Method, system, and program for supporting matching

Abstract

This method for supporting matching between users includes: reading a first personalized AI of a first user; reading a plurality of second personalized AIs of a plurality of second users; recording a conversation between the first personalized AI and each of the plurality of second personalized AIs; calculating a score for each of the plurality of second personalized AIs on the basis of the conversation; and outputting the score or information that is based on the score.

Description

Method, system, and program for assisting matching

[0001] The present invention relates to a method, a system, and a program for supporting matching, and more particularly to a method for supporting matching between a user and at least one user among a plurality of other users.

[0002] In the fields of recruitment and job hunting, romance and marriage, users are matched with each other. For example, a system for supporting matching is known, as disclosed in Patent Document 1.

[0003] JP 2023-86224 A

[0004] An object of the present invention is to make it possible to easily obtain highly accurate matching results.

[0005] The present invention provides a method for assisting matching using personalized AI.

[0006] The present invention provides, for example, the following items: (Item 1) A method for supporting matching between users, comprising: retrieving a first personalized AI of a first user; retrieving multiple second personalized AIs of multiple second users; recording a conversation between the first personalized AI and each of the multiple second personalized AIs; calculating a score for each of the multiple second personalized AIs based on the conversation; and outputting the score or information based on the score. (Item 2) The method according to any one of the above items, wherein the score includes scores from a predetermined number of perspectives, and calculating the score includes calculating the scores from the predetermined number of perspectives according to weighting. (Item 3) The method according to any one of the above items, further comprising updating the weighting. (Item 4) The method of any one of the above items, further comprising: after outputting the score or the information, receiving input from the first user selecting one of the plurality of second users, and updating the weightings includes updating the weightings so that the score with the second personalized AI of the selected second user becomes higher. (Item 5) The method of any one of the above items, further comprising: after outputting the score or the information, receiving input from the first user selecting one of the plurality of second users, and performing a re-training process on the first personalized AI so that the score with the second personalized AI of the selected second user becomes higher. (Item 6) The method of any one of the above items, further comprising: generating the first personalized AI, reading the first personalized AI includes reading the generated first personalized AI, and generating the first personalized AI includes performing a training process by fine-tuning a language model using sentences expressed by the first user.(Item 7) The method of any one of the above items, wherein generating the first personalized AI further includes: presenting a question to the first user; and receiving an answer to the question from the first user, wherein the sentence expressed by the first user includes the answer. (Item 8) The method of any one of the above items, further including providing an AI agent that supports the first user, wherein the question is provided by the AI ​​agent. (Item 9) The method of any one of the above items, further including calculating a hallucination score for the answer; and determining whether the hallucination score satisfies a threshold, wherein if the hallucination score satisfies the threshold, the answer is used to generate the first personalized AI, and if the hallucination score does not satisfy the threshold, the answer is not used to generate the first personalized AI. (Item 10) A system for assisting in matching users, comprising: a first reading means for reading a first personalized AI of a first user; a second reading means for reading a plurality of second personalized AIs of a plurality of second users; a recording means for recording a conversation between the first personalized AI and each of the plurality of second personalized AIs; a calculation means for calculating a score for each of the plurality of second personalized AIs based on the conversation; and an output means for outputting the score or information based on the score. (Item 10A) The system according to item 10, including the features according to any one of the above items.(Item 11) A program for supporting matching between users, the program being executed in a computer system having a processor unit, the program causing the processor unit to execute processes including: reading out a first personalized AI of a first user; reading out multiple second personalized AIs of multiple second users; recording conversations between the first personalized AI and each of the multiple second personalized AIs; calculating a score for each of the multiple second personalized AIs based on the conversations; and outputting the score or information based on the score. (Item 11A) The program according to item 11, including the features described in any one of the above items. (Item 11B) A non-transitory computer-readable storage medium storing the program according to item 11 or item 11A.

[0007] The present invention provides a method for assisting matching using personalized AI, which makes it possible to easily obtain highly accurate matching results. The present invention makes it possible to obtain a huge amount of interview results in a short period of time, which can bring about improvements in fields where matching is applicable.

[0008] FIG. 1 shows an example of a flow in a service using the system 100 of the present invention. FIG. 2 shows an example of a flow in which an AI agent takes the lead in generating personalized AI in the system 100. FIG. 3 shows an example of the configuration of the system 100 that supports matching between users. FIG. 4 shows an example of a specific configuration of the system 100. FIG. 5 shows an example of the configuration of the processor unit 120. FIG. 6 shows an example of the configuration of the processor unit 120', which is an alternative embodiment of the processor unit 120.

[0009] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0010] 1A. Service Supporting Matching Between Job Seekers and Employers The inventors of the present invention have developed a new service that supports matching between job seekers and employers. This service uses a "digital clone" that converses on behalf of humans. A "digital clone" can be realized by a "personalized AI" that can learn personal information, such as an individual's behavior and thoughts, and act in a way that is appropriate for that individual. For example, a "job seeker personalized AI" that learns information obtained from a job seeker can act in a way that is appropriate for that job seeker. For example, a "recruiter personalized AI" that learns information obtained from an employer can act in a way that is appropriate for that employer. This service involves a "job seeker personalized AI" and a "recruiter personalized AI" conducting interviews on behalf of the job seeker and employer, respectively, and supporting matching based on the results of those interviews.

[0011] For example, a "job seeker personalized AI" acting on behalf of the job seeker can conduct interviews with multiple "recruiter personalized AIs" for each of the multiple recruiters, allowing the job seeker to virtually conduct interviews with multiple recruiters. In the real world, it takes time and effort for a job seeker to conduct interviews with each of the multiple recruiters, but with this service, personalized AIs conduct interviews virtually with each other, reducing both time and effort. In fact, a huge amount of interview results can be obtained in a short amount of time.

[0012] This service can output a score representing the compatibility between a job seeker and each of multiple employers based on the results of a virtual interview, output information on employers who are a good match for the job seeker, and output information on job seekers who are a good match for the employer. This allows job seekers to find employers who are a good match without actually interviewing multiple employers. Employers can similarly find job seekers who are a good match for multiple job seekers without actually interviewing multiple job seekers. The "job seeker personalized AI" and the "employer personalized AI" are capable of acting like a job seeker and an employer, respectively, and the compatibility determined by this service can show trends similar to those observed when the job seeker and employer actually interview each other. If the job seeker or employer feels that the person identified as a good match by this service is not actually a good match, the "job seeker personalized AI" or the "employer personalized AI" can be retrained to adjust the "job seeker personalized AI" and the "employer personalized AI" to better match the job seeker and employer, respectively.

[0013] Furthermore, by having personalized AIs virtually interview each other, the anonymity of job seekers can be ensured. For example, if an executive of a listed company (in this invention, "executive" refers to executives, those receiving executive treatment, and department heads such as those in research and development or technological development) were to change jobs, even a leak of that information could affect the stock price. Therefore, executives of listed companies could not easily search for jobs. By using this service, an anonymously generated "job seeker personalized AI" virtually interviews with a "recruiter personalized AI," so there is no need to worry about information leaking that an executive of a listed company is looking to change jobs.

[0014] FIG. 1A shows an example of a flow of this service. In the example shown in FIG. 1A, user U is a job seeker who uses system 100 to search for a job, user C1 is a person in charge of a first company (recruiter) that uses system 100 to recruit, and user C2 is a person in charge of a second company that uses system 100 to recruit. Although not shown in FIG. 1A, people in charge of multiple companies may use system 100. Each user (user U, user C1, user C2) may communicate with system 100 via a terminal device (e.g., a smartphone, a tablet, or a personal computer).

[0015] In step S1, the user U inputs his / her own information into the system 100. The user U inputs information such as his / her name, address, desired annual salary, skills, career history, and work history. For example, this information can be input by uploading a resume and / or employment history. Alternatively, this information can be input according to an input form displayed on the screen of a terminal device.

[0016] In addition to this information, as will be described later with reference to Figure 1B, information can also be input by answering questions posed by the AI ​​agent, allowing for more detailed information to be input.

[0017] In parallel with step S1, in step S2, user C1 inputs information about the first company into system 100. User C1 inputs information such as the company name, location, job information, annual salary, job title, work location, and desired skills. For example, this information can be input by uploading a company brochure and / or recruitment requirements. Alternatively, this information can be input according to an input form displayed on the screen of a terminal device.

[0018] In addition to this information, as will be described later with reference to Figure 1B, information can also be input by answering questions posed by the AI ​​agent, allowing for more detailed information to be input.

[0019] In parallel with step S2, in step S3, user C2 inputs information about a second company into system 100. User C2 can input information in the same manner as user C1. If there are users from additional companies, each user can input information in the same manner.

[0020] When the user's information is input, the system 100 generates a personalized AI for the user U. Based on the information of the user U, the personalized AI PAI for the user U is generated. Ｕ is generated, and a personalized AI PAI for user C1 is generated based on the information of user C1. Ｃ１ is generated, and a personalized AI PAI for user C2 is generated based on the information of user C2. Ｃ２ If there are additional corporate users, a personalized AI for each user is generated in the same manner.

[0021] The PAI can be generated based on, for example, a Large Language Model (LLM). In one example, the PAI can be generated by fine-tuning the LLM using user information. When a prompt is input to the PAI generated in this way, it outputs a sentence that appears to be written by the user.

[0022] In the system 100, the user U's personalized AI PAI Ｕ and user C1's personalized AI PAI Ｃ１ In parallel with this, a virtual interview with the user U's personalized AI PAI Ｕ and user C2's personalized AI PAI Ｃ２ If there are additional enterprise users, the user U's personalized AI PAI Ｕ There will also be virtual interviews with personalized AI for additional corporate users.

[0023] For example, PAI Ｕ And P.A.I. Ｃ１ During a virtual interview with PAI, appropriate promptsＣ１ When entered into PAI Ｃ１ is PAI Ｕ The question is output as a prompt to the PAI. Ｕ When entered into PAI Ｕ The answer sentence is then sent to the PAI as a prompt. Ｃ１ When entered into PAI Ｃ１ will output further sentences. By repeating this process, the virtual interview will progress. Ｕ And P.A.I. Ｃ２ Virtual interviews with the company will proceed in the same way.

[0024] For example, a PAI-to-PAI conversation can be initiated by an AI agent initiating the conversation, e.g., in response to a user accessing system 100 or in response to a user's personalized AI being created, the AI ​​agent can speak to the PAI or input a prompt.

[0025] For example, a PAI-to-PAI interview can be terminated when a predetermined amount of information has been gathered. For example, a prompt to the PAI can be included to end the interview when a certain amount or a predetermined amount of information has been gathered, allowing the PAI to autonomously make the decision to end the interview automatically.

[0026] The sentences output from each PAI during a virtual interview are recorded for each interview, and a score for each interview is generated based on the recorded sentences. The score may represent the compatibility between the PAIs during that interview. For example, if the mutual outputs are conversationally consistent, it is assumed that the PAIs and, in turn, the users are compatible, and a high score may be calculated. Conversely, if the mutual outputs are conversationally inconsistent, it is assumed that the PAIs and, in turn, the users are not compatible, and a low score may be calculated. The score may be calculated, for example, by vectorizing the output from one PAI and vectorizing the output from the other PAI, and then calculating the distance or similarity between the vectors. Alternatively, the score may be calculated, for example, by embedding the mutual outputs. Alternatively, the score may be calculated, for example, by an LLM.

[0027] For example, the scores may be generated for each of a plurality of predetermined perspectives. For example, the predetermined plurality of perspectives may be a plurality of perspectives according to the conditions that a user U (job seeker) requires of a company (employer). In this case, the scores for each of the plurality of perspectives may represent the degree to which the company (employer) meets the conditions that the user U requires. For example, the plurality of perspectives may be perspectives such as "salary," "benefits," "ease of work," and "future prospects." The score for the "salary" perspective may represent the degree to which the company meets the "salary" conditions that the user U requires of the company. The score for the "benefits" perspective may represent the degree to which the company meets the "benefits" conditions that the user U requires of the company. The score for the "ease of work" perspective may represent the degree to which the company meets the "ease of work" conditions that the user U requires of the company. The score for the "future prospects" perspective may represent the degree to which the company meets the "future prospects" conditions that the user U requires of the company.

[0028] For example, the scores of a predetermined number of perspectives may be weighted according to the characteristics of the user U or as specified by the user U. For example, a perspective that the user U considers important may be weighted more heavily than a perspective that the user U does not consider important, thereby emphasizing the relative merits of the scores of the perspectives that the user U considers important. The characteristics of the user U are determined, for example, based on the user information input in step S1.

[0029] For example, the scores of the predetermined multiple perspectives may be weighted according to industry know-how in addition to or instead of the characteristics of the user U. The industry know-how may be accumulated as the system 100 is used, or may be predetermined.

[0030] The scores of a given number of aspects may be combined and expressed as a total score.

[0031] The calculated score may be presented to each user, or information based on the calculated score may be presented to each user. The information based on the calculated score may be, for example, information indicating which company has the highest score or a score higher than a threshold, or a recommendation to the recruiter. The information based on the calculated score may be, for example, the reason for the score. The reason for the score may be described, for example, by the LLM.

[0032] For example, in step S4, the system 100 presents the calculated score to the user U. The score is presented for each interview with each recruiter. For example, the PAI of the user C1 of the first company is Ｃ１ Score of the interview with the second company's user C2, Ｃ２ The scores of the interviews with the company, etc. are presented to the user U. The user U uses the scores as a clue to consider which company he or she would like to work for.

[0033] In addition to the calculated score or information based on the score, the result of the virtual interview itself may be presented to the user U. This allows the user U to check what kind of conversation took place between the personalized AIs in the virtual interview. For example, if the user U feels that the content of the conversation with his / her personalized AI is not like him / her or is inappropriate, he / she can modify the content of the conversation with his / her personalized AI.

[0034] In addition to the calculated score or information based on the score, the user U may be presented with reasons for matching with each recruiter. The reasons for matching may be described, for example, by the LLM.

[0035] In step S5, when user U inputs an application to a company where he or she hopes to work into system 100, user C1 or user C2 is notified in step S6 or step S7 that user U has applied to the company. As a result, user C1 or user C2 contacts user U, and the actual hiring process begins. Alternatively, when user U inputs an application to a company where he or she hopes to work into system 100 in step S5, an interview appointment may be automatically scheduled in step S6 or step S7.

[0036] In this way, compatibility with a company can be estimated through virtual interviews before the actual recruitment process, and the actual recruitment process will proceed with companies that are assumed to be a good fit, thereby reducing the possibility that the actual recruitment process will end in failure.

[0037] Furthermore, by generating personalized AI using detailed information and qualitative information of the employer or job seeker, a virtual interview will be conducted based on the detailed information and qualitative information of the employer or job seeker. As a result, the score calculated from the results of the virtual interview will also reflect the detailed information and qualitative information of the employer or job seeker, which may lead to matching based on the detailed information and qualitative information of the employer or job seeker.

[0038] For example, when generating a personalized AI, sensitive or confidential information that is difficult to talk about with others may be input. This allows for the exchange of sensitive or confidential information in interviews between personalized AIs. For example, information that would not be shared in interviews between humans can be exchanged in interviews between personalized AIs, which has the advantage of allowing the extraction of information that humans cannot extract.

[0039] The generated personalized AI may be subjected to a re-learning process to more closely resemble the user based on actions taken by the user in consideration of the results of the virtual interview. The action taken by the user in consideration of the results of the virtual interview may be, for example, applying to a company where the user wishes to work. In this case, at least one of the personalized AI of user U and the personalized AI of user C1 at the company may be adjusted so that the score of an interview between the personalized AI of the person in charge at the company to which the user applied (e.g., user C1) and the personalized AI of the job seeker (e.g., user U) is higher. For example, the weighting of the scores of a predetermined number of perspectives may be adjusted so that the score of an interview between the personalized AI of the person in charge at the company to which the user applied (e.g., user C1) and the personalized AI of the job seeker (e.g., user U) is higher.

[0040] The action taken by the user in consideration of the results of the virtual interview may be, for example, the user correcting the content of the conversation that the user's personalized AI had in the virtual interview. Ｕ And P.A.I. Ｃ１ PAI in a virtual interview with Ｕ The user U can modify the content of the conversation by, in which case the user U's personalized AI can be adjusted based on the modified content of the conversation (e.g., so that it speaks the modified content of the conversation).

[0041] The action taken by the user in consideration of the results of the virtual interview may be, for example, a conversation between the user and the personalized AI. For example, the user U may directly interact with the personalized AI of the user C1 of the company. For example, the user U may interact with the personalized AI of the user C1 of the company. Ｃ１ In this case, the personalized AI for the user U can be adjusted by learning the content of the interactions that the user U has had (for example, the type of question the user wants answered, etc.).

[0042] 1B. Service to support matching between investors and investment targets The above-mentioned service can also be applied to support matching between investors and investment targets. For example, an "investor personalized AI" that has learned information obtained from an investor can act in a way that is appropriate for that investor. For example, an "investment target personalized AI" that has learned information obtained from an investment target can act in a way that is appropriate for that investment target. In this service, the "investor personalized AI" and the "investment target personalized AI" conduct interviews on behalf of the investor or investment target, respectively, and support matching based on the results of those interviews.

[0043] For example, an "Investor Personalized AI" acting on behalf of an investor can virtually conduct interviews with multiple investment targets by conducting interviews with each of the "Investee Personalized AIs" for multiple investment targets. In the real world, it takes time and effort for an investor to conduct interviews with each of multiple investment targets, but with this service, personalized AIs conduct interviews virtually with each other, thereby reducing both time and effort. In fact, a huge amount of interview results can be obtained in a short period of time. Furthermore, while in the real world, it takes time and effort for an investor to search for multiple investment targets, this service allows investors to easily find multiple investment targets that are using this service.

[0044] This service can output a score representing the compatibility between an investor and each of multiple investment targets based on the results of a virtual interview, output information on investment targets that are a good fit for the investor, and output information on investors that are a good fit for the investment targets. This allows investors (or business companies) to find investment targets that are a good fit without actually interviewing multiple investment targets. Similarly, investment targets (e.g., startup companies) can find investors that are a good fit without actually interviewing multiple investors. The "investor personalized AI" and the "investment target personalized AI" can act like investors and investment targets, respectively, and the compatibility identified by this service can show trends similar to those observed when the investor and investment target actually meet. If an investor or investment target feels that a partner identified as a good fit by this service is not actually a good fit, the "investor personalized AI" or the "investment target personalized AI" can be retrained to adjust the "investor personalized AI" and the "investment target personalized AI" so that they are closer to the investor and investment target, respectively.

[0045] Furthermore, by having personalized AIs virtually interview each other, the confidentiality of investments can be ensured. Investments are generally conducted under confidentiality, but by using this service, an anonymously generated "investor personalized AI" virtually interviews with an "investment target personalized AI," eliminating the possibility of information leaks.

[0046] Referring again to FIG. 1A , an example of a flow of this service is shown. In this example, user U is an investor (or a person in charge of an investment company or a business company) who uses system 100 to look for investment targets, user C1 is a person in charge of a first company (an investment target) that uses system 100 to solicit investors, and user C2 is a person in charge of a second company that uses system 100 to solicit investors. Although not shown in FIG. 1A , people in charge of multiple companies may use system 100. Each user (user U, user C1, user C2) may communicate with system 100 via a terminal device (e.g., a smartphone, a tablet, or a personal computer).

[0047] In step S1, the user U inputs information about himself or her company into the system 100. The user U inputs information such as the name, investment objectives, investment field, and available investment amount. This information can be input, for example, according to an input form displayed on the screen of the terminal device.

[0048] In addition to this information, as will be described later with reference to Figure 1B, information can also be input by answering questions posed by the AI ​​agent, allowing for more detailed information to be input.

[0049] In parallel with step S1, in step S2, user C1 inputs information about the first company into the system 100. User C1 inputs information such as the company name, location, manager, technology field, business region, desired funding amount, and existing VC. This information can be input, for example, according to an input form displayed on the screen of a terminal device. In addition to this information, as will be described later with reference to FIG. 1B, information can also be input by answering questions posed by an AI agent. This allows more detailed information to be input.

[0050] In parallel with step S2, in step S3, user C2 inputs information about a second company into system 100. User C2 can input information in the same manner as user C1. If there are users from additional companies, each user can input information in the same manner.

[0051] When the user's information is input, the system 100 generates a personalized AI for the user U. Based on the information of the user U, the personalized AI PAI for the user U is generated. Ｕ is generated, and a personalized AI PAI for user C1 is generated based on the information of user C1. Ｃ１ is generated, and a personalized AI PAI for user C2 is generated based on the information of user C2. Ｃ２ If there are additional corporate users, personalized AI for each user is generated in the same manner. At this time, in addition to the user information, know-how of the industry in which the investment is made may be used.

[0052] The PAI can be generated based on, for example, a Large Language Model (LLM). In one example, the PAI can be generated by fine-tuning the LLM using user information. When a prompt is input to the PAI generated in this way, it outputs a sentence that appears to be written by the user.

[0053] In the system 100, the user U's personalized AI PAI Ｕ and user C1's personalized AI PAI Ｃ１ In parallel with this, a virtual interview with the user U's personalized AI PAI Ｕ and user C2's personalized AI PAI Ｃ２ If there are additional enterprise users, the user U's personalized AI PAI Ｕ There will also be virtual interviews with personalized AI for additional corporate users.

[0054] For example, PAI Ｕ And P.A.I. Ｃ１ During a virtual interview with PAI, appropriate prompts Ｃ１ When entered into PAI Ｃ１ is PAI Ｕ The question is output as a prompt to the PAI. ＵWhen entered into PAI Ｕ The answer sentence is then sent to the PAI as a prompt. Ｃ１ When entered into PAI Ｃ１ will output further sentences. By repeating this process, the virtual interview will progress. Ｕ And P.A.I. Ｃ２ Virtual interviews with the company will proceed in the same way.

[0055] For example, a PAI-to-PAI conversation can be initiated by an AI agent initiating the conversation, e.g., in response to a user accessing system 100 or in response to a user's personalized AI being created, the AI ​​agent can speak to the PAI or input a prompt.

[0056] For example, a PAI-to-PAI interview can be terminated when a predetermined amount of information has been gathered. For example, a prompt to the PAI can be included to end the interview when a certain amount or a predetermined amount of information has been gathered, allowing the PAI to autonomously make the decision to end the interview automatically.

[0057] The sentences output from each PAI during a virtual interview are recorded for each interview, and a score for each interview is generated based on the recorded sentences. The score may represent the compatibility between the PAIs during that interview. For example, if the mutual outputs are conversationally consistent, it is assumed that the PAIs and, in turn, the users are compatible, and a high score may be calculated. Conversely, if the mutual outputs are conversationally inconsistent, it is assumed that the PAIs and, in turn, the users are not compatible, and a low score may be calculated. The score may be calculated, for example, by vectorizing the output from one PAI and vectorizing the output from the other PAI, and then calculating the distance or similarity between the vectors. Alternatively, the score may be calculated, for example, by embedding the mutual outputs. Alternatively, the score may be calculated, for example, by an LLM.

[0058] For example, a score may be generated for each of a plurality of predetermined perspectives. For example, the predetermined plurality of perspectives may be a plurality of perspectives according to the conditions that an investor requires of an investment target. In this case, the score for each of the plurality of perspectives may represent the degree to which the investment target meets the conditions that the investor requires. For example, the plurality of perspectives may be "philosophy," "technology," "scale," "future potential," etc., and the score for the "philosophy" perspective may represent the degree to which the investment target meets the "philosophy" conditions that the investor requires of the investment target, the score for the "technology" perspective may represent the degree to which the investment target meets the "technology" conditions that the investor requires of the investment target, the score for the "scale" perspective may represent the degree to which the investment target meets the "scale" conditions that the investor requires of the investment target, and the score for the "future potential" perspective may represent the degree to which the investment target meets the "future potential" conditions that the investor requires of the investment target.

[0059] For example, the scores of a predetermined number of perspectives may be weighted according to the characteristics of the user U or as specified by the user U. For example, a perspective that the user U considers important may be weighted more heavily than a perspective that the user U does not consider important, thereby emphasizing the relative merits of the scores of the perspectives that the user U considers important. The characteristics of the user U are determined, for example, based on the user information input in step S1.

[0060] For example, the scores of the predetermined multiple perspectives may be weighted according to industry know-how in addition to or instead of the characteristics of the user U. The industry know-how may be accumulated as the system 100 is used, or may be predetermined.

[0061] The scores of a given number of aspects may be combined and expressed as a total score.

[0062] The calculated score may be presented to each user, or information based on the calculated score may be presented to each user. The information based on the calculated score may be, for example, information indicating which company has the highest score or a score higher than a threshold, or a recommendation for an investment. The information based on the calculated score may be, for example, the reason for the score. The reason for the score may be described, for example, by the LLM.

[0063] For example, in step S4, the score calculated by the system 100 is presented to the user U. The score is presented for each interview with each investment target. For example, the PAI of the user C1 of the first company is Ｃ１ Score of the interview with the second company's user C2, Ｃ２ The scores of the interviews with the investor, etc. are presented to the user U. The user U uses the scores as a clue to consider which investment to make.

[0064] In addition to the calculated score or information based on the score, the result of the virtual interview itself may be presented to the user U. This allows the user U to check what kind of conversation took place between the personalized AIs in the virtual interview. For example, if the user U feels that the content of the conversation with his / her personalized AI is not like him / her or is inappropriate, he / she can modify the content of the conversation with his / her personalized AI.

[0065] In addition to the calculated score or information based on the score, the reason for matching with each investment destination may be presented to the user U. The reason for matching may be described by the LLM, for example.

[0066] In step S5, when user U inputs an investment target in which he or she wishes to invest into system 100, user C1 or user C2 is notified in step S6 or step S7 that user U wishes to invest in the investment target. As a result, user C1 or user C2 contacts user U, and actual investment negotiations begin. Alternatively, when user U inputs an investment target in which he or she wishes to invest into system 100 in step S5, an appointment may be automatically booked or the interview (e.g., a telephone interview or a video interview) may automatically begin in step S6 or step S7.

[0067] In this way, the compatibility between the investor and the investment target can be estimated through a virtual interview before the actual investment negotiations, and the actual investment negotiations can proceed between the investor and the investment target who are assumed to be compatible, thereby reducing the possibility that the actual investment negotiations will end in failure.

[0068] Furthermore, by generating personalized AI using detailed information and qualitative information about the investee or investor, a virtual interview will be conducted based on the detailed information and qualitative information about the investee or investor. As a result, the score calculated from the result of the virtual interview will also reflect the detailed information and qualitative information about the investee or investor, which may lead to matching based on the detailed information and qualitative information about the investee or investor.

[0069] For example, when generating a personalized AI, sensitive or confidential information that is difficult to talk about with others may be input. This allows for the exchange of sensitive or confidential information in interviews between personalized AIs. For example, information that would not be shared in interviews between humans can be exchanged in interviews between personalized AIs, which has the advantage of allowing the extraction of information that humans cannot extract.

[0070] The generated personalized AI may be subjected to a re-learning process to more closely resemble the user based on actions taken by the user in consideration of the results of the virtual interview. The action taken by the user in consideration of the results of the virtual interview may, for example, be inputting a desired investment destination. In this case, at least one of the personalized AI of user U and the personalized AI of user C1 of the company may be adjusted so that the score of an interview between the personalized AI of a person in charge (e.g., user C1) of the desired investment destination company and the personalized AI of the investor (e.g., user U) is higher. For example, the weighting of the scores of a predetermined number of perspectives may be adjusted so that the score of an interview between the personalized AI of a person in charge (e.g., user C1) of the desired investment destination company and the personalized AI of the investor (e.g., user U) is higher.

[0071] The action taken by the user in consideration of the results of the virtual interview may be, for example, the user correcting the content of the conversation that the user's personalized AI had in the virtual interview. Ｕ And P.A.I. Ｃ１ PAI in a virtual interview with Ｕ The user U can modify the content of the conversation by, in which case the user U's personalized AI can be adjusted based on the modified content of the conversation (e.g., so that it speaks the modified content of the conversation).

[0072] The action taken by the user in consideration of the results of the virtual interview may be, for example, a conversation between the user and the personalized AI. For example, the user U may directly interact with the personalized AI of the user C1 of the company. For example, the user U may interact with the personalized AI of the user C1 of the company. Ｃ１ In this case, the personalized AI for the user U can be adjusted by learning the content of the interactions that the user U has had (for example, the type of question the user wants answered, etc.).

[0073] 1C. AI Agent-Driven Personalized AI Construction FIG. 1B shows an example of a flow in which an AI agent takes the lead in generating personalized AI in system 100. Here, the AI ​​agent is an AI that assists in matching users and plays a role in eliciting the qualitative needs (deep psychological thoughts) of each user. In order to match users, the AI ​​agent can identify missing user information and generate questions to ask the user for that information. The AI ​​agent can be implemented, for example, by an LLM.

[0074] For example, when user U inputs his / her own information into system 100 in step S1 described above, the AI ​​agent presents a question to user U, and user U can input his / her own information by answering the question.

[0075] Specifically, in step S11, the AI ​​agent (AI ａｇｅｎｔ ) presents a question to the user U.

[0076] In step S12, the user U answers the question, and the answer is provided to the AI ​​agent of the system 100.

[0077] Similarly, for other users, for example, user C1 of the first company, in step S13, the AI ​​agent presents a question to user U, and in step S14, user C1 answers the question, allowing user U to enter his or her own information.

[0078] The AI ​​agent determines whether sufficient information has been collected to generate a personalized AI for user U, based on the information input by user U, or based on the information input by user U and the information input by user C1. For example, by generating personalized AI for a large number of users, it is possible to grasp the average amount of information required to generate a personalized AI that behaves in a way that is characteristic of that person, and the AI ​​agent can determine whether sufficient information has been collected to generate a personalized AI for user U by determining whether the amount of information collected from user U is equal to or greater than that average amount of information.

[0079] Once it is determined that sufficient information has been collected to generate a personalized AI for the user U, the collected information is used to generate a personalized AI PAI for the user U. Ｕ On the other hand, if it is determined that sufficient information has not been collected to generate a personalized AI for user U, the AI ​​agent will present questions to user U to collect the missing information. That is, steps S11 to S12 will be repeated until it is determined that sufficient information has been collected to generate a personalized AI for user U.

[0080] The information collected in step S12 is used to create a personalized AI PAI for the user U. Ｕ By generating a personalized AI, it becomes possible to generate a personalized AI that has learned even the qualitative needs (deep psychological thoughts) of the user U. Ｕ can be generated using information of user U, for example, by fine-tuning the LLM.

[0081] Similarly, for user C1, the AI ​​agent determines whether sufficient information has been collected to generate a personalized AI for user C1 based on the information input by user C1, or based on the information input by user C1 and the information input by user U. If it determines that sufficient information has been collected to generate a personalized AI for user C1, the AI ​​agent uses the collected information to generate a personalized AI PAI for user C1. Ｃ１ If it is determined that sufficient information has not been collected to generate a personalized AI for user C1, steps S13 to S14 are repeated until it is determined that sufficient information has been collected to generate a personalized AI for user C1.

[0082] The information collected in step S14 is used to create a personalized AI PAI for the user C1. Ｃ１ By generating the personalized AI PAI, it becomes possible to generate a personalized AI that has learned even the qualitative needs (deep psychological thoughts) of the user C1. Ｃ１ can be generated using information of user C1, for example, by fine-tuning the LLM.

[0083] The results of interviews between personalized AIs generated in this way can reflect the qualitative needs (deep psychological thoughts) of each user, which can improve the success rate of matching.

[0084] For example, sensitive or confidential information that is difficult to talk about with humans is easier to talk about with an AI agent. Therefore, having an AI agent elicit information can lead to the collection of more information than a human agent could extract. The collected sensitive or confidential information is used when generating a personalized AI, and the personalized AI will exchange such sensitive or confidential information during an interview with the personalized AI.

[0085] In the above example, one personalized AI is generated for each user, but multiple personalized AIs may be generated for each user. For example, for user U, a personalized AI for an interview with user C1 of a first company, a personalized AI for an interview with user C2 of a second company, etc. may be generated. Similarly, for user C1, a personalized AI for an interview with user U, a personalized AI for an interview with another user, and a personalized AI for an interview with yet another user, etc. may be generated.

[0086] In the above examples, the matching between job seekers and employers and between investors and investment targets have been described as examples, but the system 100 of the present invention can be applied to any other matching between users. For example, the system 100 of the present invention can also be applied to matching between users seeking to meet or marry. For example, the system 100 of the present invention can also be applied to matching between users in the real estate field, the M&A field, the e-commerce field, or any other field where matching is used.

[0087] The system 100 of the present invention may have, for example, the following configuration.

[0088] 2. Configuration of a System for Supporting Matching FIG. 2 shows an example of the configuration of a system 100 for supporting matching between users. The users include at least a first user and a second user who can be a potential matching partner for the first user. Preferably, the second user may be multiple users. For example, in the example described above with reference to FIG. 1A , the first user may be user U who is a job seeker or user U who is an investor, and the multiple second users may be users C1 and C2 of multiple companies that are employers or users C1 and C2 of multiple companies that are investment targets.

[0089] The system 100 is connected to a database unit 200. The system 100 is further connected to at least one terminal device 300 via a network 400.

[0090] 2 shows three terminal devices 300, the number of terminal devices 300 is not limited to this. Any number of terminal devices 300 may be connected to the system 100 via the network 400.

[0091] The network 400 may be any type of network. For example, the network 400 may be the Internet or a LAN. The network 400 may be a wired network or a wireless network.

[0092] An example of the system 100 is, but is not limited to, a computer (e.g., a server device) installed at a provider that provides a service that supports matching. An example of the terminal device 300 is, but is not limited to, a computer (e.g., a terminal device) used by a user who is a consumer of the service that supports matching. The terminal device 300 may be, for example, the terminal device of user U (a job seeker, an investor, or a person in charge of an investment company) in the example described above with reference to FIG. 1A , as well as the terminal devices of users C1 and C2 (a person in charge of a company that is an employer or a person in charge of a rare disease that is an investment target).

[0093] Here, the computer (server device or terminal device) may be any type of computer, such as a smartphone, a tablet, a personal computer, smart glasses, a smart watch, or the like.

[0094] Data for generating personalized AI may be associated with a user and stored in the database unit 200. Alternatively, generated personalized AI may be stored in the database unit 200. The system 100 may read the personalized AI from the database unit 200 and perform processing using the read personalized AI.

[0095] FIG. 3 shows an example of a specific configuration of the system 100.

[0096] The system 100 comprises an interface section 110, a processor section 120, and a memory section 130.

[0097] The interface unit 110 exchanges information with the outside of the system 100. The processor unit 120 of the system 100 can receive information from the outside of the system 100 and can send information to the outside of the system 100 via the interface unit 110. The interface unit 110 can exchange information in any format.

[0098] The interface unit 110 includes, for example, an input unit that allows information to be input to the system 100. It does not matter how the input unit allows information to be input to the system 100. For example, if the input unit is a receiver, the receiver may input information by receiving information from outside the system 100 via a network. Alternatively, if the input unit is a data reading device, the input unit may input information by reading information from a storage medium connected to the system 100.

[0099] The interface unit 110 includes, for example, an output unit that enables information to be output from the system 100. It does not matter in what manner the output unit enables information to be output from the system 100. For example, if the output unit is a transmitter, the transmitter may output information by transmitting it to an external device outside the system 100 via a network. Alternatively, if the output unit is a data writing device, the output unit may output information by writing it to a storage medium connected to the system 100.

[0100] The system 100 can transmit information to and / or receive information from the database unit 200 via the interface unit 110, for example. The system 100 can transmit information to and / or receive information from the terminal device 300 via the interface unit 110, for example.

[0101] The system 100 can receive information or requests from the terminal device 300, for example, via the interface unit 110. The received information can be, for example, information used to generate a personalized AI for the user. The request can be a request to have the personalized AI interact with the personalized AI of another user, a request to select the personalized AI as a matching partner, a request to retrain the personalized AI, a request to change the weighting of a perspective for a score, etc. For example, the system 100 can output information, for example, via the interface unit 110. The output information can be, for example, a score calculated by the system 100, information based on the calculated score (e.g., information representing the user with the highest score), etc.

[0102] The processor unit 120 executes the processing of the system 100 and controls the overall operation of the system 100. The processor unit 120 reads and executes a program stored in the memory unit 130. This allows the system 100 to function as a system that executes desired steps. The processor unit 120 may be implemented by a single processor or by multiple processors.

[0103] The memory unit 130 stores programs required to execute the processing of the system 100, data required to execute the programs, and the like. The memory unit 130 may also store a program for causing the processor unit 120 to execute processing to support matching (e.g., a program that realizes the processing shown in FIG. 5 , which will be described later). Here, how the program is stored in the memory unit 130 is not important. For example, the program may be pre-installed in the memory unit 130. Alternatively, the program may be installed in the memory unit 130 by being downloaded via a network. In this case, the type of network does not matter. The memory unit 130 may be implemented by any storage means. Alternatively, the program may be stored in a non-transitory computer-readable storage medium and installed in the memory unit 130 by reading the storage medium.

[0104] The database unit 200 may store, for example, data for constructing a personalized AI associated with a user. The personalized AI is an AI that can learn personal information to perform activities that are characteristic of the individual, more specifically, generate sentences that are characteristic of the individual in response to questions. For example, the personalized AI may be a personalized language model (LM) or a personalized large language model (LLM) that can learn an individual's expressions and generate sentences according to the individual's expressions. For example, the database unit 200 may store personal information, expressions, thoughts, etc. as data for generating the personalized AI. The data for generating the personalized AI may be data entered by a user into the system 100 or existing data on a network (e.g., data posted by a user on a social networking site or data written by a user on a blog). The data for generating the personalized AI may be long-term memory or short-term memory. The database unit 200 may also store know-how in the field in which matching is performed. The know-how may be stored in advance in the database unit 200, or may be accumulated as matching is actually performed.

[0105] Alternatively, for example, generated personalized AI (for example, a trained model that has learned personal information) is stored in the database unit 200. The system 100 can read the personalized AI from the database unit 200 and execute processing using the read personalized AI.

[0106] In the examples shown in FIGS. 2 and 3 , the database unit 200 is provided outside the system 100, but the present invention is not limited to this. The database unit 200 can also be provided inside the system 100. In this case, the database unit 200 may be implemented by the same storage means as the storage means that implements the memory unit 130, or by a storage means different from the storage means that implements the memory unit 130. In either case, the database unit 200 is configured as a storage unit for the system 100. The configuration of the database unit 200 is not limited to a specific hardware configuration. For example, the database unit 200 may be configured as a single hardware component or multiple hardware components. For example, the database unit 200 may be configured as an external hard disk drive for the system 100, or as cloud storage connected via a network.

[0107] FIG. 4A shows an example of the configuration of the processor unit 120.

[0108] The processor unit 120 includes a first reading means 121 , a second reading means 122 , a recording means 123 , a calculating means 124 , and an output means 125 .

[0109] The first reading means 121 is configured to read out the personalized AI of the first user (i.e., the first personalized AI). The first reading means 121 can read out the first personalized AI stored in the database unit 200, for example. Alternatively, the first reading means 121 may read out the first personalized AI stored in a storage means external to the system 100 connected via the network 400, for example.

[0110] The read first personalized AI is passed to the recording means 123 .

[0111] The second reading means 122 is configured to read out the personalized AI of each of the plurality of second users (i.e., a plurality of second personalized AIs). The second reading means 122 can read out the second personalized AI stored in the database unit 200, for example. Alternatively, the second reading means 122 may read out the second personalized AI stored in a storage means external to the system 100 connected via the network 400, for example.

[0112] The read second personalized AI is passed to the recording means 123 .

[0113] The recording means 123 is configured to record a conversation between the first personalized AI and each of the plurality of second personalized AIs.

[0114] For example, the first personalized AI and the second personalized AI are connected so that an output from the first personalized AI is reflected in a prompt of one of the second personalized AIs, and an output from the second personalized AI is reflected in a prompt of the first personalized AI. Then, a conversation between the first personalized AI and the second personalized AI is initiated by inputting an appropriate triggering prompt into either the first personalized AI or the second personalized AI. Because the first personalized AI and the second personalized AI are connected as described above, the conversation between the first personalized AI and the second personalized AI progresses as the output from the first personalized AI triggers an output from the second personalized AI, and the output from the second personalized AI triggers an output from the first personalized AI. The recording means 123 stores this conversation in association with the second personalized AI.

[0115] The conversation proceeds in a similar manner with another second personalized AI among the plurality of second personalized AIs, and the recording means 123 stores this conversation in association with the other second personalized AI.

[0116] The recording means 123 may, for example, store the conversation in the database unit 200, or may store the conversation in the memory unit 130 of the system 100. For example, the recording means 123 may temporarily store the conversation in the memory unit 130 and then store it in the database unit 200.

[0117] The calculation means 124 is configured to calculate a score for each of the plurality of second personalized AIs based on the recorded conversation.

[0118] Each score may represent the degree of relevance between the output from a first personalized AI and the output from one second personalized AI. The higher the relevance, the better the compatibility between the first personalized AI and the second personalized AI, and the lower the relevance, the worse the compatibility between the first personalized AI and the second personalized AI. In this way, the score may represent the compatibility between the first personalized AI and the second personalized AI.

[0119] The calculation means 124 can calculate a score by vectorizing the output from the first personalized AI and the output from one second personalized AI, and calculating the distance or similarity between those vectors. At this time, the calculation means 124 can, for example, reflect information about the first user (e.g., information input in step S1 of FIG. 1A) in the vector, and can reflect information about the second user (e.g., information input in step S2 of FIG. 1A) in the vector. This allows the score to reflect the characteristics of the first user and the second user.

[0120] The calculation means 124 may calculate a score by, for example, embedding the output from the first personalized AI and the output from one second personalized AI. Alternatively, the calculation means 124 may calculate a score using, for example, an LLM. The LLM generally learns how relevant each type of conversation is, and therefore can generate a score by referring to the output from the first personalized AI and the output from one second personalized AI.

[0121] The calculation means 124 may calculate scores for, for example, a plurality of predetermined perspectives. The predetermined plurality of perspectives may be changed depending on the field in which matching is performed. For example, in the case of matching between a job seeker and an employer, the plurality of perspectives may be perspectives such as "salary," "welfare," "ease of working," and "future prospects." In the case of matching between an investor and an investment target, the plurality of perspectives may be perspectives such as "philosophy," "technology," "size," and "future prospects." Furthermore, the number of predetermined plurality of perspectives may be any number of two or more, and preferably five or more or ten or more. The greater the number of predetermined perspectives, the more detailed the scoring becomes. The calculation means 124 may calculate scores for the predetermined plurality of perspectives using, for example, LLM.

[0122] The calculation means 124 may calculate scores for a plurality of predetermined perspectives according to, for example, weighting. For example, weighting may be performed so that the perspective that the user considers more important receives a higher score. The weighting may be determined based on, for example, input from the user. The weighting may be determined using, for example, an LLM, which learns the degree of association between user information (e.g., user characteristics, user answers to questions, etc.) and perspectives that the user may consider important.

[0123] For example, the calculation means 124 can input user information into the LLM to determine weightings for a predetermined number of perspectives, and can calculate scores for a predetermined number of perspectives by inputting the weightings and recorded conversations between personalized AIs into the LLM.

[0124] The calculation means 124 may update the weights as described below and calculate the scores of a predetermined number of perspectives in accordance with the updated weights.

[0125] The calculation means 124 may calculate an overall score based on the scores of a plurality of predetermined perspectives. The overall score may be, for example, a score obtained by simply adding up the scores of the plurality of predetermined perspectives, or may be a score obtained by substituting the scores of the plurality of predetermined perspectives into a predetermined function. The predetermined function may be a function that specifies the weighting of each of the scores of the plurality of predetermined perspectives. The predetermined function may be, for example, constructed for each user, or for each field in which matching is performed.

[0126] The calculated score is passed to the output means 125 .

[0127] The output unit 125 is configured to output the calculated score or information based on the calculated score. The information based on the calculated score may be, for example, information indicating which of the multiple second users has a high score, or recommendation information indicating which of the multiple second users should be matched with.

[0128] The output means 125 can output the calculated score or information based on the calculated score to the terminal device 300 of the first user (i.e., user U) via the interface unit 110. The output means 125 may output the calculated score or information based on the calculated score to the terminal device 300 of the second user (i.e., user C1 or user C2).

[0129] After the calculated score or information based on the calculated score is output, the first user can input which of the multiple second users the first user wishes to match with. For example, as in the example described above with reference to FIG. 1A, user U, a job seeker, can input which of the multiple companies offering jobs the first user wishes to work for. For example, as in the example described above with reference to FIG. 1A, user U, an investor, can input which of the multiple investment companies the first user wishes to invest in.

[0130] For example, if a first user selects a second user despite a low score, it is assumed that the first personalized AI was not able to perform the same activities as the first user, or that the second personalized AI was not able to perform the same activities as the second user, i.e., it is assumed that the accuracy of the first personalized AI or the second personalized AI was not high. In this case, the accuracy of the first personalized AI or the second personalized AI can be improved by performing a re-learning process on the first personalized AI or the second personalized AI.

[0131] For example, the processor unit 120 of the system 100 may receive an input in which a first user selects one of a plurality of second users, and may perform a re-learning process on the first personalized AI so that the selected second user has a higher score with the second personalized AI, thereby improving the accuracy of the first personalized AI.

[0132] For example, the processor unit 120 of the system 100 may receive an input in which a first user selects one of a plurality of second users, and may perform a re-learning process on the second personalized AI so that the selected second user has a higher score with the second personalized AI, thereby improving the accuracy of the second personalized AI.

[0133] For example, the processor unit 120 of the system 100 may receive input from a first user selecting one of a plurality of second users, and may perform a re-learning process on both the first personalized AI and the second personalized AI so that the selected second user has a higher score with the second personalized AI, thereby improving the accuracy of both the first personalized AI and the second personalized AI.

[0134] For example, if a first user selects a second user despite a low score, it is assumed that the weighting of a plurality of predetermined aspects was inappropriate. In this case, the accuracy of the score calculation can be improved by updating the weighting.

[0135] For example, the processor unit 120 of the system 100 may receive an input in which a first user selects one of a plurality of second users, and may update the weightings for a plurality of predetermined perspectives so that the selected second user has a higher score with the second personalized AI. For example, the weightings may be adjusted so that the weight of the perspective with the highest score among the scores of the plurality of perspectives when no weighting is performed is increased and the weight of the perspective with the lowest score is decreased. Alternatively, the weightings may be adjusted using, for example, an LLM.

[0136] 4B shows an example of the configuration of processor unit 120', which is an alternative embodiment of processor unit 120. Processor unit 120' differs from processor unit 120 in that it includes generation means 126. By including generation means 126, processor unit 120', unlike processor unit 120, uses personalized AI generated by generation means 126. Here, components similar to those described above for processor unit 120 are given the same reference numerals, and detailed description thereof will be omitted.

[0137] The generating means 126 is configured to generate the first personalized AI and / or the second personalized AI.

[0138] The generation means 126 can perform a learning process by fine-tuning a language model (preferably a large-scale language model) using, for example, sentences expressed by a user. The sentences expressed by a user may be, for example, data input by the user to the system 100, or may be existing data on a network (for example, data posted by the user on a social networking site or data written by the user on a blog). The more sentences expressed by the user, the more preferable it is, and the number may be, for example, 50 or more sentences, 100 or more sentences, 1,000 or more sentences, 10,000 or more sentences, or 100,000 or more sentences.

[0139] When fine-tuning is performed using data input by a user to the system 100, the generator 126 can present questions to the user and receive answers to the questions from the user, which can be used as sentences expressed by the user.

[0140] At this time, the generating means 126 can calculate a hallucination score for the received answer. The hallucination score is a score that indicates whether the received answer is false or not. The higher the hallucination score, the more likely it is false. The generating means 126 can identify inconsistencies or contradictions between the user's past answers and the received answer, and calculate a hallucination score based on the identified results. For example, the more inconsistencies or contradictions there are, the higher the hallucination score may be.

[0141] The generation means 126 determines whether the hallucination score satisfies a threshold, and if it is determined that the threshold is met, the received answer can be used to generate the personalized AI. On the other hand, if it is determined that the threshold is not met, the received answer is not used to generate the personalized AI.

[0142] The generator 126 may, for example, implement the AI ​​agent described above with reference to Figure 1B. Thus, the questions presented to the user may be provided by the AI ​​agent.

[0143] The first reading means 121 can read the personalized AI of the first user generated by the generating means 126 .

[0144] The second reading means 122 can read out a plurality of personalized AIs of a plurality of second users generated by the generating means 126 .

[0145] The recording means 123 is configured to record a conversation between the first personalized AI and each of the plurality of second personalized AIs.

[0146] The calculation means 124 is configured to calculate a score for each of the plurality of second personalized AIs based on the recorded conversation.

[0147] The output means 125 is configured to output the calculated score or information based on the calculated score.

[0148] After the calculated score or information based on the calculated score is output, the first user can input which of the multiple second users the first user wishes to be matched with. For example, as in the example described above with reference to FIG. 1A , user U, who is a job seeker, can input which of the multiple companies hiring, the first user wishes to work for.

[0149] 4A and 4B, the components of the processor unit 120 or 120' are provided within the same processor unit 120 or 120', but the present invention is not limited to this. A configuration in which the components of the processor unit 120 or 120' are distributed across multiple processor units is also within the scope of the present invention. In this case, the multiple processor units may be located within the same hardware component, or may be located within separate hardware components located nearby or remotely.

[0150] Each component of the system 100 described above may be composed of a single hardware component, or may be composed of multiple hardware components. When composed of multiple hardware components, the manner in which the hardware components are connected does not matter. The hardware components may be connected wirelessly or by wire. The system 100 of the present invention is not limited to a specific hardware configuration. It is also within the scope of the present invention that the processor unit 120 or the processor unit 120' is configured using analog circuits rather than digital circuits. The configuration of the system 100 of the present invention is not limited to the one described above, as long as it can realize its functions.

[0151] 5 is a flowchart showing an example of a process 500 in the system 100 for supporting matching. The process 500 may be executed in the processor unit 120 or the processor unit 120′. The following description will be given taking the example of the process 500 being executed in the processor unit 120.

[0152] In step S501, the first reading means 121 of the processor unit 120 reads the personalized AI of the first user (i.e., the first personalized AI). The first reading means 121 can read the first personalized AI stored in the database unit 200, for example. Alternatively, the first reading means 121 may read the first personalized AI stored in a storage means external to the system 100 connected via the network 400, for example. Alternatively, when the process 500 is executed in the processor unit 120′, the first reading means 121 can read the personalized AI of the first user generated by the generation means 126 before step S501.

[0153] In step S502, the second reading means 122 of the processor unit 120 reads out the personalized AI of each of the multiple second users (i.e., multiple second personalized AIs). The second reading means 122 can read out the second personalized AI stored in the database unit 200, for example. Alternatively, the second reading means 122 can read out the second personalized AI stored in a storage means external to the system 100 connected via the network 400, for example. Alternatively, when the process 500 is executed in the processor unit 120', the second reading means 122 can read out the personalized AI of the second user generated by the generation means 126 before step S501.

[0154] In step S503, the recording means 123 of the processor unit 120 records conversations with the first personalized AI read out in step S501 and each of the multiple second personalized AIs read out in step S502.

[0155] For example, the first personalized AI and the second personalized AI are connected so that an output from the first personalized AI is reflected in a prompt of one of the second personalized AIs, and an output from the second personalized AI is reflected in a prompt of the first personalized AI. Then, a conversation between the first personalized AI and the second personalized AI is initiated by inputting an appropriate triggering prompt into either the first personalized AI or the second personalized AI. Because the first personalized AI and the second personalized AI are connected as described above, the conversation between the first personalized AI and the second personalized AI progresses as the output from the first personalized AI triggers an output from the second personalized AI, and the output from the second personalized AI triggers an output from the first personalized AI. The recording means 123 stores this conversation in association with the second personalized AI.

[0156] In step S504, the calculation means 124 of the processor unit 120 calculates a score for each of the plurality of second personalized AIs based on the conversation recorded in step S503. Each score may represent the degree of relevance between the output from the first personalized AI and the output from one of the second personalized AIs.

[0157] The calculation means 124 can calculate a score by vectorizing the output from the first personalized AI and the output from one second personalized AI, and calculating the distance or similarity between those vectors. At this time, the calculation means 124 can, for example, reflect information about the first user (e.g., information input in step S1 of FIG. 1A) in the vector, and can reflect information about the second user (e.g., information input in step S2 of FIG. 1A) in the vector. This allows the score to reflect the characteristics of the first user and the second user.

[0158] The calculation means 124 may calculate a score by, for example, embedding the output from the first personalized AI and the output from one second personalized AI. Alternatively, the calculation means 124 may calculate a score using, for example, an LLM. The LLM generally learns how relevant each type of conversation is, and therefore can generate a score by referring to the output from the first personalized AI and the output from one second personalized AI.

[0159] The calculation means 124 may calculate scores for a plurality of predetermined perspectives, for example. The calculation means 124 can calculate the scores for a plurality of predetermined perspectives, for example, according to weighting. For example, weighting can be performed so that the perspective that the user considers more important receives a higher score. The weighting can be determined, for example, based on input from the user. The weighting can be determined, for example, using an LLM, where the LLM learns the degree of association between the user's information and the perspective that the user may consider important.

[0160] The calculation means 124 may update the weights in accordance with the user's actions on the matching results, and calculate the scores for a plurality of predetermined perspectives in accordance with the updated weights. The weights may be updated automatically when the user takes an action on the matching results.

[0161] In step S505, the output means 125 of the processor unit 120 outputs the score calculated in step S504 or information based on the calculated score. The output means 125 can output the calculated score or information based on the calculated score to the terminal device 300 of the first user (i.e., user U) via the interface unit 110, for example. The output means 125 may also output the calculated score or information based on the calculated score to the terminal device 300 of the second user (i.e., user C1 or user C2).

[0162] After the calculated score or information based on the calculated score is output, the first user can input which of the multiple second users he or she wishes to be matched with. In this way, each user can find a compatible match before actually conducting an interview, etc. In addition, the score reflects detailed information and qualitative information about the user, which can enable matching based on the user's qualitative desires (deep psychological thoughts).

[0163] In the example described above with reference to FIG. 5, the processes are described as being performed in a specific order, but the order of the processes is not limited to that described and may be performed in any order that is logically possible.

[0164] In the example described above with reference to Fig. 5, it has been explained that the processing of each step shown in Fig. 5 is realized by the processor unit 120 or 120' and a program stored in the memory unit 130, but the present invention is not limited to this. At least one of the processing of each step shown in Fig. 5 may be realized by a hardware configuration such as a control circuit.

[0165] The present invention is not limited to the above-described embodiments. It is understood that the scope of the present invention should be interpreted only by the claims. It is understood that a person skilled in the art can implement an equivalent scope based on the description of the present invention and common technical knowledge from the description of specific preferred embodiments of the present invention.

[0166] The present invention is useful in providing a method for supporting matching between users, which enables highly accurate matching results to be easily obtained.

[0167] U, C1, C2 User PAI Ｕ , P.A.I. Ｃ１ , P.A.I. Ｃ２ Personalized AI 100 System 200 Database unit 300 Terminal device 400 Network

Claims

1. A method for assisting in matching between users, comprising: retrieving a first personalized AI of a first user; retrieving a plurality of second personalized AIs of a plurality of second users; recording a conversation between the first personalized AI and each of the plurality of second personalized AIs; calculating a score for each of the plurality of second personalized AIs based on the conversation; and outputting the score or information based on the score.

2. The method of claim 1, wherein the scores include scores of a plurality of predetermined aspects, and calculating the scores includes calculating the scores of the plurality of predetermined aspects according to weightings.

3. The method of claim 2, further comprising updating the weightings.

4. The method of claim 3, further comprising: after outputting the score or the information, receiving an input from the first user selecting one of the plurality of second users; and updating the weightings includes updating the weightings such that the selected second user has a higher score with the second personalized AI.

5. The method of claim 1, further comprising: after outputting the score or the information, receiving an input in which the first user selects one of the plurality of second users; and performing a re-learning process on the first personalized AI so that the selected second user has a higher score with the second personalized AI.

6. The method of claim 1, further comprising: generating the first personalized AI; reading the first personalized AI comprises reading the generated first personalized AI; and generating the first personalized AI comprises performing a learning process by fine-tuning a language model using sentences expressed by the first user.

7. The method of claim 6, wherein generating the first personalized AI further comprises: presenting a question to the first user; and receiving an answer to the question from the first user, wherein a sentence expressed by the first user comprises the answer.

8. The method of claim 7, further comprising providing an AI agent to support the first user, wherein the questions are provided by the AI ​​agent.

9. The method of claim 7 or claim 8, further comprising: calculating a hallucination score for the answer; and determining whether the hallucination score meets a threshold, wherein if the hallucination score meets the threshold, the answer is used to generate the first personalized AI; and if the hallucination score does not meet the threshold, the answer is not used to generate the first personalized AI.

10. A system for assisting in matching users, comprising: a first reading means for reading a first personalized AI of a first user; a second reading means for reading a plurality of second personalized AIs of a plurality of second users; a recording means for recording a conversation between the first personalized AI and each of the plurality of second personalized AIs; a calculation means for calculating a score for each of the plurality of second personalized AIs based on the conversation; and an output means for outputting the score or information based on the score.

11. A program for assisting in matching between users, the program being executed in a computer system having a processor unit, the program causing the processor unit to execute processes including: reading a first personalized AI of a first user; reading a plurality of second personalized AIs of a plurality of second users; recording conversations between the first personalized AI and each of the plurality of second personalized AIs; calculating a score for each of the plurality of second personalized AIs based on the conversations; and outputting the score or information based on the score.

Images