Estimation system, estimation method, and program

Abstract

To improve user convenience. [Solution] The inquiry information acquisition unit (101) of the estimation system (1) acquires inquiry information related to the inquiry entered by the user. The profile information acquisition unit (102) acquires profile information related to each of the profiles of multiple inquiry destinations. Based on the inquiry information and profile information, the estimation unit (103) has the AI ​​(Artificial Intelligence), which is a trained large-scale language model, estimate the appropriate inquiry destination from among the multiple inquiry destinations.

Description

Technical Field

[0001] The present disclosure relates to an estimation system, an estimation method, and a program.

Background Art

[0002] Conventionally, a technique for assisting a user who makes an inquiry to an inquiry destination has been known. For example, in Non-Patent Document 1, in a business support service in which messages such as inquiries can be exchanged between users, when a user tries to mention another user by inputting a specific symbol (for example, @), a chat that displays a list of other users who are candidates for the mention destination, which is an example of the inquiry destination, is described.

Prior Art Documents

Non-Patent Documents

[0003]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in the technique of Non-Patent Document 1, only predetermined other users are displayed as candidates, and appropriate candidates corresponding to the user's inquiry are not displayed. Therefore, with the technique of Non-Patent Document 1, the user has to think and select appropriate candidates corresponding to their own inquiry by themselves, so the convenience of the user cannot be sufficiently improved.

[0005] One of the objects of the present disclosure is to improve the convenience of the user.

Means for Solving the Problems

[0006] An estimation system relating to one aspect of this disclosure includes: an inquiry information acquisition unit that acquires inquiry information relating to an inquiry entered by a user; a profile information acquisition unit that acquires profile information relating to each of the profiles of a plurality of inquiry recipients; and an estimation unit that causes an AI (Artificial Intelligence), which is a trained large-scale language model, to estimate the appropriate inquiry recipient from among the plurality of inquiry recipients relating to the inquiry, based on the inquiry information and the profile information. [Effects of the Invention]

[0007] This disclosure can improve user convenience. [Brief explanation of the drawing]

[0008] [Figure 1] This figure shows an example of the hardware configuration of the estimation system. [Figure 2] This figure shows an example of a business support screen. [Figure 3] This figure shows an example of the functions implemented by the estimation system. [Figure 4] This figure shows an example of a comment database. [Figure 5] This figure shows an example of a profile database. [Figure 6] This figure shows an example of AI input and output. [Figure 7] This figure shows an example of the process performed by the estimation system. [Figure 8] This figure shows an example of the functionality implemented by the modified form estimation system. [Figure 9] This figure shows an example of a profile database for the first modified example. [Figure 10] This figure shows an example of the AI ​​input and output in Modification Example 1. [Figure 11] This figure shows an example of a profile database in modified example 3. [Figure 12]This is a diagram showing an example of the input and output of the AI in Modification 3. [Figure 13] This is a diagram showing an example of the input and output of the AI in Modification 4. [Figure 14] This is a diagram showing an example of the business support screen in Modification 5. [Figure 15] This is a diagram showing an example of the business support screen in Modification 6. [Figure 16] This is a diagram showing an example of the business support screen in Modification 7. [Figure 17] This is a diagram showing an example of the business support screen in Modification 8. [Figure 18] This is a diagram showing an example of the input and output of the AI in Modification 9.

Embodiments for Carrying out the Invention

[0009] [1. Hardware Configuration] An example of an embodiment of the estimation system, estimation method, and program according to the present disclosure will be described. FIG. 1 is a diagram showing an example of the hardware configuration of the estimation system. For example, the estimation system 1 includes a server 10 and a user terminal 20. Each of the server 10 and the user terminal 20 is connected to a network N such as the Internet or a LAN. In FIG. 1, each of the server 10 and the user terminal 20 is shown as one unit, but at least one of the server 10 and the user terminal 20 may exist in multiple units.

[0010] The server 10 is a server computer. For example, the server 10 includes a control unit 11, a storage unit 12, and a communication unit 13. The control unit 11 includes at least one processor. The storage unit 12 includes at least one of a volatile memory such as a RAM and a non-volatile memory such as a flash memory. The communication unit 13 includes at least one of a communication interface for wired communication and a communication interface for wireless communication.

[0011] The user terminal 20 is the user's computer. For example, the user terminal 20 is a personal computer, a tablet terminal, a smartphone, or a wearable terminal. For example, the user terminal 20 includes a control unit 21, a storage unit 22, a communication unit 23, an operation unit 24, and a display unit 25. The hardware configurations of the control unit 21, the storage unit 22, and the communication unit 23 may be the same as those of the control unit 11, the storage unit 12, and the communication unit 13, respectively. The operation unit 24 is an input device such as a mouse or a touch panel. The display unit 25 is a liquid crystal or organic EL display.

[0012] Note that the programs stored in the storage units 12 and 22 may be supplied via the network N. The hardware configurations of each of the server 10 and the user terminal 20 are not limited to the example in FIG. 1. For example, at least one of the server 10 and the user terminal 20 may include at least one of a reading unit (for example, a memory card slot) that reads a computer-readable information storage medium and an input / output unit (for example, a USB terminal) for directly connecting to an external device. The program stored in the information storage medium may be supplied to at least one of the server 10 and the user terminal 20 via at least one of the reading unit and the input / output unit.

[0013] Also, the estimation system 1 may include at least one computer. The computer included in the estimation system 1 is not limited to the example in FIG. 1. For example, the estimation system 1 may include only the server 10. In this case, the user terminal 20 exists outside the estimation system 1. The estimation system 1 may include the server 10 and another computer not shown in FIG. 1.

[0014] [2. Outline of the Estimation System] In this embodiment, we will take the example of when the estimation system 1 is applied to a business support service. The business support service is a service that supports the user's work. The business support service may be a service called groupware, or it may be a service that is not specifically classified as groupware. The business support service may be cloud-based or on-premise. For example, when a user logs in to the business support service, the user terminal 20 displays a business support screen on the display unit 25 to support the user's work.

[0015] Figure 2 shows an example of a business support screen. Figure 2 shows an example of a business support screen SC with a communication function for users to communicate with each other. For example, as shown in the upper part of Figure 2, the business support screen SC displays the content of a thread where various comments have been posted. Users communicate with other users by entering new comments in the input form F or by replying to comments made by other users. Users may communicate with other users belonging to their own organization (e.g., company) or with other users outside of their organization.

[0016] In this embodiment, we take the example of a user making an inquiry about their work as a comment posted to a thread. For example, suppose a user in charge of web content work wants to know information about HTML. In this case, the user enters an inquiry about HTML in order to get information about HTML from other users. However, if the user does not know who the other users who are knowledgeable about HTML are, they may not know who the appropriate person to ask is.

[0017] Therefore, in this embodiment, the AI ​​(Artificial Intelligence) estimates the appropriate contact person for the inquiry entered by the user. Details of the AI ​​will be described later. In the upper example of Figure 2, the user mentions the AI, such as "@AI", and enters an inquiry such as "Please tell me who is knowledgeable about HTML." After entering the inquiry text, the user can complete the input of the inquiry by performing a predetermined operation (for example, selecting a parse on the user interface to indicate completion of input, or pressing a predetermined key). Since this inquiry is about HTML, the appropriate contact person is another user who is knowledgeable about HTML. The AI ​​estimates another user who is knowledgeable about HTML as the appropriate contact person. Details of the process by which the AI ​​estimates the appropriate contact person will be described later.

[0018] For example, when the AI ​​estimates an appropriate contact point, the user terminal 20 displays the AI-estimated contact point on the business support screen SC, as shown in the lower part of Figure 2. In the example at the bottom of Figure 2, the AI's response comment indicates the AI's estimated appropriate contact point. For example, the AI ​​estimates other users who are knowledgeable about HTML based on comments previously posted to the thread. The user then uses the AI's response as a reference to mention other users knowledgeable about HTML and make an inquiry about HTML.

[0019] As described above, the estimation system 1 of this embodiment allows the AI ​​to estimate the appropriate contact point for the inquiry entered by the user. The estimation system 1 displays the appropriate contact point estimated by the AI ​​on the business support screen SC. This allows the user to know the appropriate contact point for handling their inquiry. Since the user does not need to find the appropriate contact point themselves, the estimation system 1 can improve user convenience. The details of the estimation system 1 will be described below.

[0020] [3. Functions implemented by the estimation system] Figure 3 shows an example of the functions implemented by the estimation system 1. In Figure 3, the functions implemented by the server 10 are shown among the functions implemented by the estimation system 1. For example, the server 10 includes a data storage unit 100, a query information acquisition unit 101, a profile information acquisition unit 102, and an estimation unit 103. Each of the data storage units 100 is implemented by a storage unit 12. Each of the query information acquisition unit 101, the profile information acquisition unit 102, and the estimation unit 103 is implemented by a control unit 11.

[0021] [Data Storage Unit] The data storage unit 100 stores data necessary to support business operations. For example, the data storage unit 100 stores the comment database DB1 and the profile database DB2.

[0022] Figure 4 shows an example of the comment database DB1. The comment database DB1 is a database that stores various information about comments posted on the business support service. For example, the comment database DB1 stores a thread ID that can identify a thread, a comment ID that can identify a comment, the user ID of the user who posted the comment, the content of the comment (e.g., letters, numbers, other symbols, attachments, or combinations thereof), information about the person mentioned in the comment, and the posting date and time. Note that the information stored in the comment database DB1 is not limited to the example in Figure 4. The comment database DB1 may store any information related to comments.

[0023] For example, when a user creates a new thread, server 10 issues a thread ID for that thread. When a new comment is posted to that thread, server 10 issues a comment ID for that new comment. Server 10 stores the user ID of the user who posted the new comment, the content of the new comment, the information of the people mentioned in the new comment, and the posting date and time in the comment database DB1, associating them with the thread ID and comment ID. Similarly, when a response from AI is input as a comment, server 10 also stores information such as the comment ID of the AI's comment in the comment database DB1.

[0024] Figure 5 shows an example of a profile database DB2. The profile database DB2 is a database that stores profile information for each of multiple query targets. For example, the profile database DB2 stores a query target ID that can identify the query target, and profile information. The profile database DB2 may store any data. The data stored in the profile database DB2 is not limited to the example in Figure 5. For example, the profile database DB2 may store information that serves as an index when searching in RAG (e.g., keywords that indicate the characteristics of the profile information, department name, team name, project name, job title, or work location), the name of the query target (e.g., the user's name), or the date and time the profile information was created.

[0025] The contact person's profile refers to the characteristics or attributes of the contact person. Profile information is information that describes the contact person's profile. Profile information can be in any format. For example, profile information may consist of letters, numbers, other symbols, or a combination thereof. Profile information may indicate the areas in which the contact person is knowledgeable or skilled. Profile information may also indicate the types of inquiries the contact person can handle.

[0026] In this embodiment, since each of the multiple users utilizing the business support service can be a contact point, the contact ID is the user ID. The contact profile is the user's profile. Profile information indicates the user's profile. That is, profile information is the user's characteristics or attributes. For example, profile information may indicate the content of the work the user is responsible for, the areas the user is knowledgeable or skilled in, the user's years of service, the user's career path, the user's demographic information, the self-introduction entered by the user, or a combination of these.

[0027] For example, server 10 generates profile information based on comment database DB1 and stores it in profile database DB2. For each user, server 10 retrieves the comments posted by that user from comment database DB1 and generates the user's profile information based on those comments. Server 10 may use the comments posted by the user as is for the user's profile information, or it may vectorize the comments posted by the user and use the vector as the profile information. Figure 5 shows an example where the profile information is in vector format.

[0028] A vector is information that represents the characteristics of words. Vectorization may be performed using known methods employed in the field of natural language processing. For example, server 10 may vectorize comments based on Doc2Vec, FastText, transformer-based models, or other methods. Server 10 may generate a single vector from multiple comments posted by a user, or it may generate a vector for each individual comment. Server 10 may also select only a portion of multiple comments posted by a user and generate vectors from those.

[0029] Furthermore, profile information may indicate not only the content of the profile but also which contact the profile belongs to. For example, if the contact is another user, the profile information may include information indicating who the other user is (e.g., the other user's name or user ID) and the specific content of the profile (e.g., information in vector format).

[0030] Furthermore, the data stored in the data storage unit 100 is not limited to the above example. The data storage unit 100 can store any data. In this embodiment, we take the example of a case where the AI ​​is managed by an external system located outside the estimation system 1, but when the estimation system 1 manages the AI, the data storage unit 100 may store the actual data of the AI. That is, when the estimation system 1 uses the AI ​​in a local environment where no communication with an external system occurs, the data storage unit 100 may store the actual data of the AI.

[0031] In this embodiment, we take the example of a pre-trained large-scale language model corresponding to the AI. The large-scale language model may be of any type. For example, the large-scale language model may be a transformer-type model such as GPT (Generative Pre-trained Transformer) or BERT (Bidirectional Encoder Representations from Transformers), or another model that is not classified as a transformer (e.g., a neural network). The AI ​​may be an AI created for estimation system 1, but in this embodiment, it is assumed to be a known AI.

[0032] For example, an AI includes a program that shows a series of processes from the input information, which is the information that is input to the AI, until an output is produced, and parameters that are referenced by the program. The parameters may be incorporated as part of the AI's program. The AI's parameters may be similar to publicly known parameters. For example, the AI's parameters may be weights and biases. The AI ​​may include other information besides the program and parameters. The AI ​​may be a single-modal type AI capable of processing only one particular form of input information, or a multimodal type AI capable of processing multiple forms of input information.

[0033] For example, an AI may include an input layer that processes the input information, an intermediate layer that computes an embedding representation of the input information based on parameters, and an output layer that produces an output corresponding to the embedding representation. The embedding representation is the information that the AI ​​uses to recognize the input information. Embedding representations are sometimes called features or feature vectors. Embedding representations may be multidimensional vectors or other forms (e.g., arrays, matrices, single numbers, or multiple numbers). The input information may be divided into units called tokens. The AI ​​may compute an embedding representation for each individual token. The AI ​​may predict the rest of the input information based on the order of the embedding representations of the individual tokens.

[0034] In this embodiment, it is assumed that various training data have already been trained on the AI. If the AI ​​is a model capable of natural language processing, such as a large-scale language model, then it is assumed that the AI ​​has already been trained on various texts as training data. The AI's parameters have been adjusted through training on the training data. A pre-trained general-purpose AI may be used as is, or an AI fine-tuned for estimation system 1 may be used. The AI's training method may be the same as known machine learning techniques. Similar to the definition of AI, there are various definitions of machine learning, but machine learning in this embodiment includes various known definitions. For example, deep learning is also included in machine learning.

[0035] For example, the data storage unit 100 may store default prompts that are input to the AI. Default prompts are pre-prepared prompts. The data storage unit 100 may also store data from the screen displayed on the user terminal 20. The data storage unit 100 may store a user database containing various information about each of multiple users in the business support service (for example, email addresses to which mention notifications are sent). The data storage unit 100 may also store a database containing various information about notifications in the business support service (for example, notifications received by the mentioned user).

[0036] [Inquiry Information Acquisition Department] The inquiry information acquisition unit 101 acquires inquiry information related to the inquiry entered by the user. An inquiry is a question that requires an answer. Inquiry information is information that indicates the content of the inquiry entered by the user. For example, inquiry information may be characters, numbers, other symbols, files such as documents or images, or a combination thereof. For example, the user may enter the inquiry from the operation unit 24, or may enter the inquiry by voice input or data input to the user terminal 20.

[0037] In this embodiment, since the estimation system 1 is applied to the business support service, the inquiry information acquisition unit 101 acquires inquiry information when a user of the business support service enters an inquiry related to business. An inquiry related to business is an inquiry about the content of the business supported by the business support service. For example, an inquiry related to business may be about how to proceed with the work, basic knowledge required for the work, information about the organization to which the user belongs, information about external organizations related to the work, or other inquiries.

[0038] In the example shown in Figure 2, the user enters an inquiry into the input form F on the business support screen SC. The user terminal 20 sends inquiry information related to the inquiry entered by the user to the server 10. The inquiry information acquisition unit 101 acquires the inquiry information from the user terminal 20. The inquiry information acquisition unit 101 may also acquire inquiry information from other computers besides the user terminal 20. For example, if there is another computer that stores inquiry information, the inquiry information acquisition unit 101 may acquire inquiry information from that other computer.

[0039] Furthermore, the screen on which users enter inquiries is not limited to the business support screen SC. Users may enter inquiries from screens other than the business support screen SC. For example, users may enter inquiries from bulletin boards not classified as threads, chats, comment sections of database functions, comment sections of email sharing functions, comment sections of schedule management functions, screens in other business support functions, or other screens. The inquiry information acquisition unit 101 may acquire inquiry information related to inquiries entered by users on these screens.

[0040] Furthermore, users may enter inquiries in business support functions other than the communication function within the business support service. For example, users may enter inquiries in the schedule management function for managing schedules, the database function for managing data, or the email management function for managing emails. The inquiry information acquisition unit 101 may acquire inquiry information of inquiries entered by the user in other business support functions.

[0041] [Profile Information Acquisition Unit] The profile information acquisition unit 102 acquires profile information for each of the multiple contacts. The contacts are the recipients to whom the user's inquiries are sent. For example, the contacts may be the person who answers the user's inquiries (e.g., another user, organization, department, or team), the communication location where communication takes place (e.g., a thread, a bulletin board not classified as a thread, a chat room, a space, or a channel), or the recipient to whom the user's inquiries are sent (e.g., an email address, a messaging app account, or a phone number).

[0042] In this embodiment, since the profile information is stored in the profile database DB2, the profile information acquisition unit 102 acquires the profile information for each of the multiple query targets from the profile database DB2. The profile information acquisition unit 102 may acquire all the profile information stored in the profile database DB2. The profile information acquisition unit 102 may randomly acquire some of the profile information from the profile database DB2. The profile information may be stored in a database other than the profile database DB2, a computer other than the server 10, or an information storage medium. The profile information acquisition unit 102 may acquire profile information from a database other than the profile database DB2, a computer other than the server 10, or an information storage medium.

[0043] In this embodiment, we take an example where the profile information acquisition unit 102 acquires at least one profile information by searching the profile database DB2 using the RAG (Retrieval Augmented Generation) mechanism. For example, the profile information acquisition unit 102 searches the profile database DB2 based on the query information. The search method may be the same as a known method.

[0044] For example, the profile information acquisition unit 102 may vectorize the query information and retrieve at least one profile information by searching the profile database DB2 based on the vector of the query information. In this case, the profile information acquisition unit 102 retrieves profile information that is relatively close to the vector of the query information in the vector space.

[0045] For example, the profile information acquisition unit 102 may acquire keywords from the query information and search the profile database DB2 based on those keywords. The method for acquiring keywords may also be the same as known methods. For example, the profile information acquisition unit 102 may acquire keywords by performing morphological analysis on the query information. When keywords are used, it is assumed that keywords that will serve as an index during the search are associated with the profile information in advance.

[0046] In this embodiment, since the estimation system 1 is applied to the business support service, the profile information acquisition unit 102 acquires profile information based on the business support service. Profile information based on the business support service is profile information created based on data registered in the business support service. For example, profile information created based on comment data stored in the comment database DB1 corresponds to profile information based on the business support service.

[0047] Furthermore, profile information based on the business support service may be created based on data stored in databases other than the comment database DB1. For example, profile information based on the business support service may be created based on a database containing posts registered on bulletin boards, chats, or other communication tools that are not classified as threads.

[0048] For example, profile information based on business support services may be created based on a database managed by the database function. Profile information based on business support services may be created based on a database of schedules registered in the schedule management function. Profile information based on business support services may be created based on a database of shared emails registered in the shared email function.

[0049] In this embodiment, we take the example of a case where multiple contacts are multiple other users different from the user. Therefore, the profile information acquisition unit 102 acquires profile information for each of the multiple other users' profiles. The profile information indicates the characteristics or attributes of the other users. For example, the profile information may indicate the content of the work that the other users are in charge of, or it may indicate the fields that the other users are knowledgeable about or specialize in.

[0050] Furthermore, profile information may be generated based on information registered in services other than the business support service. For example, if the business support service is linked with an SNS (Social Networking Service), profile information may be generated based on posts on the SNS of various users who use the business support service. The profile information acquisition unit 102 may acquire profile information generated based on information registered in other services.

[0051] Furthermore, the profile information acquisition unit 102 can acquire profile information at any time. For example, the profile information acquisition unit 102 acquires profile information when AI estimation is performed. The profile information may not be generated in advance, but may be generated on the spot when AI estimation is performed. The profile information may be generated in advance and then adjusted by the administrator. The profile information may not be generated based on comments, but may use information registered in the business support service (for example, employee rosters, personnel documents, or evaluation information) as is.

[0052] Furthermore, the profile information acquisition unit 102 may acquire profile information before the user enters an inquiry. The server 10 may pre-train the AI ​​with the profile information acquired by the profile information acquisition unit 102. For example, the server 10 may retrain (e.g., fine-tune) an AI that has been trained using documents other than profile information as training data, based on the profile information. That is, the server 10 may fine-tune the AI's parameters based on the profile information. The AI ​​retraining process may be performed by the estimation unit 103, which will be described later.

[0053] [Estimation part] The estimation unit 103 causes the AI, a pre-trained large-scale language model, to estimate the appropriate query destination from among multiple query destinations based on the query information and profile information. For example, the estimation unit 103 inputs the query information and profile information as input information to the AI. The estimation unit 103 may input the query information and profile information together into one input to the AI, or it may input the query information and profile information separately without combining them into one input to the AI.

[0054] In this embodiment, we take the example of a case where the AI ​​is managed by an external system, so the estimation unit 103 inputs query information and profile information to the AI ​​by transmitting query information and profile information to the external system. The estimation unit 103 obtains information from the external system indicating the appropriate query destination estimated by the AI. If the AI ​​is stored in the data storage unit 100, the estimation unit 103 only needs to input query information and profile information to the AI ​​stored in the data storage unit 100. The estimation unit 103 only needs to obtain information indicating the appropriate query destination output from the AI ​​stored in the data storage unit 100.

[0055] Figure 6 shows an example of AI input and output. In this embodiment, the estimation unit 103 inputs a default prompt, query information, and profile information to the AI ​​as input information. The default prompt indicates the AI's task. In other words, the default prompt indicates what output the AI ​​should produce based on the input information. The default prompt is written in natural language. At least one of the query information and profile information may be incorporated into the default prompt. The default prompt may indicate what each of the query information and profile information is. Instead of a default prompt, a prompt entered by the user may be input to the AI.

[0056] For example, the default prompt includes natural language text indicating that the system will estimate the appropriate contact based on the inquiry information and profile information. The default prompt may be prepared by the administrator of Estimation System 1 or by the administrator of the organization to which the user belongs. In the example in Figure 6, the default prompt shows text such as, "You are the AI ​​that estimates the appropriate contact. Please estimate the appropriate contact based on the inquiry information and profile information you have entered." The default prompt can be any text. The default prompt is not limited to the example in Figure 6.

[0057] For example, when a default prompt, query information, and profile information are input to the AI, the AI ​​calculates embedded representations of the default prompt, query information, and profile information based on pre-adjusted parameters. The AI ​​may also divide the default prompt, query information, and profile information into units called tokens and then calculate the embedded representation of each token. The AI ​​outputs information indicating the appropriate query destination according to the embedded representation. Based on the embedded representation of the default prompt, the AI ​​can predict the continuation of the sentence as needed and estimate the appropriate query destination according to the query information and profile information. The estimation unit 103 acquires the information indicating the appropriate query destination output by the AI.

[0058] In the example in Figure 6, the profile information shows the name of another user and the profile of that other user. For example, the AI ​​estimates that user U5 is the appropriate contact by predicting the next sentence as needed based on the default prompt, inquiry information, and embedded representations of the profile information. The AI ​​outputs a sentence indicating that user U5 is the appropriate contact. The estimation unit 103 acquires the sentence output by the AI. The estimation unit 103 sends the data of the sentence output by the AI ​​to the user terminal 20. As shown in the lower part of Figure 2, the user terminal 20 displays the sentence on the business support screen SC.

[0059] In this embodiment, since the inquiry information and profile information are information related to the business support service, the estimation unit 103 causes the AI ​​to estimate the appropriate contact point for the business support service. The appropriate contact point for the business support service may be a user, organization, department, team, communication location (e.g., thread, bulletin board not classified as a thread, chat room, space, or channel), email address, or other information. The AI ​​estimates the appropriate contact point for the business support service as the appropriate contact point according to the embedded representation.

[0060] In this embodiment, the estimation unit 103 causes the AI ​​to estimate an appropriate other user from among a plurality of other users as the appropriate query destination. The plurality of other users that make up the population may belong to the same organization as the user, or they may be persons who have no particular relationship with the organization to which the user belongs. The AI ​​estimates at least one other user. If no appropriate other user exists, the AI ​​may output information indicating that no appropriate other user was estimated. The AI ​​estimates an appropriate other user according to the embedded representation.

[0061] For example, if profile information is generated for each department, the AI ​​estimates the department as the appropriate contact point. The estimation unit 103 sends data indicating the department estimated by the AI ​​to the user terminal 20. The user terminal 20 displays the department, rather than a specific individual, as the appropriate contact point on the business support screen SC. If profile information is generated for each communication location (e.g., thread), the AI ​​estimates the location as the appropriate contact point. The estimation unit 103 sends data indicating the communication location estimated by the AI ​​to the user terminal 20. The user terminal 20 displays the communication location, rather than a specific individual, as the appropriate contact point on the business support screen SC.

[0062] In this embodiment, we take as an example the case in which estimation by the estimation unit 103 is performed when the user mentions AI, such as "@AI". For example, when the user completes inputting the inquiry, the estimation unit 103 determines whether or not a specific string such as "@AI" is included in the inquiry information. The specific string is a string that indicates a mention of AI. If the estimation unit 103 determines that the specific string is not included in the inquiry information, it does not perform estimation by AI. If the estimation unit 103 determines that the specific string is included in the inquiry information, it performs estimation by AI.

[0063] Furthermore, the estimation unit 103 may perform AI-based estimation even if the user does not specifically mention AI. For example, if a user interface part for requesting AI-based estimation is displayed on the business support screen SC, the estimation unit 103 may perform AI-based estimation when that part is selected. The estimation unit 103 may also perform AI-based estimation without receiving explicit instructions from the user. For example, the estimation unit 103 may perform AI-based estimation when the user enters any inquiry, even if the user does not specifically mention AI.

[0064] [4. Processes executed by the estimation system] Figure 7 shows an example of the processing performed by the estimation system 1. The processing in Figure 7 is performed by the control units 11 and 21 executing programs stored in the storage units 12 and 22, respectively. It is assumed that, in order for the processing in Figure 7 to be performed, the server 10 generates profile information based on the comment database DB1.

[0065] As shown in Figure 7, the server 10 performs a login process with the user terminal 20 to allow the user to log in to the business support service (S1). When the user selects a thread of the business support service, the server 10 performs a process with the user terminal 20 to display the business support screen SC showing the thread selected by the user (S2). The following describes the process when the user mentions the AI ​​from the business support screen SC to have the AI ​​estimate the target of the inquiry.

[0066] The user terminal 20 sends query information to the server 10 based on the input to input form F (S3). The server 10 retrieves the query information from the user terminal 20 (S4). The server 10 retrieves profile information from the profile database DB2 (S5). In S5, the server 10 vectorizes the query information and searches for the profile information. Note that the server 10 may also retrieve the profile information from the profile database DB2 without using the RAG mechanism.

[0067] Server 10 causes the AI ​​to estimate an appropriate contact point based on the inquiry information obtained in S4 and the profile information obtained in S5 (S6). In S6, Server 10 inputs the inquiry information and profile information to the AI ​​along with the default prompt stored in the memory unit 12. Server 10 obtains information indicating the appropriate contact point output by the AI. Server 10 performs a process to display the appropriate contact point estimated by the AI ​​on the business support screen SC in communication with the user terminal 20 (S7), and this process ends. In S7, Server 10 displays the appropriate contact point estimated by the AI ​​on the business support screen SC.

[0068] [5. Summary of Embodiments] The estimation system 1 of this embodiment acquires inquiry information. The estimation system 1 acquires profile information for each of the multiple inquiry destinations. Based on the inquiry information and profile information, the estimation system 1 has the AI ​​estimate the appropriate inquiry destination from among the multiple inquiry destinations. This allows the user to know the appropriate inquiry destination for their inquiry. Since the user does not need to find the appropriate inquiry destination themselves, the estimation system 1 can improve user convenience.

[0069] Furthermore, Estimation System 1 acquires inquiry information when a user of the business support service enters an inquiry related to their work. Estimation System 1 acquires profile information based on the business support service. Estimation System 1 has the AI ​​estimate the appropriate contact point for inquiries within the business support service. This allows users to know the appropriate contact point for handling inquiries related to their work. Since users no longer need to research the appropriate contact point themselves, Estimation System 1 can improve user convenience.

[0070] Furthermore, the multiple contact points are multiple other users different from the user. Estimation System 1 obtains profile information about each of the multiple other users' profiles. Estimation System 1 has the AI ​​estimate which of the multiple other users is appropriate for the inquiry. This allows the user to know which other user is appropriate to handle their inquiry. Since the user does not need to find the appropriate other user themselves, Estimation System 1 can improve user convenience.

[0071] [6. Variant] This disclosure is not limited to the embodiments described above. This disclosure may be modified as appropriate without departing from the spirit of this disclosure.

[0072] Figure 8 shows an example of the functions implemented in the modified estimation system 1. As shown in Figure 8, in the modified version described later, the server 10 implements a user profile information acquisition unit 104, a mention execution unit 105, a mention target presentation unit 106, and an accuracy acquisition unit 107. Each of the user profile information acquisition unit 104, the mention execution unit 105, the mention target presentation unit 106, and the accuracy acquisition unit 107 is implemented by the control unit 11.

[0073] [6-1. Variation 1] For example, the profile of a contact person may change from time to time. If the contact person is another user, the tasks that other users are responsible for may change from time to time. The tasks that other users are knowledgeable about or skilled in may also change from time to time. The same applies if the contact person is something other than another user (for example, an organization, department, team, thread, bulletin board not categorized as a thread, or chat room). Therefore, in Variation 1, we will explain the case where profile information is prepared for each period.

[0074] Figure 9 shows an example of the profile database DB2 in Modification Example 1. As shown in Figure 9, the profile database DB2 in Modification Example 1 stores the query ID, period, and profile information in association. That is, profile information is associated with each query ID for each period. The period can be of any length. In Modification Example 1, the period is given as an example of one month, but the period can be shorter or longer than one month. For example, the period can be a day, a week, a quarter, or a year.

[0075] For example, server 10 generates profile information for a certain period from comments posted during that period based on comment database DB1 and stores it in profile database DB2. In variation 1, since the period is one month, server 10 generates profile information for other users for each month from comments posted by those users during that month and stores it in profile database DB2. Server 10 may store profile information for the entire past period in profile database DB2, or it may store profile information for the most recent period (for example, each month of the most recent year) in profile database DB2. For a single period, there is profile information for each of multiple query targets.

[0076] The profile information acquisition unit 102 in Modification 1 acquires profile information for each of the multiple query destinations for each of the multiple periods. The profile information acquisition unit 102 may acquire profile information for each of the multiple query destinations for all periods, or it may acquire profile information for each of the multiple query destinations for some periods. For example, the profile information acquisition unit 102 acquires profile information for each month of the most recent year from the profile database DB2. The profile information acquisition unit 102 may also acquire profile information for each month of the most recent year from other databases, other computers other than server 10, or information storage media.

[0077] Figure 10 shows an example of AI input and output in Modification Example 1. As shown in Figure 10, the default prompt in Modification Example 1 indicates that the AI ​​should estimate the contact by prioritizing recent profile information. For example, the default prompt might say, "You will be given inquiry information for each of several time periods. Please estimate the appropriate contact by prioritizing recent profile information." The default prompt may also say something else. For example, the default prompt might indicate that the AI ​​should estimate the contact by prioritizing older profile information, or by prioritizing profile information from a specific date and time or around that date and time. The specific date and time may be specified by the user.

[0078] In the modified example 1, the estimation unit 103 causes the AI ​​to estimate an appropriate query target based on profile information relating to each of the multiple query targets' profiles for each of the multiple query targets in each of the multiple time periods. For example, the estimation unit 103 inputs the default prompt shown in Figure 10, query information, and each of the multiple query targets' profile information for each of the multiple time periods to the AI. The AI ​​calculates these embedded representations based on pre-adjusted parameters. The AI ​​emphasizes the profile information for a particular time period based on the embedded representation of the default prompt and estimates an appropriate query target. The AI ​​outputs information indicating an appropriate query target corresponding to the calculated embedded representation. The estimation unit 103 obtains the information indicating an appropriate query target output from the AI.

[0079] For example, suppose another user used to be in the sales department and was knowledgeable about sales, but recently moved to the development department and became knowledgeable about websites. In this case, the other user's old profile information would indicate that they were knowledgeable about sales. The other user's recent profile information would indicate that they were knowledgeable about websites. If the inquiry information indicates an inquiry about HTML related to a website, the AI ​​estimates that the other user is the appropriate contact person from the embedded representation of the other user's recent profile information. The estimation unit 103 obtains information indicating the other user estimated by the AI. The processing after the appropriate contact person has been estimated may be the same as in the embodiment.

[0080] In the modified example 1, the estimation system 1 acquires profile information for each of the multiple query targets' profiles for each of several time periods. Based on the profile information for each of the multiple query targets' profiles for each of the multiple time periods, the estimation system 1 allows the AI ​​to estimate the appropriate query target. This allows the estimation system 1 to allow the AI ​​to estimate the appropriate query target based on the profile information for each period, thereby improving the accuracy of the AI's estimation. For example, if the default prompt indicates that recent profile information should be emphasized, the estimation system 1 can allow the AI ​​to estimate the appropriate query target according to recent trends.

[0081] [6-2. Variation 2] For example, in Modification Example 1, we used the case where profile information is prepared for each period, but conversely, the profile information acquisition unit 102 may acquire profile information that is not dependent on a period. Profile information that is not dependent on a period is profile information that is not prepared for each period. There is one set of profile information for each query destination.

[0082] The profile information acquisition unit 102 in the modified example 2 acquires a profile information set up for the entire past period, of which only one is available. The period-independent profile information represents a universal profile. For example, the period-independent profile information may represent hobbies or life's work. The profile information described in the embodiment is also an example of period-independent profile information, as it is not prepared for each period.

[0083] In Modification 2, the estimation unit 103 causes the AI ​​to estimate an appropriate query destination based on time-independent profile information. This differs from the Embodiment and Modification 1 in that the profile information is time-independent, but the processing after the time-independent profile information, etc., is input to the AI ​​may be the same as in the Embodiment and Modification 1. The AI ​​calculates an embedded representation of the time-independent profile information, etc., and outputs information indicating an appropriate query destination according to the embedded representation. The processing after an appropriate query destination is estimated may be the same as in the Embodiment.

[0084] In the modified example 2, Estimation System 1 acquires time-independent profile information. Based on this time-independent profile information, Estimation System 1 allows the AI ​​to estimate the appropriate contact point. This improves the accuracy of the AI's estimation because it allows Estimation System 1 to allow the AI ​​to estimate the appropriate contact point based on time-independent profile information. For example, Estimation System 1 can allow the AI ​​to estimate the appropriate contact point based on universal profile information such as the hobbies or lifestyle of other users.

[0085] [6-3. Modification 3] For example, there may be multiple types of contact points, such as other users, communication locations, organizations, departments, and teams. The type of contact point can also be described as a classification or attribute of the contact point. In variation 3, we take the example of a case where there are two types of contact points: other users and communication locations. There may be three or more types of contact points.

[0086] In the modified example 3, the profile information acquisition unit 102 acquires profile information for each of the multiple types of contact points. For each type, there are multiple contact points. For example, for the type of contact point "other users," there are multiple employees of a certain company. For the type of contact point "communication location," there are multiple threads used by a certain company. Similarly, for other types of contact points, there are multiple contact points for each other type.

[0087] Figure 11 shows an example of the profile database DB2 in Modification 3. As shown in Figure 11, the data storage unit 100 in Modification 3 stores a profile database DB2 for each type of query target. In Modification 3, there are two types of query targets: other users and communication locations. Therefore, the data storage unit 100 stores a profile database DB2 containing the profile information of other users and a profile database DB2 containing the profile information of communication locations. The method for generating the profile database DB2 containing the profile information of other users may be the same as in the embodiment.

[0088] For example, server 10 generates profile information for a communication location from comments posted to that location based on the comment database DB1 and stores it in the profile database DB2. If the communication location is a thread, server 10 generates profile information for each thread from comments posted to that thread and stores it in the profile database DB2. If the communication location is a space, server 10 generates profile information for each space from comments posted to that space and stores it in the profile database DB2.

[0089] For example, the profile information acquisition unit 102 acquires profile information for each of the multiple query destinations of a given type from each of the multiple types of profile databases DB2. The profile information acquisition unit 102 may also acquire profile information for each of the multiple types of query destinations from other databases, other computers other than the server 10, or information storage media. When the RAG mechanism is used, as in the embodiment, the profile information acquisition unit 102 searches for query information from each type of profile database DB2 based on the query information. The profile information acquisition unit 102 acquires the profile information of the given type that has been retrieved from each of the multiple types of profile databases DB2. The profile information may indicate which type of query destination the profile information belongs to.

[0090] Figure 12 shows an example of AI input and output in Modification 3. As shown in Figure 12, the default prompt in Modification 3 indicates that the AI ​​should estimate the appropriate type and the appropriate query destination for each of the multiple types of profile information. For example, the default prompt might say, "You will be given multiple types of profile information. Please estimate and output the appropriate type and the appropriate query destination for that type." The default prompt may also say something else. The default prompt may also indicate what type of profile the profile information is.

[0091] In Modification 3, the estimation unit 103 causes the AI ​​to estimate the appropriate query target from among multiple query targets of multiple types, based on profile information relating to the profiles of each of the multiple query targets of multiple types. For example, the estimation unit 103 inputs the default prompt shown in Figure 12, query information, and profile information for each of the multiple types to the AI. The AI ​​calculates these embedded representations based on pre-adjusted parameters. The AI ​​estimates that a certain type and a specific query target within that type are appropriate based on the embedded representation of the default prompt. The AI ​​outputs information indicating the appropriate query target of the appropriate type according to the calculated embedded representation. The estimation unit 103 obtains the information indicating the appropriate query target of the appropriate type output by the AI. In the example in Figure 12, the AI ​​estimates that a certain thread is appropriate, rather than a certain other user. The processing after the appropriate query target has been estimated may be the same as in the embodiment.

[0092] In the modified example 3, Estimation System 1 acquires profile information for each of the multiple types of multiple query destinations. Based on the profile information for each of the multiple types of multiple query destinations, Estimation System 1 has the AI ​​estimate the appropriate query destination from among the multiple types of multiple query destinations. This allows the user to know the appropriate query destination for the appropriate type, thus improving user convenience. For example, depending on the content of a query, it may be more appropriate for the user to query a specific thread rather than another user. In this case, by having the AI ​​estimate a specific thread rather than another user, the user can know that the specific thread is the appropriate query destination.

[0093] [6-4. Modification 4] For example, Embodiments and Modifications 1-3 describe a case where the profile information of the contact person is input to the AI. The information input to the AI ​​is not limited to the examples in Embodiments and Modifications 1-3. Modification 4 describes a case where the profile information of the user who is making the inquiry is input to the AI.

[0094] The estimation system 1 of Modification 4 includes a user profile information acquisition unit 104. The user profile information acquisition unit 104 acquires user profile information relating to the user's profile. The user profile information is the profile information of the user who entered the inquiry. The user profile information differs from the profile information of other users described in Embodiment and Modifications 1 to 3 in that it is the profile information of the user who entered the inquiry, but the content indicated by the profile information may be the same as that described in Embodiment and Modifications 1 to 3. Hereafter, where it is simply referred to as "profile information," it means the profile information of a user other than the user who entered the inquiry.

[0095] In Modification 4, since user profile information is stored in the profile database DB2, the user profile information acquisition unit 104 acquires user profile information from the profile database DB2. User profile information may also be stored in a database other than the profile database DB2, a computer other than the server 10, or an information storage medium. The user profile information acquisition unit 104 may acquire user profile information from the other database, the other computer, or the information storage medium.

[0096] Figure 13 shows an example of the AI's input and output in Modification 4. As shown in Figure 13, the default prompt in Modification 4 includes natural language text indicating that the AI ​​should estimate the appropriate contact based on the inquiry information, profile information, and user profile information. For example, the default prompt might be a sentence like, "You are an AI that estimates the appropriate contact. Please estimate the appropriate contact based on the inquiry information, profile information, and user profile information you have entered." The default prompt may also be a different sentence.

[0097] In Modification 4, the estimation unit 103 causes the AI ​​to estimate an appropriate query destination based on user profile information. For example, the estimation unit 103 inputs the default prompt, query information, profile information, and user profile information to the AI. The AI ​​calculates the embedded representations of these based on pre-adjusted parameters. The AI ​​may divide these into units called tokens and then calculate the embedded representation of each token. The AI ​​outputs information indicating an appropriate query destination corresponding to the embedded representation. The AI ​​can estimate an appropriate query destination not only based on the query information and profile information, but also based on the user prompt information, using the embedded representation of the default prompt. The estimation unit 103 obtains the information indicating an appropriate query destination output from the AI. The processing after an appropriate query destination has been estimated may be the same as in the embodiment.

[0098] In the modified example 4, Estimation System 1 acquires user profile information. Estimation System 1 then uses the user profile information to have the AI ​​estimate the appropriate contact point. This allows Estimation System 1 to have the AI ​​estimate the appropriate contact point according to the user who entered the inquiry, thereby improving the accuracy of the AI's estimation. For example, even for the same inquiry, the appropriate contact point may differ depending on the user. In the case of an HTML inquiry like the one in Figure 13, the user profile information indicates whether the user who made the inquiry has some knowledge of HTML or no knowledge of HTML at all. In such cases, the AI ​​can estimate the appropriate contact point according to the user who made the inquiry.

[0099] [6-5. Variation 5] For example, as briefly explained in the embodiment, the estimation unit 103 may have the AI ​​estimate an appropriate mention recipient as an appropriate query recipient. The mention recipient is not limited to other users as described in the embodiment. The mention recipient may be a group, not just an individual. For example, the mention recipient may be an organization, department, team, mailing list, chat room participant, or other group. In the modified example 5, it is assumed that the profile information of these candidate mention recipients is stored in the profile database DB2.

[0100] Figure 14 shows an example of the business support screen SC of Modification 5. As shown in the upper part of Figure 14, when a user enters an inquiry into the input form F, the estimation unit 103 causes the AI ​​to estimate the mention recipient. The default prompt of Modification 5 is shown to estimate an appropriate mention recipient based on the inquiry information and profile information. The AI ​​calculates the embedded representations of the default prompt, inquiry information, and profile information, and estimates the mention recipient according to these embedded representations.

[0101] For example, as shown in the lower part of Figure 14, the estimation unit 103 displays the mention recipient estimated by the AI ​​on the business support screen SC. In the example of Figure 14, as in the embodiment, user U5 is estimated as the appropriate contact, so the estimation unit 103 displays it on the input form F, with the mention symbol "@" followed by the string "User U5", indicating that it is the appropriate contact recipient. The user may complete the inquiry input as is, or they may change or delete the mention recipient if they think the mention recipient estimated by the AI ​​is not appropriate. In the example of Figure 14, the estimation unit 103 also displays a window W indicating that the AI ​​has estimated an appropriate mention recipient.

[0102] The estimation system 1 of Modification 5 includes a mention execution unit 105. The mention execution unit 105 executes a mention to the appropriate mention recipient when the user completes inputting an inquiry with the appropriate mention recipient specified. For example, the mention execution unit 105 executes a mention by sending a predetermined notification to the mention recipient estimated by the AI. The notification may be made by any notification means. For example, the mention execution unit 105 may execute a mention based on email, a notification function in a business support service, a notification function of the user terminal 20, or other notification means.

[0103] In the modified example 5, Estimation System 1 has the AI ​​estimate the appropriate mention recipient as the appropriate contact for the inquiry. When the user completes the inquiry input with the appropriate mention recipient specified, Estimation System 1 makes a mention to the appropriate mention recipient. This allows the user to know the appropriate mention recipient, thus improving user convenience.

[0104] [6-6. Variation 6] For example, in Modification 5, the case where the appropriate mention recipient estimated by the AI ​​is presented in a manner that the user can recognize is described. However, the appropriate mention recipient estimated by the AI ​​may also be presented in a manner that the user cannot recognize. In Modification 6, the mention recipient is estimated in the same manner as in Modification 5.

[0105] The estimation system 1 of the modified example 6 includes a mention target presentation unit 106. The mention target presentation unit 106 presents an appropriate mention target to the user in a manner in which the user does not recognize the appropriate mention target. A manner in which the user does not recognize the appropriate mention target means a manner in which the information indicating the appropriate mention target estimated by the AI ​​(e.g., letters, numbers, other symbols, images such as icons, or combinations thereof) is not included. In the lower example of Figure 14, a manner in which the information of the mention target is not included, such as "User U5", corresponds to a manner in which the user does not recognize the appropriate mention target.

[0106] Figure 15 shows an example of the business support screen SC of Modification 6. As shown in the upper part of Figure 15, the mention destination display unit 106 presents the appropriate mention destination to the user in a manner in which the user does not recognize the appropriate mention destination, by placing the string "HTTP expert" in response to the inquiry entered by the user after the mention symbol "@", rather than the name of another user estimated by the AI. Behind the scenes of the business support screen SC, it is assumed that user U5 is estimated as the appropriate inquiry destination, as in Modification 5. The user does not recognize a specific mention destination from a string such as "HTTP expert". This string may be estimated by the AI ​​or entered by the user.

[0107] As shown in the lower part of Figure 15, the mention destination display unit 106 may present appropriate mention destinations to the user in a manner that allows the user to recognize the appropriate mention destination when the user focuses the mouse cursor on the mention destination string "HTTP expert". In the example in the lower part of Figure 15, the mention destination display unit 106 displays a list L showing the mention destinations estimated by AI on the user terminal 20. For example, the user may delete or change the mention destinations displayed in list L. The mention destination display unit 106 may then present the mention destinations that the user has deleted or changed in list L again. The information of the mention destinations that the user has deleted or changed may be stored in the data storage unit 100.

[0108] In Modification 6, the mention execution unit 105 executes a mention to the appropriate mention target after the appropriate mention target has been presented to the user in a manner unrecognized by the user, and the user has completed inputting the inquiry with the appropriate mention target specified. This differs from Modification 5 in that the appropriate mention target is presented to the user in a manner unrecognized by the user, but the method of executing the mention after the user has completed inputting the inquiry is the same as in Modification 5.

[0109] In the modified example 6, estimation system 1 presents the appropriate mention to the user in a manner in which the user is unaware of the appropriate mention. After the appropriate mention is presented to the user in a manner in which the user is unaware, estimation system 1 executes a mention to the appropriate mention when the user completes the input of the inquiry with the appropriate mention specified. As a result, the user can complete the input of the inquiry without being aware of the details of the appropriate mention, thus improving user convenience.

[0110] [6-7. Variation 7] For example, as explained to some extent in Modification Example 5, the appropriate mention recipient estimated by the AI ​​may be presented in a way that is recognizable to the user. The estimation system 1 in Modification Example 7 includes a mention recipient presentation unit 106. The mention recipient presentation unit 106 presents the appropriate mention recipient to the user before the user completes the input of the inquiry. Completing the input of the inquiry means that the operation of posting the inquiry is performed.

[0111] Figure 16 shows an example of the business support screen SC of Modification 7. In the upper example of Figure 16, the mention target display unit 106 displays the appropriate contact target estimated by the AI ​​in window W before the user enters and completes the input of the inquiry, thereby presenting the appropriate contact target. When the user performs an operation to set the appropriate contact target displayed in window W as the mention target (for example, by selecting "Yes" in window W in Figure 16), the user terminal 20 sets the appropriate contact target estimated by the AI ​​as the mention target, as shown in the lower part of Figure 16. The mention target display unit 106 may also present the appropriate contact target by other methods other than those shown in Figure 16. For example, the mention target display unit 106 may display the appropriate contact target in a list L as shown in Figure 17.

[0112] In Modification 7, the mention execution unit 105 executes a mention to the appropriate mention recipient after the appropriate mention recipient has been presented to the user and the user has completed inputting the query with the appropriate mention recipient specified. This differs from Modification 5 in that the appropriate mention recipient is specified after the appropriate mention recipient has been presented to the user, but the method of executing the mention after the user has completed inputting the query is the same as in Modification 5.

[0113] In the modified example 7, estimation system 1 presents the user with appropriate mentions before the user completes the input of the inquiry. After the user is presented with appropriate mentions and the user has completed the input of the inquiry with the appropriate mentions specified, estimation system 1 executes a mention on the appropriate mentions. This allows the user to complete the input of the inquiry after knowing and specifying the appropriate mentions, thus improving user convenience.

[0114] [6-8. Variation 8] For example, in Modification 7, the estimation unit 103 may cause the AI ​​to estimate multiple appropriate mentions. The default prompt in Modification 8 indicates that multiple appropriate mentions should be estimated. The AI ​​estimates multiple appropriate mentions based on the embedded representation of the default prompt. In Modification 8, the accuracy of the estimation result for each mention is calculated. Accuracy is an indicator of the accuracy of the AI's estimation of appropriate query targets. Accuracy is sometimes called probability or score. Accuracy can also be described as the reliability of the AI's estimation result of appropriate query targets.

[0115] The estimation system 1 of Modification 8 includes a confidence acquisition unit 107. The confidence acquisition unit 107 acquires the confidence level of each of several appropriate mentioned targets from the AI. The method for calculating confidence level may be a known method. For example, the confidence acquisition unit 107 may acquire the confidence level of the AI ​​estimation result based on a softmax function, a sigmoid function, temperature scaling, a Bayesian approach, distance in the feature space, or other calculation methods. Modification 8 gives an example where the confidence level is expressed numerically. For example, the higher the confidence level, the more reliable the AI ​​estimation result is. The lower the confidence level, the less reliable the AI ​​estimation result is. The confidence level may be expressed in letters or other symbols.

[0116] Figure 17 shows an example of the business support screen SC of Modification 8. In Modification 8, the mention destination display unit 106 presents the user with multiple appropriate mention destinations based on the probability of each of the multiple appropriate mention destinations when the user enters a symbol related to a mention. In the example in Figure 17, the mention destination display unit 106 displays a list L on the user terminal 20, which is sorted in descending order of the probability of each of the multiple appropriate mention destinations.

[0117] The mention destination display unit 106 may display a list L on the user terminal 20, sorted so that each of the multiple appropriate mention destinations is ordered in ascending order of their respective likelihood levels. The mention destination display unit 106 may display in list L only those mention destinations whose likelihood level is above a threshold. The mention destination display unit 106 may display in list L only those mention destinations whose likelihood level is above a predetermined rank. The mention destination display unit 106 may display in list L only those mention destinations whose likelihood level is above a threshold and whose likelihood level is above a predetermined rank (for example, mention destinations with a likelihood level of 80% or higher and whose likelihood level is among the top 10).

[0118] In the modified example 8, Estimation System 1 has the AI ​​estimate multiple appropriate mention recipients. Estimation System 1 obtains the probability of each of the multiple appropriate mention recipients from the AI. When the user enters a symbol related to a mention, Estimation System 1 presents each of the multiple appropriate mention recipients to the user based on the probability of each of the multiple appropriate mention recipients. This allows the user to understand the appropriate mention recipient according to the probability, thus improving user convenience.

[0119] [6-9. Modification 9] For example, if the person making the inquiry is another user, the user may be able to make new discoveries or obtain new information if the AI ​​estimates other users with whom the user does not usually communicate. Therefore, the estimation unit 103 may, based on a prompt indicating that it wants to estimate other users with whom the user does not usually communicate as appropriate users, have the AI ​​estimate other users with whom the user does not usually communicate as appropriate users.

[0120] Figure 18 shows an example of the AI's input and output in Modification 9. As shown in Figure 18, the default prompt in Modification 9 indicates that the AI ​​should estimate other users with whom the user does not communicate often. For example, the default prompt might read, "You are an AI that estimates appropriate other users. Please estimate the appropriate contact person among other users with whom the user who entered the inquiry does not usually communicate." The default prompt may also read other sentences.

[0121] For example, when a default prompt, inquiry information, and profile information are input to the AI, the AI ​​calculates their embedded representations based on pre-adjusted parameters. The AI ​​may divide these into units called tokens and then calculate the embedded representation for each token. The AI ​​outputs information indicating the appropriate contact person according to the embedded representation. Based on the embedded representation of the default prompt, the AI ​​prioritizes estimating other users with whom the user does not communicate often. The estimation unit 103 acquires the information indicating the appropriate contact person output by the AI. The processing after the appropriate contact person has been estimated may be the same as in the embodiment.

[0122] In the modified version 9, estimation system 1 causes the AI ​​to estimate appropriate other users based on a prompt indicating that the user should estimate other users with whom the user does not communicate as appropriate other users. This allows the user to estimate other users with whom they do not communicate much as appropriate other users, thus improving user convenience.

[0123] [6-10. Other variations] For example, you may combine two or more of the above variations.

[0124] For example, while we have described an application of Estimation System 1 to business support services, Estimation System 1 may also be applied to other services besides business support services. Estimation System 1 may also be used to have the AI ​​estimate the appropriate contact point when a user enters an inquiry that is not particularly related to business (for example, an inquiry about a service that the user uses privately). The inquiry information may indicate an inquiry that is not related to business. The profile information may indicate a profile that is not related to business.

[0125] For example, a function described as being implemented on server 10 may be implemented on user terminal 20. In this case, the function can be implemented by a browser script or an application installed on user terminal 20. For example, each function may be shared among multiple computers, or it may be implemented by a single computer. [Explanation of Symbols]

[0126] 1 Estimation system, 10 Server, 11,21 Control unit, 12,22 Storage unit, 13,23 Communication unit, 20 User terminal, 24 Operation unit, 25 Display unit, F Input form, L List, N Network, W Window, SC Business support screen, 100 Data storage unit, 101 Inquiry information acquisition unit, 102 Profile information acquisition unit, 103 Estimation unit, 104 User profile information acquisition unit, 105 Mention execution unit, 106 Mention target presentation unit, 107 Accuracy acquisition unit, DB1 Comment database, DB2 Profile database.

Claims

1. An inquiry information acquisition unit that acquires inquiry information related to the inquiry entered by the user, A profile information acquisition unit acquires profile information for each of the profiles of multiple contacts based on a vector representing the characteristics of the comment in the comment database where the comment posted in the business support service is stored, and other information other than the comment registered in the business support service. An estimation unit causes an AI (Artificial Intelligence), which is a trained large-scale language model, to estimate the appropriate query destination from among the plurality of query destinations, based on the query information and the profile information. An estimation system that includes this.

2. The inquiry information acquisition unit acquires the inquiry information when the user using the business support service enters the inquiry related to the business. The profile information acquisition unit acquires the profile information based on the business support service, The estimation unit causes the AI ​​to estimate the appropriate contact point in the business support service. The estimation system according to claim 1.

3. The aforementioned multiple contacts are multiple other users different from the aforementioned user, The profile information acquisition unit acquires the profile information relating to each of the profiles of the plurality of other users, The estimation unit causes the AI ​​to estimate, from among the plurality of other users, the appropriate other user relating to the inquiry, as the appropriate inquiry recipient. The estimation system according to claim 1 or 2.

4. The profile information acquisition unit acquires the profile information relating to each of the profiles of the plurality of query destinations for each of the plurality of periods, The estimation unit causes the AI ​​to estimate the appropriate contact based on the profile information relating to each of the profiles of the plurality of contacts in each of the plurality of periods. The estimation system according to claim 1 or 2.

5. The profile information acquisition unit acquires the profile information which is independent of the period, The estimation unit causes the AI ​​to estimate the appropriate contact based on the profile information which is independent of the period. The estimation system according to claim 1 or 2.

6. The profile information acquisition unit acquires the profile information relating to each of the multiple types of the multiple query destinations, The estimation unit causes the AI ​​to estimate the appropriate query destination from among the multiple types of query destinations, based on the profile information relating to the profile of each of the multiple types of query destinations. The estimation system according to claim 1 or 2.

7. The estimation system further includes a user profile information acquisition unit that acquires user profile information relating to the user's profile, The estimation unit further causes the AI ​​to estimate the appropriate contact based on the user profile information. The estimation system according to claim 1 or 2.

8. An inquiry information acquisition unit that acquires inquiry information relating to an inquiry entered by a user, A profile information acquisition unit that acquires profile information for each of the profiles of multiple contacts, An estimation unit causes an AI (Artificial Intelligence), which is a trained large-scale language model, to estimate an appropriate mention target from among the plurality of query targets as the appropriate query target for the query, based on the query information and the profile information. When the user completes inputting the inquiry with the appropriate mention recipient specified, a mention execution unit executes a mention to the appropriate mention recipient. An estimation system that includes this.

9. The estimation system further includes a mention-to-user unit that presents the appropriate mention to the user in a manner in which the user does not recognize the appropriate mention-to-user, The mention execution unit executes a mention to the appropriate mention recipient after the appropriate mention recipient has been presented to the user in a manner not recognized by the user, and the user has completed inputting the query with the appropriate mention recipient specified. The estimation system according to claim 8.

10. The estimation system further includes a mention suggestion unit that presents the appropriate mention to the user before the user completes inputting the query, The mention execution unit executes a mention to the appropriate mention recipient after the appropriate mention recipient has been presented to the user, and the user has completed inputting the query with the appropriate mention recipient specified. The estimation system according to claim 8 or 9.

11. The estimation unit causes the AI ​​to estimate a plurality of appropriate mention recipients. The estimation system further includes a confidence acquisition unit that acquires the confidence level of each of the multiple appropriate mention targets from the AI, The mention recipient presentation unit, when the user inputs a symbol related to a mention, presents each of the multiple appropriate mention recipients to the user based on the probability of each of the multiple appropriate mention recipients. The estimation system according to claim 10.

12. Based on a prompt indicating that the estimation unit should estimate the appropriate other user among the plurality of other users with whom the user does not communicate, the estimation unit causes the AI ​​to estimate the appropriate other user. The estimation system according to claim 3.

13. We retrieve inquiry information related to the inquiry entered by the user. Based on a vector representing the characteristics of a comment in a comment database where comments posted in the business support service are stored, and other information other than the comment registered in the business support service, profile information for each of the profiles of multiple contacts is obtained. Based on the aforementioned query information and the aforementioned profile information, the AI ​​(Artificial Intelligence), which is a trained large-scale language model, is made to estimate the appropriate query destination from among the multiple query destinations for the query. Estimation method.

14. Inquiry information acquisition unit that acquires inquiry information related to inquiries entered by the user. A profile information acquisition unit that acquires profile information for each of the profiles of multiple contacts, based on a vector representing the characteristics of the comment in a comment database where comments posted in the business support service are stored, and other information other than the comment registered in the business support service, An estimation unit causes an AI (Artificial Intelligence), which is a trained large-scale language model, to estimate the appropriate query destination from among the plurality of query destinations, based on the query information and the profile information. A program that makes a computer function.

Images