Person matching support device

Abstract

The present invention provides a person matching support device that assists in individually associating multiple individuals included in different lists of people. [Solution] The person matching support device 1 includes: a first prompt generation module 11 that acquires first characteristic data for each of a plurality of first people and second characteristic data for each of a plurality of second people, and generates a first prompt that causes a first generation AI 3 to generate first persona data for each of the plurality of first people and second persona data for each of the plurality of second people; and a second prompt generation module 12 that generates a second prompt that requests the second generation AI 3 to individually associate the plurality of first people with the plurality of second people based on the first persona data and the second persona data.

Description

[Technical Field]

[0001] The present invention relates to a plurality of first persons a included in a first person list. j And, multiple second person b included in the second person list i This relates to a person matching support device that assists in individually associating individuals with each other. [Background technology]

[0002] Multiple first person a j And several second person b i When the same person is included in both lists, integrating these lists allows for more precise data analysis. Patent No. 6158464 of this applicant is a technology that combines and analyzes information when a first entity holds information about a first person and a second entity holds information about a second person, while minimizing the extent to which each entity's information is disclosed to the other. [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] Patent No. 6158464 [Overview of the project] [Problems that the invention aims to solve]

[0004] In the aforementioned Patent No. 6158464, the same person is identified as the first person and the second person by using an email address or a combination of address and name as a unique key. However, how to identify the same person when such a unique key does not exist has not been considered.

[0005] One aspect of the present invention described below is the ability to identify multiple first persons a without a unique key j and several second person b i This relates to individually associating with the first entity. In this invention, the first entity is the first person aj having the information of j , and not limited to the case where the second entity has the information of the second person b i but also applicable to the case where the same entity has the list of the first persons a j and the list of the second persons b i of i . In the present invention, it is not always necessary to determine the same person, and it is only necessary to be able to perform the association.

Means for Solving the Problem

[0006] A person association support device according to one aspect of the present invention includes a first prompt generation module that acquires the first feature data of each of a plurality of first persons a j and the second feature data of each of a plurality of second persons b i and generates a first prompt for causing the first generation AI to generate the first persona data of each of the plurality of first persons a j and the second persona data of each of the plurality of second persons b i ; and a second prompt generation module that generates a second prompt for requesting the second generation AI to individually associate the plurality of first persons a j with the plurality of second persons b i based on the first and second persona data. The person association support device includes the above.

Brief Description of the Drawings

[0007] [Figure 1] FIG. 1 shows a person association support device 1 according to a first embodiment and an external device connected to the person association support device 1. [Figure 2] FIG. 2 shows the functions and operations of the person association support device 1 according to the first embodiment. [Figure 3] FIG. 3 shows an example of the first feature data of a plurality of first persons aj. [Figure 4] FIG. 4 shows an example of the second feature data of a plurality of second persons bi. [Figure 5]Figure 5 shows an example of a first prompt for generating first persona data for multiple first individuals aj. [Figure 6] Figure 6 shows an example of the first prompt for generating second persona data for multiple second individuals (BIs). [Figure 7] Figure 7 shows the persona definition sent to the generating AI3 along with the first prompt. [Figure 8] Figure 8 shows an example of the first persona data for multiple first individuals aj. [Figure 9] Figure 9 shows examples of second persona data for multiple second individuals (BIs). [Figure 10] Figure 10 shows the specifications of the input file that is sent to the generating AI3 along with the second prompt. [Figure 11] Figure 11 shows part of the second prompt, which requires that multiple first persons aj and multiple second persons bi be individually associated. [Figure 12] Figure 12 shows part of the second prompt, which requires that multiple first persons aj and multiple second persons bi be individually associated. [Figure 13] Figure 13 shows part of the second prompt, which requires that multiple first persons aj and multiple second persons bi be individually associated. [Figure 14] Figure 14 shows part of the second prompt, which requires that multiple first persons aj and multiple second persons bi be individually associated. [Figure 15] Figure 15 shows an example of a C file returned by the generated AI3 in response to the second prompt. [Figure 16] Figure 16 shows the functions and operation of the person matching support device 1a according to the second embodiment. [Figure 17] Figure 17 shows an example of the first judgment data. [Figure 18] Figure 18 shows an example of the evaluation weight sij obtained from the first judgment data. [Figure 19] Figure 19 shows an example of the second judgment data. [Figure 20] Figure 20 shows an example of a confidence score (CJI) obtained from the second judgment data. [Figure 21] Figure 21 shows an example of a judgment score Jij calculated from evaluation weights sij and confidence score cji. [Figure 22] Figure 22 shows an example of data for scoring by the scoring module 15 in the third embodiment. [Figure 23] Figure 23 is a flowchart showing scoring example 1 in the third embodiment. [Figure 24] Figure 24 is a flowchart showing example 2 of scoring in the third embodiment. [Figure 25] Figure 25 is a flowchart showing scoring example 3 in the third embodiment. [Modes for carrying out the invention]

[0008] Embodiments of the present invention will be described in detail below with reference to the drawings. Each embodiment described below is an example of the present invention and does not limit the scope of the present invention. Furthermore, not all of the configurations and operations described in each embodiment are necessarily essential to the configurations and operations of the present invention. The same reference numerals are used for identical components, and redundant explanations are omitted.

[0009] <1. First Embodiment> <1-1. Structure> Figure 1 shows a person matching support device 1 according to the first embodiment and external devices connected to the person matching support device 1. The person matching support device 1 is a computer system equipped with a CPU, memory, etc. (not shown). The person matching support device 1 may consist of a single computer or multiple computers connected via a network. The person matching support device 1 is connected to external devices such as a database 2 and a generation AI 3.

[0010] Database 2 stores the first and second feature data. The first and second feature data will be explained with reference to Figures 3 and 4. Database 2 is not limited to data stored in a single storage device; it may also be data distributed across multiple storage devices.

[0011] The Generative AI3 includes a Large-Scale Language Model (LLM). The LLM is a language model constructed using a large amount of text data and deep learning techniques, and it processes tasks such as answering questions, translating, generating text, summarizing text, and sentiment analysis in response to prompts sent from the Human Correspondence Support Device 1. It is desirable that the LLM includes an attention mechanism. The attention mechanism is a mechanism that extracts important parts from the input and greatly contributes to improving the processing speed and accuracy of the LLM.

[0012] The person matching support device 1 acquires various data from the database 2, generates a prompt and sends it to the generation AI 3, and obtains a response from the generation AI 3.

[0013] <1-2. Functions and Operation> Figure 2 shows the functions and operation of the person matching support device 1 according to the first embodiment. The person matching support device 1 includes a first prompt generation module 11, a second prompt generation module 12, and a scoring module 15. These modules are implemented by loading programs into the memory included in the person matching support device 1 and executing them by the CPU.

[0014] The first prompt generation module 11 generates multiple first persons a j Each of the first characteristic data and multiple second person b i The second feature data for each of the first person a is obtained from database 2 to generate the first prompt. The first prompt is obtained from multiple first person a j Each of the first persona data and multiple second person b iThis is a prompt requesting the Generator AI3 to generate the second persona data for each of the two individuals. The first prompt will be explained with reference to Figures 5 to 7, and the first and second persona data generated by the Generator AI3 will be explained with reference to Figures 8 and 9.

[0015] The second prompt generation module 12 generates a second prompt. The second prompt generates multiple first person a based on the first and second persona data. j and several second person b i This prompt requests the generating AI3 to individually associate the two elements. The second prompt will be explained with reference to Figures 10 to 15.

[0016] The scoring module 15 obtains the first explanatory variable and target variable for each of the multiple learning targets, performs machine learning on the combination of the first explanatory variable and target variable for each learning target, and creates a model that shows the relationship between the first explanatory variable and the target variable. Furthermore, the scoring module 15 obtains the second explanatory variable for the prediction target and calculates a score for the prediction target by inputting the second explanatory variable into the model. The scoring will be explained with reference to Figures 22 to 25.

[0017] In this application, the generation AI3 that generates a persona in response to the first prompt is sometimes referred to as the first generation AI, and the generation AI3 that performs the matching in response to the second prompt is sometimes referred to as the second generation AI. Furthermore, the generation AI3 that performs the matching in response to the third prompt (see Figure 16) is sometimes referred to as the third generation AI. The first to third generation AIs may be the same generation AI or may be different generation AIs.

[0018] <1-3. Characteristic data for the first and second features> Figure 3 shows multiple first persons a j An example of the first feature data is shown. The first feature data is multiple first person a j Each of these includes "ID_aj", "age", and numerical data representing the results of the employee's job aptitude test. "ID_aj" is the first person aj This is an ID that identifies each of them. The results of the job aptitude test include scales such as "degree of social avoidance" and "degree of self-reflection."

[0019] Figure 4 shows multiple second persons b i An example of the second feature data is shown. The second feature data is multiple second person b i For each of these, the data includes "ID_bi", "Department", "Position", and "Age", as well as evaluation data from the HR department based on performance reviews. "ID_bi" is the second person b i Each of these is an ID that identifies it, and is described as a natural number. The HR department's evaluation data includes metrics such as "performance evaluation" and "skill evaluation."

[0020] Multiple first person a j and several second person b i The relationship is unclear. The first and second characteristic data do not necessarily have to include personally identifiable information such as names or dates of birth. Also, in the first and second characteristic data, "age" is expressed in 10-year increments such as 20s, 30s, 40s, etc. The "age" data is an example of category data in this invention.

[0021] The first and second characteristic data do not have to be employee data; for example, they could be customer data.

[0022] <1-4. First Prompt> Figure 5 shows multiple first persons a j An example of the first prompt for generating the first persona data is shown. Figure 6 shows multiple second person b i An example of the first prompt for generating the second persona data is shown. Figure 7 shows the persona definition sent to the generating AI3 along with the first prompt.

[0023] In both the first prompt shown in Figures 5 and 6, the generating AI3 is asked to determine which of the "Type 1" to "Type 5" categories included in the persona definition in Figure 7 each person belongs to. "Type 1" to "Type 5" represent a general classification of each person's persona.

[0024] In both the first prompts shown in Figures 5 and 6, the generating AI3 is requested to generate text describing each person's persona. In Figure 5, multiple first person a j The task requires a description of each of the following: their strengths, weaknesses, and points to consider as a superior. Figure 6 shows multiple second-person b i We are requesting that you describe each individual's characteristics and provide comments that encourage their future growth.

[0025] While the case of generating different first prompts in Figures 5 and 6 to obtain first and second persona data has been described, the present invention is not limited to this. A common first prompt may be generated to obtain first and second persona data.

[0026] <1-5. Persona Data for the First and Second Personas> Figure 8 shows multiple first persons a j An example of the first persona data is shown. The first persona data generated by Generator AI3 consists of multiple first individuals a j Each entry includes "ID_aj," "Type," "Strengths," "Weaknesses," and "Comments." The "Comments" are text describing points to note from the supervisor's perspective.

[0027] Figure 9 shows multiple second persons b i Here is an example of the second persona data. The second persona data generated by Generator AI3 is a combination of multiple second person b i Each of these includes "ID_bi", "Type", and "Comment". The "Comment" is text that describes comments to encourage the individual to grow in the future.

[0028] In the first and second persona data, the "Type" is one of "Type 1" through "Type 5". The "Type" data is an example of category data in this invention.

[0029] <1-6. Second Prompt> Figure 10 shows the specifications of the input files sent to the generating AI3 along with the second prompt. The input files include files A, B, and C. File A contains multiple first person a j In addition to the first persona data, it also includes the "age" data contained in the first characteristic data (see Figures 3 and 8). File B contains data on multiple second individuals b i In addition to the second persona data, it also includes the "age" data included in the second characteristic data (see Figures 4 and 9). File C contains multiple first person a j and several second person b i This file records the results of individual mappings and includes data for "ID_bi", "age", "type", and "ID_aj". The C file will be explained with reference to Figure 15.

[0030] Figures 11-14 show multiple first persons a j and several second person b i An example of a second prompt for requiring individual mappings is shown. Figures 11-14 together constitute a single prompt. The second prompt includes descriptions of the "input file description," "analysis method," and "start and end" shown in Figure 11, the "procedure" shown in Figures 12 and 13, and the "release conditions," "end conditions," and "output conditions" shown in Figure 14.

[0031] The "procedure" shown in Figures 12 and 13 will be explained in detail. The "procedure" includes steps 1 to 11 and processes 1 to 6 which define the "recursive processing". Depending on the processing capabilities of the generating AI3, it is desirable to divide the "procedure" in Figures 12 and 13 into multiple prompts and process them sequentially while obtaining responses, rather than providing all steps at once in a single prompt.

[0032] In step 1, the ID_bi to be analyzed is fixed to a natural number k. If the initial value of k is, for example, ID_b11, then ID_b11 is the first to be analyzed. In step 2, for ID_bi=k which was fixed in step 1, the type and age are read from file B. In step 3, select ID_aj from file A that is of the same type and age as the subject to be analyzed, and set it as the "initial set". The initial set may be ID_a11 to ID_a14. In step 4, ID_aj, which is written in the C file, is excluded from the initial set and designated as a "candidate for determination". In other words, individuals who have not yet been associated are designated as candidates for determination. If no individuals have already been associated, ID_aj is not written in the C file, so the candidates for determination will be ID_a11 to ID_a14. In step 5, the person whose persona data is most similar to the target of analysis is extracted from the list of candidates as the "judgment result." For example, if ID_a11 is the person whose profile is most similar to ID_b11, then ID_a11 will be the judgment result.

[0033] Steps 1 through 5 require selecting a candidate for judgment from an initial set of individuals of the same type and age as the subject of analysis, and then extracting the judgment result. In other words, the second prompt is to select multiple first individuals a j and several second person b i The generation AI3 is being asked to associate people who are assigned to the same category with each other.

[0034] Steps 1 to 5 correspond to the first process in the present invention. The object to be analyzed in the first process corresponds to the first object to be analyzed in the present invention, the candidate for determination in the first process corresponds to the first candidate for determination in the present invention, and the result of determination in the first process corresponds to the first result of determination in the present invention.

[0035] In step 8, the extracted ID_aj, for example ID_a11, is written to the C file in association with ID_b11, thereby removing it from the list of candidates for judgment. Then, k is incremented by 1, and the process returns to step 1.

[0036] Since k is incremented by 1, for example, ID_b12 becomes the next target of analysis. If ID_b11 and ID_b12 are of the same type and age, the same initial set is used. Since the extracted ID_aj is excluded from the candidates, the candidates become ID_a12 to ID_a14. If ID_a12 is the person whose profile is most similar to ID_b12 among the candidates, then ID_a12 is used as the result, and ID_a12 is written to the C file in association with ID_b12.

[0037] If either the type or age of ID_b11 or ID_b12 differs, a different initial set will be selected from file A. In this way, an initial set is created for each combination of type and age.

[0038] In step 9, if the number of candidates for judgment is two or less, the process proceeds to recursion. For example, if ID_a11 is determined for ID_b11, and ID_a12 is determined for ID_b12, and there are two candidates for judgment for ID_b13, ID_a13 and ID_a14, the recursion process will be explained with reference to Figure 13.

[0039] In process 1, a determination is made regarding the current ID_bi. For example, suppose the current ID_b13 is determined to be similar to ID_a13 among the determination candidates ID_a13 and ID_a14. ID_b13 corresponds to the second analysis target in the present invention, ID_a13 and ID_a14 correspond to the second determination candidates in the present invention, and ID_a13 corresponds to the second determination result in the present invention.

[0040] In process 1, if there is anything unnatural about the second judgment result for the second analysis target, the initial set is removed from the second judgment candidate and re-selected as a new judgment candidate. For example, suppose that for the current ID_b13, ID_a11 and ID_a12 are selected as new judgment candidates, and as a result of the re-selection, it is determined to be similar to ID_a11. ID_a11 and ID_a12 correspond to the third judgment candidate in the present invention, and ID_a11 corresponds to the third judgment result in the present invention.

[0041] In process 1, we consider which of the second or third judgment results is better. If the result of the re-selection is rejected and the second judgment result is adopted, ID_a13 is associated with ID_b13 and written to the C file. If the third determination result, which is the result of the re-selection, is adopted, ID_a11 is associated with ID_b13 and written to the C file. At this time, ID_b11, which was already associated with ID_a11, is in conflict with ID_b13 in terms of association with ID_a11. Therefore, the association between ID_a11 and ID_b11 is deleted from the C file, and the following processes 2 and 3 are performed. Process 1 corresponds to the second process in the present invention.

[0042] In process 2, for the conflicting ID_b11, ID_a12 to ID_a14 are re-selected as new judgment candidates, excluding ID_a11 obtained through re-selection from the initial set, and the C file is updated. ID_b11 corresponds to the third analysis target in the present invention, ID_a12 to ID_a14 correspond to the fourth judgment candidates in the present invention, and the judgment result selected from the fourth judgment candidates corresponds to the fourth judgment result in the present invention. Process 2 corresponds to the third process in the present invention.

[0043] If the fourth judgment result creates a new conflict, repeat process 2. Once the conflict is resolved, there is only one candidate remaining, so proceed to process 4 and perform the same processing as processes 1 to 3.

[0044] Figure 15 shows an example of a C file returned by the generated AI3 in response to the second prompt. The ID_aj of the judgment result is stored in ascending order of the ID_bi of the data to be analyzed. There may be ID_bi or ID_aj that could not be matched.

[0045] <1-7. Effects> (1) According to the first embodiment, the person matching support device 1 is Multiple first person a jEach of the first characteristic data and multiple second person b i The second feature data of each of the following is obtained, and multiple first person a j Each of the first persona data and multiple second person b i A first prompt generation module 11 generates a first prompt that causes AI3 to generate second persona data for each of the following: Based on the first and second persona data, multiple first person a j and several second person b i A second prompt generation module 12 generates a second prompt that requests the generation AI3 to individually associate and Includes.

[0046] According to this, even lists of individuals that previously were difficult to associate due to the lack of unique keys such as email addresses can now be associated via persona data. Two individuals associated via persona data do not necessarily have to be the same person. Two individuals with similar personas, for example, have similar tendencies or similar behavioral patterns, so when predicting human behavior, for example, treating these two individuals as the same person in data analysis can be expected to yield sufficient predictive accuracy. Rather, the first person a j and the second person b i By combining these methods, we can increase the number of variables for analysis, which can lead to improved prediction accuracy.

[0047] The first and second feature data shown in Figures 3 and 4 do not need to have all their fields filled in. Even if some fields have missing data, the AI3 can generate persona data from the data present in the other fields, making it possible to create a correspondence based on the persona data.

[0048] Since the first and second characteristic data may contain confidential information or personally identifiable information, information exchange can be difficult if the first entity possesses the first characteristic data and the second entity possesses the characteristic data. However, with persona data, information can be blurred to the extent that correspondence is possible, which can facilitate information exchange between different entities.

[0049] While analysis using Generative AI3 can sometimes lack clear reasoning and make verification of the results difficult, using persona data allows us to understand the reasoning behind the decisions and enables verification of the analysis results.

[0050] (2) According to the first embodiment, Both the first feature data or first persona data and the second feature data or second persona data include category data. The second prompt is multiple first person a j and several second person b i The generation AI3 is requested to associate people who are assigned the same category with each other.

[0051] According to this, the accuracy of the judgment can be improved by matching people who have been assigned the same category. If there are too many subjects to analyze or candidates for judgment, judgment conflicts are likely to occur, which can increase the load on the generating AI3 or make it difficult to produce a judgment result at all, but this can be improved.

[0052] (3) According to the first embodiment, the second prompt is Multiple second person b i We will use one of the individuals who has not yet been matched as the first subject of analysis, and multiple first individuals a j The first process (steps 1 to 5) is repeated to associate the first determination result, which is one person among the multiple first determination candidates that have first persona data similar to the second persona data of the first analysis target, with the first analysis target. Multiple first person a j If the number of individuals who have not yet been matched falls below a threshold, the second process (process 1) is performed, which involves multiple second individuals b i One of the individuals who has not yet been matched will be the second subject of analysis, and multiple first individuals a jAmong the second group of candidates for determination, who have not yet been matched, the second determination result is one person whose first persona data is similar to the second persona data of the second subject of analysis, and multiple first person a j A second process (process 1) is performed to associate the third person with the second analysis target by selecting one of the third judgment results, which is one of the third judgment candidates that has the first persona data similar to the second persona data of the second analysis target, from among the third judgment candidates that have already been matched. If the third judgment result is adopted, the first analysis subject that was associated with the third judgment result will become the third analysis subject, and multiple first individuals a j Among the fourth set of judgement candidates, excluding those with the third judgment result, a third process (process 2) is performed to associate the fourth judgment result, which is similar to the third set of analysis targets, with the third set of analysis targets. This is requested of the generating AI3.

[0053] According to this, the first process (steps 1 to 5) results in multiple second persons b i Since each person is selected individually for analysis and a corresponding judgment result is produced, it can be expected that a highly accurate judgment will be made for each individual. Second person b i Even for individuals who are selected later in the analysis process, if they are subject to competition in the second process (process 1), they can be re-selected to ensure that the results are properly matched.

[0054] <2. Second Embodiment> <2-1. Functions and Operation> Figure 16 shows the functions and operation of the person matching support device 1a according to the second embodiment. In addition to the various modules included in the person matching support device 1 described with reference to Figure 2, the person matching support device 1a further includes a third prompt generation module 13 and a judgment score calculation module 14. These modules are implemented by loading a program into the memory included in the person matching support device 1a and executing it by the CPU.

[0055] The third prompt generation module 13 generates a third prompt. The third prompt generates multiple first person a based on the first and second persona data. j and several second person b i This is a prompt that requests the generating AI3 to individually associate with and . The second prompt is a prompt that requests multiple second person b i For each of these, multiple first person a j The first prompt requests the AI3 to generate determination data to determine which of the following it corresponds to, while the third prompt requests the AI3 to generate determination data to determine which of the following it corresponds to. j For each of these, multiple second person b i The AI3 generates a second judgment data indicating which of the following it corresponds to. In other words, the direction of the judgment differs between the second and third prompts. The first judgment data will be explained with reference to Figure 17, and the second judgment data will be explained with reference to Figure 18.

[0056] The judgment score calculation module 14, based on the first and second judgment data generated by the generating AI 3, determines multiple first persons a j and several second person b i A comprehensive judgment score J for individually matching and ij Calculate the judgment score J. ij This is calculated by applying Bayes' theorem, as follows:

[0057] Set B={b1,b2,...,b n element b of} i Given that it is a set A = {a1, a2, ..., a n Element a of} j The probability of this being true is given by the confidence matrix conf B→A It can be expressed as: Confidence matrix conf B→A elements (conf B→A ) ij The following equation 1 is given. (conf B→A ) ij =P(a j |b i ) = P(b i |aj )P(a j ) / P(b i ) ···(Equation 1) Here, P(a j |b i ) is the probability that event a occurs after event b (conditional probability, posterior probability), P(b i |a j ) is the likelihood, P(a i ) is the probability of occurrence of a (subjective probability, prior probability), and P(b j ) is the probability of occurrence of b. j ) is the probability of occurrence of a (subjective probability, prior probability), and P(b j ) is the probability of occurrence of b. i ) is the probability of occurrence of b. i is the probability of occurrence of b.

[0058] Conversely, when judging from element a j to element b i , if the strength of confidence is observed and taken as confidence level c ji , then the subjective probability P(a j ) and the likelihood P(b i |a j ) are calculated as follows. P(a j )=(1 / C)Σ i (c ji ), C=Σ ij (c ji ) ···(Equation 2) P(b i |a j )=c ji ···(Equation 3)

[0059] Furthermore, P(b i ) is calculated as follows. P(b i )=Σ j (P(b i |a j )P(a j ))=(1 / C)Σ jk (c ji c jk ) ···(Equation 4)

[0060] By substituting Equations 2 to 4 into Equation 1, the element P(a j |b i ) of the confidence matrix can be obtained. Note that Σy (x yz ) is x yz This is the sum of all y values, Σ yz (x yz ) is x yz This is the sum of all y values ​​and all z values.

[0061] In the second embodiment, instead of simply applying the confidence matrix obtained from Bayes' theorem, the reliability of the decision result is evaluated using the evaluation weight s ij This is expressed as [expression], and this is multiplied by the confidence level to obtain the judgment score J. ij To obtain. J ij =s ij P(a j |b i )

[0062] Judgment score J ij This refers to multiple second persons b i Multiple first person a for each of j This can be interpreted as the degree of probability for each of them.

[0063] <2-2. Evaluation weights s ij , confidence level c ji , and judgment score J ij Example > Figure 17 shows an example of the first judgment data. The first judgment data consists of multiple second person b i For each of these, multiple first person a j This data is obtained by repeatedly requesting the Generator AI3 to determine which of the following it corresponds to. The multiple requests to the Generator AI3 may involve giving the same prompt to the same Generator AI3, giving multiple prompts with different procedures, or giving requests to multiple different Generator AI3s.

[0064] Figure 18 shows the evaluation weights s obtained from the first judgment data. ij An example is shown. By aggregating the first judgment data, the evaluation weight s ij You can obtain this.

[0065] Figure 19 shows an example of the second judgment data. The second judgment data consists of multiple first persons a j For each of these, multiple second person b i This data is obtained by requesting the Generator AI3 to provide the result of determining which of the options it corresponds to, along with its confidence level. Here, we show the case where data including the confidence level is requested from the Generator AI3 only once, but as in Figure 17, it is also possible to make multiple requests and calculate the average of the confidence levels.

[0066] Figure 20 shows the confidence level c obtained from the second judgment data. ji Here is an example. Figure 21 shows the evaluation weights s ij and confidence level c ji Judgment score J calculated from ij Here is an example.

[0067] Evaluation weight s ij This is not limited to cases obtained from the judgment data shown in Figure 17, but also, as in Figure 19, multiple second persons b i For each of these, multiple first person a j It could also be obtained by requesting the generating AI3 to indicate its confidence level regarding which of the following it can be associated with.

[0068] confidence level c ji This is not limited to cases obtained from the judgment data shown in Figure 19, but also, as in Figure 17, multiple first persons a j For each of these, multiple second person b i The result of determining which of the following corresponds to which can be obtained by generating AI3 multiple times and aggregating the data obtained.

[0069] In other respects, the second embodiment is the same as the first embodiment.

[0070] <2-3. Effects> (4) According to the second embodiment, the person matching support device 1a is Based on the first and second persona data, multiple first person a j and several second person b iA third prompt generation module 13 generates a third prompt that requests the generation AI3 to individually associate and Based on the response from Generating AI3, multiple first person a j and several second person b i Judgment score J for determining the correspondence ij A judgment score calculation module 14 that calculates the following, It further includes, The second prompt is multiple second person b i For each of these, multiple first person a j The AI3 is requested to generate first determination data to determine which of the following it corresponds to. The third prompt is multiple first person a j For each of these, multiple second person b i The AI3 is then asked to generate a second set of judgment data to determine which of the following it corresponds to. The judgment score calculation module 14 is: Based on the first judgment data, multiple second person b i For each of these, multiple first person a j Evaluation weights s for each correspondence ij Calculate, Based on the second judgment data, multiple first person a j For each of these, multiple second person b i The confidence level of each correspondence c ji Calculate, The judgment score J is calculated using the following formula. ij Calculate. J ij =s ij P(a j |b i ) however, P(a j |b i ) = P(b i |a j )P(a j ) / P(b i ) P(a j )=(1 / C)Σ i (c ji ) C = Σ ij (c ji ) P(b i |a j )=c ji P(b i )=(1 / C)Σ jk (c ji c jk )

[0071] According to this, the matching judgment score J is obtained by combining multiple judgment data obtained from the generated AI3. ij Since it calculates this, stable results can be obtained even if the output from Generator AI3 changes from time to time. Also, even if the accuracy of the judgment by Generator AI3 is not sufficient, the accuracy of the correspondence can be ensured by combining multiple judgment data. Rather, by requesting judgment results many times from Generator AI3, which is designed to operate at high speed rather than accuracy or versatility, and combining the resulting data, an improvement in accuracy can be expected.

[0072] <3. Third Embodiment> <3-1. Data used for scoring> Figure 22 shows an example of data for scoring by the scoring module 15 (see Figures 2 and 16) in the third embodiment.

[0073] Multiple first person a j For each of these, there may be a first characteristic data and a first persona data. The first characteristic data is, for example, the data shown in Figure 3, and in Figure 22, these are represented by variables V1, V2, etc. The first persona data is, for example, the numerical representation of the data shown in Figure 8, and in Figure 22, these are represented by variables P1, P2, etc.

[0074] Multiple second person b iFor each of these, there may be second characteristic data, second persona data, and correct answer data. The second characteristic data is a numerical representation of the data shown in Figure 4, for example, and is represented by variables V3, V4, etc. in Figure 22. The second persona data is a numerical representation of the data shown in Figure 9, for example, and is represented by variables P3, P4, etc. in Figure 22. The correct answer data is the data that serves as the target variable in scoring, for example, with resignations set to 1 and current employees to 0. The correct answer data may also be data included in the second characteristic data.

[0075] According to the first or second embodiment, a plurality of first persons a j and several second person b i As a result of individually associating these, data for individuals in the following three groups, Group 1 to Group 3, can be generated.

[0076] (Group 1) ID_a1, ID_a2, ID_b1, ID_b2 If ID_a1 and ID_b1 are associated, they can be analyzed as the same person. If ID_a2 and ID_b2 are associated, they can be analyzed as the same person. For each person in Group 1, there may be first and second feature data, first and second persona data, and ground truth data.

[0077] (Group 2) ID_b3, ID_b4 Multiple first person a j ID_b3 and ID_b4 may also exist that do not correspond to any of the above. For each person in Group 2, there may be second feature data, second persona data, and ground truth data. First feature data and first persona data are not available.

[0078] (Group 3) ID_a5, ID_a6 Multiple second person b iID_a5 and ID_a6 may also exist that do not correspond to any of the above. For each person in Group 3, there may be a first characteristic data and a first persona data. The second characteristic data, the second persona data, and the ground truth data have not been obtained.

[0079] <3-2. Scoring Examples> For example, if we want to determine the resignation risk of ID_a5 and ID_a6 (Group 3) as the score for ID_a5 and ID_a6, then the second person b i It is conceivable to perform machine learning using the correct data (retired employees, current employees). ID_a5 and ID_a6 and the second person b i Even if the variables are different, scoring is possible by using the results of the correspondence as follows. ID_a5 and ID_a6 are examples of a third person in this invention.

[0080] Furthermore, if the first and second characteristic data are customer data rather than employee data, and the correct data is set to 1 for purchasers and 0 for non-purchasers, then the purchase probability of a third person can be scored.

[0081] <3-2-1. Example 1> Figure 23 is a flowchart showing scoring example 1 in the third embodiment.

[0082] In S11, the scoring module 15 obtains a first explanatory variable from the first feature data or first persona data of ID_a1 and ID_a2 belonging to group 1, and obtains the target variable from the ground truth data of ID_b1 and ID_b2 belonging to group 1, and creates a model. The first explanatory variable includes variables V1 and V2, or variables P1 and P2.

[0083] In S13, the scoring module 15 obtains second explanatory variables from the first feature data or first persona data of ID_a5 and ID_a6, inputs them into the model, and calculates a score. The second explanatory variables include variables V1 and V2, or variables P1 and P2.

[0084] In this way, scoring becomes possible by using variables V1 and V2, or variables P1 and P2, which are common variables for ID_a1 and ID_a2 and ID_a5 and ID_a6, and by using the ground truth data for ID_b1 and ID_b2, which are associated with ID_a1 and ID_a2 respectively.

[0085] <3-2-2. Example 2> Figure 24 is a flowchart showing example 2 of scoring in the third embodiment.

[0086] In S21, the scoring module 15 obtains a first explanatory variable from the first feature data or first persona data of ID_a1 and ID_a2 belonging to group 1, and the second feature data or second persona data of ID_b1 and ID_b2 belonging to group 1, and obtains the target variable from the ground truth data of ID_b1 and ID_b2 belonging to group 1 to create a model. The first explanatory variable includes variables V1, V2, V3, and V4, or variables P1, P2, P3, and P4.

[0087] In S22, the scoring module 15 selects ID_a1 or ID_a2 from group 1 that have feature data or persona data similar to the first feature data or first persona data of ID_a5, and estimates the second feature data or second persona data of ID_b5 from the second feature data or second persona data of ID_b1 or ID_b2 associated with the selected ID_a1 or ID_a2. The estimation is performed, for example, by weighting the variables V3 and V4 or variables P3 and P4 of ID_b1 or ID_b2 according to the similarity of variables or personas between ID_a5 and ID_a1 or ID_a2. The same estimation is performed for ID_b6. ID_b5 and ID_b6 are hypothetical individuals identified with ID_a5 and ID_a6.

[0088] In S23, the scoring module 15 obtains a second explanatory variable from the first feature data or first persona data of ID_a5 and ID_a6 and the estimated second feature data or second persona data of the virtual ID_b5 and ID_b6, inputs it into the model, and calculates a score. The second explanatory variable includes variables V1 and V2 and estimated variables V3 and V4, or variables P1 and P2 and estimated variables P3 and P4.

[0089] In this way, by estimating the variables V3 and V4, or variables P3 and P4, for the virtual IDs ID_b5 and ID_b6, scoring becomes possible using the variables V1, V2, V3, and V4, or variables P1, P2, P3, and P4.

[0090] The second example described the calculation of scores for ID_a5 and ID_a6, but the present invention is not limited thereto. Similar to S22, first feature data or first persona data for virtual ID_a3 and ID_a4 may be estimated. In this case, similar to S23, a second explanatory variable can be obtained from the estimated first feature data or first persona data for virtual ID_a3 and ID_a4 and the second feature data or second persona data for ID_b3 and ID_b4, and input into the model to calculate scores for ID_b3 and ID_b4.

[0091] If the first feature data or first persona data for the hypothetical IDs_a3 and ID_a4 are estimated, the estimated variables may be used as the first explanatory variables in creating the model. In other words, in addition to using the variables of individuals belonging to Group 1, if the first explanatory variables are obtained from the estimated first feature data or first persona data of the hypothetical IDs_a3 and ID_a4 belonging to Group 2, and from the second feature data or second persona data of IDs_b3 and ID_b4 belonging to Group 2, and the target variable is obtained from the ground truth data of IDs_b3 and ID_b4 belonging to Group 2, and a model is created using these variables as well, a more accurate model can be created.

[0092] <3-2-3. Example 3> Figure 25 is a flowchart showing scoring example 3 in the third embodiment.

[0093] In S31, the scoring module 15 obtains a first explanatory variable from the first feature data or first persona data of ID_a1 and ID_a2 belonging to group 1, and the second feature data or second persona data of ID_b1 and ID_b2 belonging to group 1, and obtains the target variable from the ground truth data of ID_b1 and ID_b2 belonging to group 1 to create a model. This is the same as in S21 of Example 2.

[0094] In S33, the scoring module 15 obtains a second explanatory variable from the first feature data or first persona data of ID_a1 and ID_a2 belonging to group 1, and the second feature data or second persona data of ID_b1 and ID_b2 belonging to group 1, inputs it into the model, and calculates the score.

[0095] In S34, the scoring module 15 selects ID_a1 or ID_a2 from group 1 that have feature data or persona data similar to the first feature data or first persona data of ID_a5, and estimates the score of ID_a5 from the scores of the selected ID_a1 or ID_a2. The estimation is performed, for example, by weighting the scores according to the similarity of the variables between ID_a5 and ID_a1 or ID_a2, or the similarity of the personas. The same estimation is performed for ID_a6.

[0096] By scoring ID_a1 and ID_a2 in this way, it becomes possible to estimate the scores of ID_a5 and ID_a6 based on the similarity of the variables between ID_a1 and ID_a2 and ID_a5 and ID_a6.

[0097] The third example described the calculation of the scores for ID_a5 and ID_a6, but the present invention is not limited thereto. Similar to S34, the scores for ID_b3 and ID_b4 may be estimated based on the similarity of the variables.

[0098] <3-2-4. Others> In Examples 1 to 3, scoring may be performed using both the first or second feature data and the first or second persona data as explanatory variables, or only the first or second feature data may be used, or only the first or second persona data may be used. If only the first or second persona data is used, the degree of anonymization of the information will be high, so the first subject will be the first person a. j The second subject possesses the information of the second person b i Even if they possess the same information, they can encourage each other to share that information and utilize it in their respective businesses. In other respects, the third embodiment is the same as the first or second embodiment.

[0099] <3-3. Effects> (5) According to the third embodiment, the person matching support device 1 or 1a is The system includes a scoring module 15 that creates a model by performing machine learning on a combination of a first explanatory variable and an objective variable to be learned, and calculates a score regarding the target variable by inputting a second explanatory variable to be predicted into the model. Multiple first person a mapped based on the response from Generating AI3 j One of the second person b i Using one person as the same training subject, the first explanatory variables V1_a1, V2_a1, P1_a1, and P2_a1 are obtained from the first feature data or the first persona data, and the target variable is obtained from the second feature data to create a model. The second explanatory variables V1_a5, V2_a5, P1_a5, and P2_a5 are obtained from the first characteristic data or first persona data of the third person, and the score of the prediction target is calculated using the third person as the prediction target.

[0100] According to this, if there is first characteristic data or first persona data for the third person, then the first person a belonging to group 1 jThe first characteristic data or first persona data of person b, and the second person b belonging to group 1. i By using the correct data, we can make predictions about a third person.

[0101] (6) According to the third embodiment, the person matching support device 1 or 1a is The system includes a scoring module 15 that creates a model by performing machine learning on a combination of a first explanatory variable and an objective variable to be learned, and calculates a score regarding the target variable by inputting a second explanatory variable to be predicted into the model. Multiple first person a mapped based on the response from Generating AI3 j One of the second person b i Using one person as the same training subject, the first explanatory variables V1_a1, V2_a1, P1_a1, P2_a1, V3_b1, V4_b1, P3_b1, P4_b1 are obtained from the first and second feature data or the first and second persona data, and the target variable is obtained from the second feature data to create a model. Multiple first persons a who have first characteristic data or first persona data similar to the first characteristic data or first persona data of a third person. j Multiple second person b associated with this i Obtain the second characteristic data or second persona data of a third person from the second characteristic data or second persona data of the third person, The second explanatory variables V1_a5, V2_a5, P1_a5, P2_a5, V3_b5, V4_b5, P3_b5, and P4_b5 of the third person are obtained from the first characteristic data or first persona data of the third person and the second characteristic data or second persona data of the third person, and the score of the prediction target is calculated with the third person as the prediction target.

[0102] According to this, if there is first characteristic data or first persona data for the third person, then the first person a belonging to group 1 j The first characteristic data or first persona data of person b, and the second person b belonging to group 1. iBy using the second characteristic data or second persona data of the first person, the second characteristic data or second persona data of the third person can be calculated, and since the variables for the third person become more diverse, the prediction accuracy can be improved.

[0103] (7) According to the third embodiment, the person matching support device 1 or 1a is The system includes a scoring module 15 that creates a model by performing machine learning on a combination of a first explanatory variable and an objective variable to be learned, and calculates a score regarding the target variable by inputting a second explanatory variable to be predicted into the model. Multiple first person a mapped based on the response from Generating AI3 j One of the second person b i Using one person as the same training subject, the first explanatory variables V1_a1, V2_a1, P1_a1, P2_a1, V3_b1, V4_b1, P3_b1, P4_b1 are obtained from the first and second feature data or the first and second persona data, and the target variable is obtained from the second feature data to create a model. Multiple first person a mapped based on the response from Generating AI3 j One of the second person b i Using one person as the same prediction target, the second explanatory variables V1_a1, V2_a1, P1_a1, P2_a1, V3_b1, V4_b1, P3_b1, and P4_b1 are obtained from the first and second feature data or the first and second persona data, and the score of the prediction target is calculated. The score of the third person is obtained from the score of the predicted target, which has first characteristic data or first persona data similar to the first characteristic data or first persona data of the third person.

[0104] According to this, the first person a belonging to group 1 j Since the score can be calculated with high accuracy from a variety of variables, the first characteristic data or first persona data of the third person and the first person a j A score for the third person can be appropriately calculated based on the similarity to the first characteristic data or first persona data. [Explanation of symbols]

[0105] 1, 1a...Person matching support device, 2...Database, 3...Generating AI, 11...First prompt generation module, 12...Second prompt generation module, 13...Third prompt generation module, 14...Judgment score calculation module, 15...Scoring module

Claims

1. Multiple first person a ｊ Each of the first characteristic data and multiple second person b ｉ The second characteristic data of each of the multiple first persons a is obtained, and the plurality of first persons a ｊ Each of the first persona data and the plurality of second person b ｉ A first prompt generation module that generates a first prompt that causes the first generation AI to generate the second persona data for each of the following, Based on the first and second persona data, the plurality of first individuals a ｊ and the aforementioned multiple second persons b ｉ A second prompt generation module generates a second prompt that requests the second generation AI to individually associate and A person matching support device, including...

2. A person matching support device as described in claim 1, Both the first feature data or the first persona data and the second feature data or the second persona data include category data. The second prompt is the plurality of first persons a ｊ and the aforementioned multiple second persons b ｉ The second generation AI is requested to associate individuals who have been assigned the same category with each other. Person-matching support device.

3. A person matching support device as described in claim 1, The second prompt is, the plurality of second persons b ｉ One of them that has not yet been associated is set as the first analysis target, and among the first determination candidates, which are a plurality of persons among the plurality of first persons a ｊ that have not yet been associated, the first determination result, which is one person having the first persona data similar to the second persona data of the first analysis target, is repeatedly subjected to the first process of associating with the first analysis target The aforementioned plurality of first persons a ｊ If the number of individuals among them who have not yet been matched falls below a threshold, the second process is performed, and the plurality of second individuals b ｉ One of the individuals who has not yet been matched will be the second subject of analysis, and the aforementioned multiple first individuals a ｊ Among the second group of candidates for determination, which are individuals who have not yet been matched, the second determination result is one individual whose first persona data is similar to the second persona data of the second subject of analysis, and the plurality of first individuals a ｊ The second process is performed to associate the third person with the second analysis target by selecting either the third determination result, which is one of the third determination candidates that has already been matched with the second persona data of the second analysis target, and the third determination result which is one of the third candidates that has already been matched with the second persona data of the second analysis target. If the third judgment result is adopted, the first analysis target that was associated with the third judgment result becomes the third analysis target, and the plurality of first persons a ｊ The third process involves associating the fourth determination result, which is the person remaining after excluding the third determination result, with the third determination result, which is similar to the third analysis target, with the third analysis target. A person matching support device that requests the second generation AI to do the above.

4. A person matching support device as described in claim 1, Based on the first and second persona data, the plurality of first individuals a ｊ and the aforementioned multiple second persons b ｉ A third prompt generation module generates a third prompt that requests a third generation AI to individually associate and Based on the responses from the second generation AI and the third generation AI, the plurality of first people a ｊ and the aforementioned multiple second persons b ｉ Judgment score J for determining the correspondence ｉｊ A judgment score calculation module that calculates the following, It further includes, The second prompt is the plurality of second persons b ｉ For each of the above, the plurality of first persons a ｊ The first determination data, which of the following it corresponds to, is requested from the second generating AI. The third prompt is the plurality of first persons a ｊ For each of the above, the multiple second person b ｉ The third generating AI is requested to provide a second determination data indicating which of the following it corresponds to. The aforementioned judgment score calculation module is: Based on the first determination data, the plurality of second persons b ｉ For each of the above, the plurality of first persons a ｊ Evaluation weights s for each correspondence ｉｊ Calculate, Based on the second determination data, the plurality of first persons a ｊ For each of the above, the multiple second person b ｉ Confidence level c for matching each of them ｊｉ Calculate, The judgment score J is calculated using the following formula. ｉｊ Calculate J ｉｊ =s ｉｊ P(a ｊ |b ｉ ) however, P(a ｊ |b ｉ )=P(b ｉ |a ｊ )P(a ｊ ) / P(b ｉ ) 6(D) ｊ (11)Σ ｉ (s) ｊｉ ) C=S ｉｊ (c) ｊｉ ) P(b ｉ |a ｊ )=c ｊｉ 6(D) ｉ (11)Σ ｊｋ (s) ｊｉ セ ｊｋ ) Person-matching support device.

5. A person matching support device as described in claim 1, The system further includes a scoring module that performs machine learning to create a model from a combination of a first explanatory variable and an objective variable to be learned, and calculates a score for the target variable by inputting a second explanatory variable to be predicted into the model, wherein the scoring module The plurality of first persons a that are associated based on the response from the second generation AI ｊ One of the aforementioned second persons b ｉ Using one of the individuals as the same learning target, the first explanatory variable is obtained from the first feature data or the first persona data, and the target variable is obtained from the second feature data to create the model. The second explanatory variable is obtained from the first characteristic data or first persona data of the third person, and the score of the prediction target is calculated using the third person as the prediction target. Person-matching support device.

6. A person matching support device as described in claim 1, The system further includes a scoring module that performs machine learning to create a model from a combination of a first explanatory variable and an objective variable to be learned, and calculates a score for the target variable by inputting a second explanatory variable to be predicted into the model, wherein the scoring module The plurality of first persons a that are associated based on the response from the second generation AI ｊ One of the aforementioned second persons b ｉ Using one of the individuals as the same learning target, the first explanatory variable is obtained from the first and second feature data or the first and second persona data, and the target variable is obtained from the second feature data to create the model. The plurality of first persons a that have first characteristic data or first persona data similar to the first characteristic data or first persona data of a third person ｊ The aforementioned multiple second persons b that are associated with ｉ Obtain the second characteristic data or second persona data of the third person from the second characteristic data or second persona data of the third person, Obtain the second explanatory variable of the third person from the first characteristic data or first persona data of the third person and the second characteristic data or second persona data of the third person, and calculate the score of the prediction target, with the third person as the prediction target. Person-matching support device.

7. A person matching support device as described in claim 1, The system further includes a scoring module that performs machine learning to create a model from a combination of a first explanatory variable and an objective variable to be learned, and calculates a score for the target variable by inputting a second explanatory variable to be predicted into the model, wherein the scoring module The plurality of first persons a that are associated based on the response from the second generation AI ｊ One of the aforementioned second persons b ｉ Using one of the individuals as the same learning target, the first explanatory variable is obtained from the first and second feature data or the first and second persona data, and the target variable is obtained from the second feature data to create the model. The plurality of first persons a that are associated based on the response from the second generation AI ｊ One of the aforementioned second persons b ｉ Using one of the individuals as the same prediction target, the second explanatory variable is obtained from the first and second feature data or the first and second persona data, and the score of the prediction target is calculated. Obtain the score of the third person from the score of the prediction target which has the first characteristic data or first persona data similar to the first characteristic data or first persona data of the third person. Person-matching support device.

Images