Subject response estimation system

Abstract

To provide a subject response estimation system that can realize a natural state in which a questioner exchanges questions with a real survey subject. [Solution] The machine learning device 10 of the survey subject response estimation system 1 is configured to include the functions of a learning model memory unit 11 that stores and retains the digital clone Ci of each survey subject Pi, a question input unit 13 that inputs question data indicating the content of a question given by a user Q to a required digital clone Cx (= C1, or C2, or, ..., or CN) among the digital clones C1 to CN, and a response output unit 14 that outputs the answer obtained from the digital clone Cx in accordance with the question data.

Description

[Technical Field]

[0001] The present invention relates to a survey recipient response estimation system that estimates and outputs responses from survey recipients. [Background technology]

[0002] In surveys in various fields, such as marketing surveys and television audience rating surveys, it has been common practice to collect answers to questions from a number of pre-selected survey subjects (so-called panelists or panelists), as shown in Patent Document 1 below. In order to reduce the burden on the survey subjects while ensuring the same quality and reliability as when collecting answers from the survey subjects, the applicant and inventor of the present application have developed the patent invention shown in Patent Document 2 below.

[0003] The survey aggregation system of Patent Document 2 below compares the answers of multiple survey subjects to questions posed to each of the multiple survey subjects with the answers generated by each of the multiple learning models to the questions, and from each combination of the multiple survey subjects and each of the multiple learning models, extracts pair sets, which are pairs of survey subjects and learning models whose answers to the questions match at a level higher than a predetermined threshold, and is configured to obtain the answers generated by each learning model of the pair sets to the questions as substitutes for the answers of the survey subjects. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2002-312557 [Patent Document 2] Patent No. 7290704 Summary of the Invention [Problem to be solved by the invention]

[0005] The inventors of the present application have now developed such a survey compilation system and have come to the realization that various survey data can be obtained only in a natural environment, as if the questioner were exchanging questions with the actual survey subjects.

[0006] In view of the above circumstances, an object of the present invention is to provide a subject response estimation system that allows a questioner to have a natural exchange of questions, as if he or she were actually conducting a survey with the survey subject. [Means for solving the problem]

[0007] The survey subject response estimation system of the first invention is a survey subject response estimation system that estimates and outputs responses from survey subjects, a machine learning device that constructs and stores learning models of the plurality of survey subjects by machine learning processing, The machine learning device includes: Multiple survey subjects Life Log a learning model storage unit that stores a plurality of learning models constructed so as to be able to estimate answers given by each survey subject to questions by learning the above-mentioned model as digital clones of the plurality of survey subjects; a question input unit that acquires questions for the plurality of survey subjects and inputs question data indicating the contents of the questions into the plurality of learning models; an answer output unit that outputs answers estimated by the plurality of learning models as answers of the plurality of survey subjects in response to question data input from the question input unit; It is equipped with The answer output unit is characterized in that it extracts features of the question content from the question data and outputs a question corresponding to the features of the question content. According to this invention, by extracting the characteristics of the question asked by the questioner and having the digital clone ask the question corresponding to the extracted characteristics to the questioner in return, it is possible to realize a natural conversation as if the questioner and the original survey subject were exchanging questions with each other. The survey subject response estimation system of the second invention is a survey subject response estimation system that estimates and outputs responses from survey subjects, a machine learning device that constructs and stores learning models of the plurality of survey subjects by machine learning processing, The machine learning device includes: Multiple survey subjects Life Log a learning model storage unit that stores a plurality of learning models constructed so as to be able to estimate answers given by each survey subject to questions by learning the above-mentioned model as digital clones of the plurality of survey subjects; a question input unit that acquires questions for the plurality of survey subjects and inputs question data indicating the contents of the questions into the plurality of learning models; an answer output unit that outputs answers estimated by the plurality of learning models as answers of the plurality of survey subjects in response to question data input from the question input unit; It is equipped with The answer output unit outputs Estimate The system is characterized in that it extracts the characteristics of the content of the answer from the answer and outputs a question corresponding to the characteristics of the content. According to this invention, the characteristics of the content of the answer are extracted from the output predicted answer, and the extracted By having the digital clone ask the questioner questions that correspond to their characteristics, the questioner and the original It is possible to realize a natural conversation between the survey subjects as if they were asking each other questions. can.

[0017] No. 3 The survey subject response estimation system of the invention is 1 or 2 In the invention, The answer from the questioner to the question output from the answer output unit is used as the question data, and the conversation is repeated.

[0018] According to this invention, by using the answer from the questioner to the question output from the answer output unit as question data, the answer from the questioner can be used as new question data to repeatedly continue the conversation, thereby realizing a more natural situation in which the questioner exchanges questions with the actual survey subject.

[0019] No. 4 The survey subject response estimation system of the invention is 3 In the invention, The machine learning device is characterized in that it stores a learning model constructed to be able to estimate the answer that the questioner will give to the question by learning the answer from the questioner to the question output by the answer output unit in the learning model memory unit as a digital clone of the questioner.

[0020] According to this invention, a natural state is realized in which the questioner exchanges questions with the actual survey subject, making it possible to construct a learning model that learns the questions and answers posed by the questioner as various survey data, and to create a digital clone of the questioner.

[0021] No. 5 The survey subject response estimation system of the invention is or second In the invention, The answer output unit is characterized by including an attribute estimation unit that extracts features of question content from the question data and estimates attributes of the questioner from the features of the question content.

[0022] According to this invention, a natural situation is realized in which the questioner is exchanging questions with the actual survey subject, and the attributes of the questioner (true attributes that the questioner himself is not aware of) can be inferred from the characteristics of the content of the question as various survey data.

[0023] No. 6 The survey subject response estimation system of the invention is 1 or 2 In the invention, The system is characterized by comprising an attribute estimation unit that estimates the attribute of the questioner from the characteristics of the answer content from the questioner to the question output by the answer output unit.

[0024] According to this invention, a natural situation is realized in which the questioner is exchanging questions with the actual survey subject, and the attributes of the questioner (true attributes that the questioner himself is not aware of) can be inferred from the characteristics of the content of the questioner's answers as various survey data.

[0025] No. 7 The survey subject response estimation system of the invention is or second In the invention, The learning model storage unit is characterized by changing the constructed learning models over time.

[0026] According to this invention, by changing the constructed multiple learning models over time, the digital clone can be adapted to match the passage of time, allowing the questioner to feel as if they are exchanging questions with the original survey subject, including the passage of time. The survey respondent response estimation system of the eighth invention is the first or second invention, The answer output unit is characterized in that it outputs multiple estimated answers from the multiple learning models to the same question data via digital clones that represent the personas of the multiple survey subjects, whose answers were learned to construct the multiple learning models, as human images. According to this invention, the answer output unit simultaneously outputs multiple estimated answers from multiple learning models for the same question data, so that the questioner can naturally realize a situation in which he or she asks the same question to multiple survey subjects and receives questions from them simultaneously (without selecting any special settings). In such a situation, multiple estimated answers from multiple learning models to the same question data are output via digital clones that represent the personas of multiple survey subjects, whose answers were learned to build multiple learning models, as human images. In this way, answers are output from a digital clone of a person's image that reflects the persona of the original survey subject that corresponds to the learning model, making it possible to achieve a more natural exchange of questions, as if the questioner were actually asking the original survey subject. A survey subject response estimation system according to a ninth aspect of the present invention is the eighth aspect of the present invention, The answer output unit is characterized in that, when the question data is not present, it outputs information about the survey subject including a self-introduction associated with the persona. According to this invention, if there is no question data from the questioner, there is a possibility that the conversation will not start. However, if there is no question data, the digital clone can output information about the survey subject, including a self-introduction associated with the persona, allowing the questioner to understand the survey subject (or more accurately, the digital clone corresponding to the survey subject) in a natural way, prompting them to ask questions and starting the conversation. [Brief explanation of the drawings]

[0027] [Figure 1] 1 is a block diagram showing the configuration of a survey recipient response estimation system according to an embodiment of the present invention; [Figure 2] FIG. 2 is an explanatory diagram regarding the processing of the pair set extraction unit shown in FIG. [Figure 3] FIG. 2 is a block diagram showing the configuration of the machine learning device shown in FIG. 1. [Figure 4]FIG. 2 is an explanatory diagram showing the processing of the machine learning device of FIG. 1. [Figure 5] FIG. 2 is an explanatory diagram showing the processing of the machine learning device of FIG. 1. DETAILED DESCRIPTION OF THE INVENTION

[0028] An embodiment of the present invention will be described below with reference to FIGS.

[0029] 1, a survey recipient answer estimation system 1 of this embodiment includes a machine learning device 10 that constructs (generates) and stores multiple learning models LM1, LM2, ..., LMN that can generate answers to questions through machine learning processing; a response aggregation unit 20 that acquires and aggregates answers to questions from all or some of multiple pre-selected survey recipients P1, P2, ..., Pn; a question data generation unit 30 that generates question data indicating questions to all or some of the survey recipients P1 to Pn; a question recipient database 40 (hereinafter simply referred to as database 40) that stores and stores information about each of the survey recipients P1 to Pn; and a communication device 50 that can communicate with an external network NW (wide area network) composed of the Internet, a telephone communication network, etc. Hereinafter, any one of the multiple learning models LM1 to LMN will be referred to as LMi, and any one of the multiple survey recipients P1 to Pn will be referred to as Pj.

[0030] The communication device 50 is capable of communicating with the external network NW, and is also capable of communicating with the communication terminals 60 used by the survey subjects P1 to Pn via the external network NW.

[0031] Here, the communication terminal 60 used by each survey subject Pj is configured, for example, by a smartphone, tablet terminal, personal computer, etc., and includes a display unit 60a configured by a liquid crystal display, etc., and a sound output unit 60b configured by a speaker, etc. A predetermined application for the survey (hereinafter referred to as the survey app) is pre-installed in this communication terminal 60.

[0032] Then, with the survey app running, the communication terminal 60 can communicate with the communication device 50 via the external network NW. In this case, the communication terminal 60 has a function of receiving question data transmitted from the survey recipient answer estimation system 1 via the communication device 50 and notifying the survey recipient Pj of the content of the question indicated by the question data, a function of accepting the survey recipient Pj's answer to the question and transmitting it to the communication device 50, and the like.

[0033] The machine learning device 10, response aggregation unit 20, question data generation unit 30, and database 40 of the survey recipient response estimation system 1 are configured by one or more computers including, for example, a processor such as a microcomputer, a storage device such as a memory, an interface circuit, etc.

[0034] For example, the machine learning device 10, the answer counting unit 20, the question data generation unit 30, and the database 40 may each be configured as separate computers that can communicate with each other. In this case, the computer that constitutes the machine learning device 10 is configured to function as the machine learning device 10 by the hardware configuration and programs (software configuration) implemented therein. The same applies to the answer counting unit 20, the question data generation unit 30, and the database 40.

[0035] However, the computer may be configured so that two or more of the machine learning device 10, the answer compilation unit 20, the question data generation unit 30, and the database 40 are included in, for example, a single computer. Furthermore, the communication device 50 may be included in or attached to any of the computers that make up the machine learning device 10, the answer compilation unit 20, the question data generation unit 30, and the database 40. Alternatively, the communication device 50 may be included in or attached to each of the multiple computers that make up the machine learning device 10, the answer compilation unit 20, the question data generation unit 30, and the database 40.

[0036] The database 40 stores and holds information about each of a plurality of pre-selected survey subjects P1 to Pn, such as attribute information about each survey subject Pj, such as place of residence, gender, age, and family structure, as well as answer history information showing the history of answers to questions posed to each survey subject Pj. The database 40 is capable of outputting any information (attribute information, answer history information, etc.) about any survey subject Pj in response to a request from the machine learning device 10, the answer counting unit 20, etc.

[0037] The question data generation unit 30 is configured to generate question data indicating the contents of multiple questions for a required survey, such as a marketing survey or a television audience rating survey, in response to instructions from an operator. The question data can be generated as text data or audio signal data, for example. The question data generation unit 30 can store and retain a history of the question data generated for each survey conducted.

[0038] The machine learning device 10 is configured to construct (generate) multiple learning models LM1 to LMN that can generate answers to questions through machine learning processing that uses the life logs of an unspecified number of people (an unspecified number of individuals) as learning data.

[0039] The life log is information that contains a lot of unique or characteristic information about an unspecified number of individuals. The life log may include SNS (Social Networking Service) information posted by individuals.

[0040] Such a life log can be input by an operator at any time via an appropriate input device to the machine learning device 10. Alternatively, the machine learning device 10 can automatically collect the life log from the Internet or the like of an external network NW.

[0041] The machine learning device 10 extracts life logs to be used as learning data for each learning model LMi from a large number of given life logs, and constructs each learning model LMi by performing machine learning processing for each learning model LMi using the extracted life logs as learning data.

[0042] In this case, in extracting the life log for the learning data of each learning model LMi, for example, the life log for the learning data of each learning model LMi may be extracted so that at least a part of the life logs used as learning data by each learning model LMi is different from each other among the learning models LM1 to LMN. As a result, each of the learning models LM1 to LMN may be configured to have different characteristics from each other regarding the generation of answers to questions.

[0043] Additionally, when the life logs of the survey subjects P1 to PN can be collected, for example, the life log of one of the survey subjects Pi may be used as the learning data of at least some of the learning models LM1 to LMN. In this case, when there are multiple learning models that use the life log of the survey subject Pi as learning data, the life logs of different survey subjects are used as the learning data of each of the multiple learning models.

[0044] The input of new life logs into each learning model LMi and the machine learning process are continuously performed, so that each learning model LMi is updated so that it can generate answers that are consistent with the content of various questions.

[0045] The machine learning device 10 is configured to construct multiple learning models LM1 to LMN through machine learning processing as described above. Note that, as a more specific technique for generating each learning model LMi, the technique proposed in, for example, Japanese Patent Application Laid-Open No. 2018-190457 can be adopted.

[0046] The machine learning device 10 of this embodiment is further configured to include the function of a learning model memory unit 11 that stores and retains the constructed learning models LM1 to LMN, and the function of a pair set extraction unit 12 that extracts pair sets, which are combinations of survey subjects P1 to Pn and learning models whose answers to questions have a degree of agreement higher than a predetermined threshold, from combinations of each of the survey subjects P1 to Pn and each of the learning models LM1 to LMN.

[0047] Here, a specific process of the pair set extraction unit 12 will be described. The pair set extraction unit 12 executes the pair set extraction process, for example, periodically, at a timing instructed by an operator, or after each survey is conducted. In this process, the pair set extraction unit 12 acquires question data from a survey conducted in the past (for example, the most recent survey or the most recent multiple surveys) on all or some of the survey subjects P1 to Pn from the question data generation unit 30, and inputs multiple questions (text questions) indicated by the question data into the learning models LM1 to LMN.

[0048] Furthermore, the pair set extraction unit 12 acquires the answers generated by each learning model LMi in response to each of the multiple questions, and acquires the answers obtained from all or some of the survey subjects P1 to PN in response to the multiple questions from the database 40, and compares the answers obtained from each of the learning models LM1 to LMN with the answers obtained from all or some of the survey subjects P1 to PN.

[0049] Then, the pair set extraction unit 12 extracts, from the combinations of each of the multiple survey subjects P1 to PN and each of the multiple learning models LM1 to LMN, pairs of survey subjects and learning models whose answers to questions have a degree of agreement higher than a predetermined threshold as pair sets.

[0050] In this case, the index value representing the degree of agreement between the answers of each learning model LMi and the answers of each survey subject Pj may be, for example, the proportion of the total number of answers that match between the learning model LMi and the survey subject Pj (proportion to the total number of questions; hereinafter, referred to as the matching answer proportion).The pair set extraction unit 12 then extracts the pair of the survey subject Pj and the learning model LMi as a pair set (Pj, LMi) when the matching answer proportion for a pair of a survey subject Pj among the survey subjects P1 to PN and a learning model LMi among the learning models LM1 to LMN is equal to or greater than a predetermined threshold (for example, when the matching answer proportion is equal to or greater than a threshold close to 100%).

[0051] For example, referring to Figure 2, if the matching response rate for the pair of survey subject Pa and learning model LMa, the matching response rate for the pair of survey subject Pb and learning model LMb, and the matching response rate for the pair of survey subject Pc and learning model LMc are each above a predetermined threshold, these three pairs (Pa, LMa), (Pb, LMb), and (Pc, LMc) are extracted as pair sets.

[0052] Then, for a learning model LMi that forms a pair set with any of the learning models LM1 to LMN and a survey subject Pj, the learning model storage unit 11 stores and holds the learning model LMi in association with the survey subject Pj that forms the pair set. For example, Fig. 1 shows that the learning model LM2 is stored and held in the learning model storage unit 11 in association with the survey subject P5 that forms the pair set with the learning model LM2.

[0053] Additionally, the learning model LMi that forms a pair set with one of the survey subjects Pj has a high degree of agreement with the survey subject Pj in its answers to questions, and therefore can be considered to be equivalent to an artificial intelligence that mimics the thinking and behavior of the survey subject Pj (in other words, a digital clone of the survey subject Pj).

[0054] The response counting unit 20, details of which will be described later, is configured so that when conducting a required survey, it selects multiple survey subjects P1 to Pn registered in the database 40 as response acquisition subjects from whom answers to questions should be obtained, and acquires and counts answers to multiple questions from the multiple response acquisition subjects.

[0055] In addition, if there are a certain number or more answer acquisition subjects who form pair sets with the selected multiple answer acquisition subjects, the answer aggregation unit 20 can also acquire and aggregate answers to multiple questions only from the learning models corresponding to the answer acquisition subjects who form the pair sets.

[0056] Next, the details and operation of the machine learning device 10 will be described.

[0057] As shown in FIG. 3, the machine learning device 10 is configured to be able to construct (generate) digital clones Ci as learning models LM1 to LMN constructed for each of the survey subjects P1 to Pn.

[0058] The machine learning device 10 is also configured to include the functions of a learning model memory unit 11 that stores and retains the digital clone Ci of each survey subject Pi, a question input unit 13 that inputs question data indicating the content of a question given by a user Q (corresponding to the questioner of the present invention) via a communication terminal not shown in the figure into a required digital clone Cx (= C1, or C2, or, ..., or CN) among the digital clones C1 to CN, and an answer output unit 14 that outputs the answer obtained from the digital clone Cx in accordance with the question data.

[0059] Next, the operation of the survey recipient response estimation system 1 when conducting a required survey will be specifically described.

[0060] As shown in Figure 4, user Q first inputs the survey subject conditions. Specifically, user Q inputs the group name, then selects gender, age, place of residence, and occupation, and finally inputs the text "studying using a learning app" in the attribute detail setting field. This selects a pair set of survey subjects and learning models that meet the survey subject conditions from the attribute information and other information about the survey subjects stored in database 40. As a result, in this example, a group with attributes such as "students & learning app users" is set.

[0061] In a typical qualitative survey, it can take several weeks just to recruit the survey subjects to be interviewed, but with the survey subject response estimation system 1 of the present invention, the selection of survey subjects can be completed in just a few tens of seconds.

[0062] Here, the attribute detail settings have no input restrictions and allow you to enter text freely, so the statement "studying using a learning app" can be expressed in bullet points, for example, as "learning app user, likes to study, has been using the app for over a year," or in a sentence, such as "someone who uses a learning app on the train when commuting to school and finds studying with the app enjoyable."

[0063] Next, once the selection of survey subjects has been completed (all AI consultants have been gathered), user Q uses the chat function to ask a question, which is input to the machine learning device 10 via question input unit 13 (more precisely, input to the selected learning model), and the estimated answer of the survey subject is output as the output result (the response of the selected learning model) via answer output unit 14. As a result, the answers to the questions are output to user Q all at once and instantly.

[0064] Furthermore, since the learning model corresponding to each answer acquisition target with a pair set is a learning model that can simulate the thinking and behavior of the answer acquisition target, the answers generated by the learning model in response to each of the multiple questions are likely to match the answers of the answer acquisition target. Therefore, the answers of each answer acquisition target with a pair set can be pseudo-obtained from the learning model corresponding to the answer acquisition target.

[0065] In Figure 5, the survey participants immediately and simultaneously responded to the question, "How often do you read the newspaper?"

[0066] Another question I asked was, "What is your goal in studying?" Ayaka Yamada (18 years old, female): "I am studying with the goal of passing the university entrance exam." Takada Takashi (19 years old, male): "I am studying hard in psychology, my favorite subject, with the aim of going on to university." Emi Toda (13 years old, female): "I study with the goal of improving my grades at school and becoming better at speaking English." Ryoichi Nakamura (19 years old, male): "I'm aiming to take the university entrance exam, so I'm studying with the goal of acquiring the necessary knowledge to pass." As you can see, each person surveyed gives a different answer.

[0067] In this way, the answer output unit 14 outputs multiple estimated answers from multiple learning models for the same question data at once, so that user Q can naturally experience a situation in which he or she asks the same question to multiple survey subjects and receives questions from them simultaneously (without selecting any special settings).

[0068] The digital clones corresponding to the survey subjects were each given a name and a photograph, and each persona of the survey subjects P1 to Pn was expressed as a human image. This allows the interviewer to feel as if they are having a real conversation, even though they are interacting with an AI.

[0069] In this way, according to the survey subject response estimation system 1 of this embodiment, answers are output from a digital clone of a person image that reflects the persona of the original survey subject corresponding to the learning model, making it possible to achieve a more natural appearance as if user Q were exchanging questions with the original survey subject.

[0070] It is preferable that the constructed learning model corresponding to the survey subject change over time in accordance with the passage of time since its construction. In other words, by changing the constructed multiple learning models over time, the digital clone can be made to match the passage of time, allowing the questioner to experience a more natural exchange of questions, including the passage of time, as if they were actually exchanging questions with the original survey subject. For example, consider the case where one year has passed since the learning model was constructed. Pi, a survey subject of learning model Li, is considered to have aged +1 year. Here, from an overall analysis of survey subjects P1 to Pn, changes in attributes such as hobbies and preferences that accompany aging are analyzed and extracted. For example, suppose a trend is obtained that survey subjects who are age A + 1 year older than a certain age A, showing a greater awareness and interest in health. Then, by incorporating (relearning) the changes in the characteristics and trends of Pi's attributes or similar attributes at age + 1 year older than Pi into Pi's learning model Li as changes over one year, the effect of the passage of time can be applied.

[0071] Furthermore, if user Q wants to have a more detailed conversation with a specific survey subject, such as a depth interview in which the moderator and the survey subject have a one-on-one in-depth discussion about a specific topic, he or she can use the "individual chat" function to ask questions only to that specific survey subject.

[0072] For example, in the example above, the AI ​​consultant, Mr. Yamada, answered, "I'm using a learning app to get into university," so Q. What subjects do you use apps to study? A. Mainly math and English Q. Why Math and English? A. Because I think it is important for entrance exams and future advancement In this way, just like in a group interview with real people, we can delve deeper into Yamada's individual answers and explore the deeper levels of his consciousness, asking "why does he think that way?"

[0073] If there is no question data, the answer output unit 14 outputs, as an answer, information about the survey subject, including a self-introduction associated with the persona of the survey subject, by referring to the database 40. Of course, if there is a self-introduction request as question data, the answer output unit 14 outputs, as an answer, similar information about the survey subject by referring to the database 40.

[0074] This means that if there is no question data from user Q, there is a possibility that the conversation will not start.However, if there is no question data, the digital clone can output information about the survey subject, including a self-introduction associated with the persona, allowing user Q to understand the survey subject (or more accurately, the digital clone associated with the survey subject) in a natural way, encouraging them to ask questions and starting the conversation.

[0075] In addition, when there is no question data, it is the case that the question data from user Q is not input, does not start, or is not completed within the given period of time.

[0076] Furthermore, the answer output unit 14 may extract features of the question content from the question data of the question of the user Q, and output a question corresponding to the features of the question content.

[0077] Specifically, by employing various existing technologies related to language analysis and natural language processing, including text mining, it is possible to extract characteristics of the questions posed by user Q, and then have the digital clone ask user Q questions corresponding to the extracted characteristics, thereby realizing a natural conversation as if user Q and the original survey subjects were exchanging questions with each other.

[0078] At this time, the question genre is selected from the ones that are of high interest according to the characteristics of the question content by user Q, and may also be selected from the ones that are of high interest according to the characteristics when the characteristics of the question content by user Q have a tendency, characteristic, or bias (when the information entropy is small) and when there is no tendency, characteristic, or bias (when the information entropy is large).

[0079] Furthermore, the answer output unit 14 may extract features of the content of the output predicted answer from the answer, and output a question corresponding to the features of the content.

[0080] Similarly, by adopting various existing technologies related to language analysis and natural language processing, it is possible to extract characteristics of the content of the output predicted answer and have the digital clone ask user Q questions corresponding to the extracted characteristics, thereby realizing a natural conversation as if user Q and the original survey subject were exchanging questions with each other.

[0081] In this way, when a question is output from the answer output unit 14, the conversation is repeated by using the answer from the user Q to the question as new question data. This makes it possible to encourage questions from the user Q.

[0082] In other words, by inputting the answer from user Q to the question output from the answer output unit 14 as question data via the question input unit 13, the answer from user Q can be used as new question data to repeat and continue the conversation, thereby realizing a more natural situation as if user Q were exchanging questions with the original survey subject.

[0083] Here, when a question is output by the answer output unit 14, the machine learning device 10 may construct a learning model of user Q by learning the answer from user Q. In other words, the learning model constructed so as to be able to estimate the answer that user Q will give to the question may be stored in the learning model storage unit as a digital clone of user Q.

[0084] This allows user Q to appear more natural, as if he were exchanging questions with the original survey subject, making it possible to build a learning model that learns the questions and answers posed by user Q as various survey data, and to create a digital clone of user Q.

[0085] Additionally, the answer output unit 14 may include an attribute estimation unit (not shown) that extracts features of the question content from the question data of the user Q and estimates the attribute of the questioner from the features of the question content.

[0086] As mentioned above, by realizing a more natural situation in which user Q is exchanging questions with the actual survey subjects, it is possible to infer the attributes of the questioner (true attributes that the questioner himself is not aware of) from the characteristics of the questions asked by user Q as various survey data.

[0087] The user attributes (persona) estimated here include demographic attributes such as gender and age group, as well as psychological attributes such as interests.

[0088] For example, the attribute estimation unit is configured with a machine learning model that has been trained by machine learning on feature data and correct answer data based on survey results such as a questionnaire.

[0089] The above is an explanation of the details and operation of the survey subject response estimation system 1 and its machine learning device 10 of this embodiment. The survey subject response estimation system 1 and its machine learning device 10 can achieve a natural appearance, as if the questioner were exchanging questions with the actual survey subject.

[0090] In this embodiment, the questioner Q is assumed to be a person, but the questioner Q may be a digital clone. In this case, since a copy of the digital clone can be created, for example, a digital clone formed from a large-scale language model (learning model) can be copied to a slightly smaller large-scale language model (with a smaller amount of parameters) by repeating the question conversation, thereby creating a copy of the digital clone with the smaller language model. [Explanation of symbols]

[0091] 1...Survey subject response estimation system 10...Machine learning device, 11...Learning model memory unit, 12...Pair set extraction unit, 13...Question input unit, 14...Response output unit, 20...Response aggregation unit, 30...Question data generation unit, 40...Question subject database, 50...Communication device, 60...Communication terminal, P...Survey subject, Q...User (questioner), LM...Learning model, C...Digital clone

Claims

1. A survey subject response estimation system that estimates and outputs responses from survey subjects, a machine learning device that constructs and stores learning models of the plurality of survey subjects by machine learning processing, The machine learning device includes: a learning model storage unit that stores, as digital clones of the plurality of survey subjects, a plurality of learning models constructed so as to be able to estimate answers given by each of the plurality of survey subjects to questions by learning the life logs of the plurality of survey subjects; a question input unit that acquires questions for the plurality of survey subjects and inputs question data indicating the contents of the questions into the plurality of learning models; an answer output unit that outputs answers estimated by the plurality of learning models as answers of the plurality of survey subjects in response to question data input from the question input unit; It is equipped with The survey subject response estimation system is characterized in that the response output unit extracts features of the question content from the question data and outputs a question corresponding to the features of the question content.

2. A survey subject response estimation system that estimates and outputs responses from survey subjects, a machine learning device that constructs and stores learning models of the plurality of survey subjects by machine learning processing, The machine learning device includes: a learning model storage unit that stores, as digital clones of the plurality of survey subjects, a plurality of learning models constructed so as to be able to estimate answers given by each of the plurality of survey subjects to questions by learning the life logs of the plurality of survey subjects; a question input unit that acquires questions for the plurality of survey subjects and inputs question data indicating the contents of the questions into the plurality of learning models; an answer output unit that outputs answers estimated by the plurality of learning models as answers of the plurality of survey subjects in response to question data input from the question input unit; It is equipped with The survey subject response estimation system is characterized in that the response output unit extracts characteristics of the content of the response from the output estimated response, and outputs questions corresponding to the characteristics of the content.

3. 3. The survey subject response estimation system according to claim 1, A survey subject response estimation system, characterized in that a conversation is repeated by using the response from the questioner to the question outputted from the response output unit as the question data.

4. 4. The survey subject response estimation system according to claim 3, The machine learning device learns the answers from the questioner to the questions output by the answer output unit, and stores the learning model constructed to be able to estimate the answer that the questioner will give to the question in the learning model storage unit as a digital clone of the questioner.

5. 3. The survey subject response estimation system according to claim 1, the answer output unit is further provided with an attribute estimation unit that extracts features of the question content from the question data and estimates the attributes of the questioner from the features of the question content.

6. 3. The survey subject response estimation system according to claim 1, A survey subject response estimation system comprising an attribute estimation unit that estimates attributes of a questioner from features of the response content of the questioner's response to the question output by the response output unit.

7. 3. The survey subject response estimation system according to claim 1, A survey respondent response estimation system, characterized in that the learning model storage unit changes the constructed learning models over time.

8. 3. The survey subject response estimation system according to claim 1, a survey subject response estimation system, wherein the response output unit outputs a plurality of estimated responses of the plurality of learning models to the same question data via digital clones that represent, as human images, the personas of the plurality of survey subjects whose answers have been learned to construct the plurality of learning models.

9. 9. The survey subject response estimation system according to claim 8, A survey subject response estimation system, characterized in that the response output unit outputs information about the survey subject including a self-introduction associated with the persona when the question data is not available.

Images