportation. Keywords: Transportation optimization, Demand forecasting, AI

[0058] In this embodiment, the relevance estimation unit 112 can generate a prompt to select target information related to the search information based on the values ​​included in each item of the target information 121. In this case, it is desirable that the selection reason be created taking into consideration the values ​​included in each item of the target information 121.

[0059] Next, the relevance estimation unit 112 outputs a command and obtains the result (step S15). Specifically, the relevance estimation unit 112 sends the prompt generated in step S13 to the generation AI system 30 and receives the result.

[0060] Next, the relevance estimation unit 112 determines the number of extractable data points using the limit value and the number of groups (step S16). Specifically, the relevance estimation unit 112 obtains the limit value and uses the limit value and the number of groups to determine the number of target information points that can be extracted per group. For example, if the limit value is the maximum number of data points per processing step, the relevance estimation unit 112 can determine the number of extractable data points by dividing the limit value by the number of groups.

[0061] Next, the target information extraction unit 115 extracts a number of target information items for each group based on the number of extractable data items (step S17). Specifically, for each of the multiple groups obtained in step S13, the target information extraction unit 115 extracts a number of target information items determined in step S16 from among the multiple target information items included in the group, based on predetermined criteria. For example, if the number of extractable data items determined in step S16 is "2", the target information extraction unit 115 extracts the target information item with the highest degree of relevance and the target information item with the lowest degree of relevance. Alternatively, the target information extraction unit 115 may extract target information items such that the degrees of relevance of the extracted target information items are spaced approximately equally. Alternatively, the target information extraction unit 115 may extract target information items randomly.

[0062] Figure 5 shows an example of the relevance estimation process. The top right column and the center column of Figure 5 represent one of the groups formed by dividing multiple target information contained in the "Data A" database, respectively, while the right column represents the target information contained in the "Data B" database. The relevance estimation unit 112 can generate different system prompts for "Data A" and "Data B" to compensate for differences due to differences in data format.

[0063] In Figure 5, "Data A1" is labeled "(Estimated 4 - Expected 6)". This indicates that for "Data A1", the correlation estimated by the generating AI system 30 is "4", while the expected correlation is "6". The expected correlation is, for example, the correlation that would be estimated if the generating AI were to simultaneously estimate correlation for all target information. The expected correlation is a theoretical value used for explanation and is not the actual value output by the correlation estimation device 10.

[0064] Many of the target information items shown in the top row of this figure have different estimated relevance levels than expected. This is presumably because the generating AI system 30 estimates relevance levels for target information belonging to the same group, and the bias in the content of the target information in each group leads to a bias in the estimated relevance levels. For example, if a certain group contains a concentration of cases related to specific search information (e.g., keywords), then when the cases are divided into groups and the relevance level of each case is calculated, cases that would have had a high relevance level if they had been included in other groups may be extracted with a low relevance level.

[0065] In step Y1 of Figure 5, the target information extraction unit 115 extracts two pieces of target information from each group. In this example, the target information extraction unit 115 extracts the target information with the highest degree of relevance and the target information with the lowest degree of estimation from among the multiple pieces of target information contained in each group. From the group in the left column of Figure 5, "Data A5" is extracted as the target information with the highest degree of relevance, and "Data A1" is extracted as the target information with the lowest degree of relevance. From the group in the center column of Figure 5, "Data A100" is extracted as the target information with the highest degree of relevance, and "Data A110" is extracted as the target information with the lowest degree of relevance. From the group in the right column of Figure 5, "Data B2" is extracted as the target information with the highest degree of relevance, and "Data B1" is extracted as the target information with the lowest degree of relevance.

[0066] Let's return to the explanation in Figure 4. Next, the relevance estimation unit 112 generates a command to re-estimate the relevance of the extracted target information (step S18). Specifically, the relevance estimation unit 112 generates a prompt to re-estimate the relevance of all the target information extracted in step S17. Note that the relevance estimation unit 112 does not group the target information when re-estimating the relevance. That is, all the target information extracted in step S17 is subject to re-estimated relevance.

[0067] The process performed in step S19 is the same as the process performed in step S15, so its explanation is omitted. As a result, the relevance estimation unit 112 obtains the re-estimated relevance for each of the multiple target information extracted in step S17.

[0068] In step Y2 of Figure 5, the re-estimated relevance is estimated by the generating AI system 30 for the target information extracted in step Y1. Specifically, the relevance estimation unit 112 instructs the generating AI system 30 to estimate the re-estimated relevance of the extracted target information, namely "Data A1", "Data A5", "Data A100", "Data A110", "Data B1", and "Data B2". In the example shown in this figure, "(Re-estimated 5 - Expected 6)" is written for "Data A1". This indicates that the re-estimated relevance obtained for "Data A1" is "5".

[0069] Let's return to the explanation in Figure 4. Next, the relevance correction unit 114 corrects the relevance of each target information for each group using the re-estimated relevance of the extracted target information (step S20). That is, the relevance correction unit 114 corrects the relevance using the re-estimated relevance, using one group each composed of multiple target information whose relevance has been estimated as the estimation unit. More specifically, the relevance correction unit 114 performs the correction by normalizing the relevance of target information belonging to the same group using the re-estimated relevance. The relevance correction unit 114 can correct the relevance using known methods. For example, the relevance correction unit 114 uses the piecewise function to correct the relevance of each case that was not extracted by the target information extraction unit 115 based on the re-estimated relevance of the extracted cases.

[0070] For example, the relevance correction unit 114 determines the re-estimated relevance obtained in step S18 as the corrected relevance for the target information extracted in step S17. The relevance correction unit 114 corrects the relevance by scaling the uncorrected relevance of the target information that was not extracted in step S17, based on the relationship between the uncorrected and corrected relevances of the extracted target information. In other words, the relevance correction unit 114 normalizes the relevance of each target information based on a predetermined correction rule obtained for each group.

[0071] As a result, the relevance estimation unit 112 obtains the selection reason and the corrected relevance for each of the selected target information obtained in step S15 from the target information obtained as search targets in step S12. Alternatively, the relevance estimation unit 112 may treat all of the target information obtained as search targets in step S12 as selected and obtain the selection reason and the corrected relevance for each of the target information. Alternatively, the relevance estimation unit 112 may select target information using the corrected relevance obtained for all of the target information obtained as search targets in step S12.

[0072] In step Y3 shown in Figure 5, the relevance estimation device 10 corrects the relevance of each target information for each group using a pre-set correction rule and re-estimated relevance. For example, the relevance estimation device 10 corrects the relevance of each target information by scaling.

[0073] In the group shown in the center column of Figure 5, the re-estimated relevance of "Data A100" is "10," and the re-estimated relevance of "Data A110" is "8." The relevance estimation device 10 treats the relevance of "Data A100" as being the same as the re-estimated relevance of "10," and the relevance of "Data A110" as being the same as the re-estimated relevance of "8." The relevance estimation device 10 corrects the relevance of "Data A105," which is the target information belonging to the group, to "9" by using a correction rule. Note that the description regarding the correction of relevance is omitted for the left and right columns of Figure 7.

[0074] Let's return to the explanation in Figure 4. Next, the response unit 113 generates a response that associates the selection reason with the re-estimated relevance for a predetermined number of target information items with a high relevance after correction (step S21). In addition, the response unit 113 can include the target information selected using the corrected relevance in the response. Specifically, the response unit 113 uses the selection results obtained by the relevance estimation unit 112 in step S15 to generate a response that associates the selection reason with the relevance for a predetermined number of cases with a high relevance. The response unit 113 may also include the selection results received from the generation AI system 30 as is in the response content. As an example, one case included in the response generated by the response unit 113 is as follows.

[0075] Case ID: AC321J Project Name: Visualization of Line Inspection Results Using Belt Conveyors Project Overview: Real-time graph display of work performance management using cameras. Reason for selection: The objective is to improve operational efficiency in the manufacturing industry, and this may be related to increasing the speed of work on parts that can be transported by conveyor belts.

[0076] Next, the response unit 113 outputs a response (step S22). For example, the response unit 113 sends the response generated in step S21 to the target information request device 20 that sent the inquiry information. After that, the processing unit 110 terminates the processing of this flowchart.

[0077] In step Y4 of Figure 5, target information is selected using the corrected relevance score, or from among all search target information included in each group, using the corrected relevance score, and transmitted to the target information request device 20.

[0078] Furthermore, when generating prompts in step S14, the relevance estimation unit 112 may use the case summaries of the target information 121, which have been pre-input by the person in charge of each case, to have the generating AI generate reasons for selection. The case summaries of the target information 121 may also be summaries of cases separately generated by the generating AI, and the relevance estimation unit 112 may use these case summaries to have the generating AI generate reasons for selection.

[0079] Furthermore, in step S18, when the relevance estimation unit 112 causes the generating AI system 30 to estimate the re-estimated relevance, it may generate a prompt to cause the system to estimate the re-estimated relevance using the selection reason and relevance of the case obtained as a result of outputting a command in step S15. This reduces the processing load when causing the system to estimate the re-estimated relevance.

[0080] As described above, according to this embodiment, it is possible to select target information that is highly relevant to the user's search criteria, thereby improving convenience. Furthermore, since the degree of relevance of the selected target information to the user's request is presented to the user, the user can recognize the depth of the relationship between the request and the target information. In addition, since the reason for selecting the target information is displayed, the user can understand why the target information was selected, which not only improves convenience but also increases the reliability of the search results.

[0081] Furthermore, according to this embodiment, even if the target information to be searched exceeds the limit value, which is the upper limit of the amount of data that the generating AI system 30 can process simultaneously, more reliable search results can be obtained. In other words, according to this embodiment, satisfaction when obtaining target information related to an inquiry can be improved.

[0082] <Variation> Figure 6 shows an example of the target information selection process in a modified version of this embodiment. In this modified version, the relevance estimation device 10 performs multi-stage processing during relevance normalization by determining whether the extracted target information exceeds a limit value when causing the generation AI system 30 to estimate the re-estimated relevance. The differences from the above-described embodiment will be explained below.

[0083] The processing in steps 1 to 17 is the same as in steps 1 to 17 in Figure 5.

[0084] As described above, in step S16, the relevance estimation unit 112 obtains a limit value and uses the limit value and the number of groups to determine the number of target information that can be extracted per group. In this modified example, a lower limit may be set for the number of extractable data. For example, the number of extractable data may be set in advance to "2" or more. In that case, if the calculated number of extractable data is "less than 2", the relevance estimation unit 112 can determine the number of extractable data to be the lower limit of "2".

[0085] Next, the relevance estimation unit 112 determines whether the total number of extracted target information exceeds a limit (step S171). Specifically, the relevance estimation unit 112 obtains the total number of extracted target information by multiplying the number of target information for each group, based on the number of extractable data extracted in step S17, by the number of groups obtained by division in step S13. The relevance extraction unit determines whether the total number of extracted target information exceeds a limit.

[0086] If the relevance estimation unit 112 determines that the total number of extracted target information exceeds the limit value (if the answer is "YES" in step S171), it proceeds to step S13. Accordingly, the relevance estimation unit 112 further divides the target information extracted in step S17 into groups so that the total number of extracted target information remains within the limit value. As a result, the relevance estimation unit 112 repeats the division of target information until the total number of target information extracted by the target information extraction unit 115 is less than or equal to the limit value.

[0087] If the relevance estimation unit 112 determines that the total number of extracted target information is less than or equal to the limit value (i.e., "NO" in step S171), it proceeds to step S18. Therefore, even if the total number of target information in step S171 becomes less than or equal to the limit value as a result of grouping the target information multiple times (i.e., multiple stages) in step S13, the relevance estimation unit 112 also executes the process in step S18. That is, for each of the target information included in the group in which the total number of target information is less than or equal to the limit value, the relevance estimation unit 112 generates a prompt to cause the relevance generation AI system 30 to re-estimate the degree of relevance.

[0088] If the target information is grouped multiple times, the relevance correction unit 114 performs a multi-stage process in step S20 to correct the relevance using the re-estimated relevance. For example, if the division in step S13 is performed three times, in step S19, the relevance estimation unit 112 obtains the re-estimated relevance of the target information extracted from the group obtained as a result of the third division. The relevance correction unit 114 uses the re-estimated relevance to correct the relevance of the target information included in the group obtained after the third division. This group includes the target information extracted from the group obtained as a result of the second division.

[0089] The relevance correction unit 114 corrects the relevance of the target information included in the group using the corrected relevance of the target information extracted from the group obtained as a result of the second division. By repeating this process, the relevance correction unit 114 obtains normalized relevance for the target information acquired in step S12.

[0090] As described above, this modified version allows for obtaining a normalized relevance score regardless of the amount of data in the target information. Therefore, according to the relevance estimation system 1 in this modified version, it is possible to obtain highly reliable selection results even when the amount of data in the target information is large.

[0091] Although the embodiments of the present invention have been described above, the present invention is not limited to the examples of embodiments described above, and various modifications are included. For example, the examples of embodiments described above are explained in detail to make the present invention easier to understand, and the present invention is not limited to having all the configurations described herein. Furthermore, it is possible to replace a part of the configuration of one example of an embodiment with the configuration of another example. It is also possible to add a configuration of another example to the configuration of one example of an embodiment. Furthermore, it is possible to add, delete, or replace a part of the configuration of one example of each embodiment with a configuration of another example. In addition, some or all of the above configurations, functions, processing units, processing means, etc., may be realized in hardware, for example, by designing them as integrated circuits. Also, the control lines and information lines in the figures are shown only if they are considered necessary for explanation, and do not necessarily show all of them. It can be assumed that almost all of the configurations are interconnected.

[0092] Furthermore, the functional configurations of the relevance estimation device 10, the target information request device 20, and the generation AI system 30 described above are classified according to their main processing content for ease of understanding. The present invention is not limited by the way the components are classified or named. As shown above, the configurations of the relevance estimation device 10, the target information request device 20, and the generation AI system 30 can be further classified into many more components depending on the processing content. Alternatively, they can be classified so that a single component performs even more processing.

[0093] For example, the relevance estimation device 10 may function as a target information request device 20. In that case, the relevance estimation device 10 can cause the generating AI system 30 to select target information in response to search information transmitted from the user's terminal device. [Explanation of Symbols]

[0094] 1: Relevance estimation system, 10: Relevance estimation device, 20: Target information request device, 30: Generation AI system, 110-210: Processing unit, 111: Inquiry acquisition unit, 112: Relevance estimation unit, 113: Response unit, 114: Relevance correction unit, 115: Target information extraction unit, 116: Target information acquisition unit, 120-220: Storage unit, 121: Target information, 130-230: Input unit, 140-240: Output unit, 150-250: Communication unit, 211: Inquiry management unit, 101: Processor, 102: Memory, 103: Storage, 104: Communication device, 105: Bus

Claims

1. An inquiry acquisition unit that acquires inquiry information related to the inquiry, A relevance estimation unit causes the generating AI to estimate the degree of relevance between each of the one or more target information related to the aforementioned inquiry and the aforementioned inquiry. An information processing device comprising: a relevance correction unit that corrects the degree of relevance of the generated AI according to the estimation unit of the target information for which the degree of relevance has been estimated.

2. An information processing apparatus according to claim 1, The information processing device is characterized in that the correlation correction unit corrects the correlation using a plurality of groups, each composed of a plurality of the target information whose correlation has been estimated, as the estimation unit.

3. An information processing apparatus according to claim 2, Each of the aforementioned multiple groups is provided with a target information extraction unit that extracts multiple pieces of the aforementioned target information, The correlation estimation unit has the generating AI re-estimate the correlation for each of the multiple pieces of target information extracted by the target information extraction unit to obtain a re-estimated correlation. The information processing device is characterized in that the correlation correction unit performs correction using the re-estimated correlation.

4. An information processing apparatus according to claim 3, The information processing device is characterized in that the correlation correction unit performs a correction by normalizing the correlation of other subject information belonging to the same group using the re-estimated correlation of the subject information.

5. An information processing apparatus according to claim 1, The system includes a target information acquisition unit that acquires document information relating to past cases as the target information, The information processing device is characterized in that the relevance estimation unit causes the generating AI to estimate the degree of relevance for a plurality of document pieces of information related to the query, as the target information.

6. An information processing apparatus according to claim 3, The correlation estimation unit obtains a limit value, which is the maximum amount of information per processing, and determines the number of extractable data using the limit value and the number of groups. The information processing device is characterized in that the target information extraction unit extracts the number of target information data that can be extracted for each of the plurality of groups.

7. An information processing apparatus according to claim 6, The correlation estimation unit divides the extracted target information into groups if the total number of target information extracted by the target information extraction unit exceeds the limit value. The target information extraction unit extracts multiple pieces of the target information for each group, The information processing device is characterized in that the correlation estimation unit repeats the division until the total number of target information extracted by the target information extraction unit is less than or equal to the limit value, and for each of the target information included in the group in which the total number of target information is less than or equal to the limit value, the generating AI re-estimates the correlation to obtain a re-estimated correlation.

8. An information processing apparatus according to claim 3, It is equipped with a target information acquisition unit that acquires document information related to past cases, The aforementioned relevance estimation unit, The generating AI is instructed to select the document information, and for each selected document, it is instructed to generate a reason for selection and to estimate the degree of relevance. An information processing device characterized by causing a generating AI to estimate the re-estimation of the relevance using the selection reason and the relevance.

9. A computer-based relevance estimation system, The aforementioned computer, The inquiry retrieval step involves obtaining inquiry information related to the inquiry, A relevance estimation step in which the generating AI estimates the degree of relevance of each of the multiple pieces of target information related to the aforementioned inquiry to the aforementioned inquiry, A relevance estimation system characterized by performing a relevance correction step in which the relevance of the generated AI is corrected according to the estimation unit of the target information for which the relevance has been estimated.

10. A method for estimating relevance performed by an information processing device, The procedure for obtaining inquiry information about an inquiry, A correlation estimation procedure that causes the generating AI to estimate the degree of correlation between each of the multiple pieces of information related to the aforementioned inquiry and the aforementioned inquiry, A method for estimating relevance, characterized by comprising: a relevance correction procedure for correcting the relevance of the generated AI according to the estimation unit of the target information for which the relevance has been estimated.

11. A program that causes a computer's processing unit to execute a correlation estimation method, The procedure for obtaining inquiry information about an inquiry, A correlation estimation procedure that causes the generating AI to estimate the degree of correlation between each of the multiple pieces of information related to the aforementioned inquiry and the aforementioned inquiry, A program characterized by causing a generating AI to execute a relevance correction procedure that corrects the relevance according to the estimation unit of the target information for which the relevance has been estimated.

Images