Survey management server, survey analysis method, and program

Abstract

A questionnaire management server is provided that can easily acquire profile data that indicates the tendency of targets that meet specific conditions. [Solution] The survey management server includes an extraction unit that compares attribute information of a target identified by a first condition with attribute information of a comparison target identified by a second condition from multiple monitors to extract characteristics of the target, a prompt generation unit that generates a prompt to be input into a generative AI model based on the extracted characteristics of the target, and a profile data acquisition unit that acquires profile data representing the target generated by the generative AI model by inputting the generated prompt into the generative AI model.

Description

[Technical Field]

[0001] The present invention relates to a questionnaire management server, a questionnaire analysis method, and a program. [Background technology]

[0002] One method for understanding the trends of a marketing target is to conduct a survey of multiple respondents (monitors) and analyze the response results. Patent Document 1 describes a marketing support system that tally up each respondent by brand usage frequency and attribute based on the survey response results and the respondent's attributes. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2020-119563 Summary of the Invention [Problem to be solved by the invention]

[0004] In order to provide effective marketing support to clients, it is desirable to analyze the trends of targets (respondents) that meet certain conditions based on questionnaire response results or attribute information, rather than simply collecting aggregated values, and to create and provide a document that shows the target's profile. However, with conventional technology, this work has been done manually, which is extremely time-consuming.

[0005] Therefore, one object is to provide a survey management server, a survey analysis method, and a program that can easily acquire profile data that indicates the tendencies of targets that meet specific conditions. [Means for solving the problem]

[0006] According to one aspect of the present disclosure, a survey management server includes an extraction unit that compares attribute information of a target identified by a first condition with attribute information of a comparison target identified by a second condition from multiple monitors to extract characteristics of the target, a prompt generation unit that generates a prompt to be input into a generative AI model based on the characteristics of the target extracted by the extraction unit, and a profile data acquisition unit that acquires profile data representing the target generated by the generative AI model by inputting the prompt generated by the prompt generation unit into the generative AI model.

[0007] According to one aspect of the present disclosure, a survey analysis method includes the steps of extracting characteristics of a target from multiple monitors by comparing attribute information of the target identified by a first condition with attribute information of a comparison target identified by a second condition; generating a prompt to be input into a generative AI model based on the extracted characteristics of the target; and acquiring profile data representing the target generated by the generative AI model by inputting the generated prompt into the generative AI model.

[0008] According to one aspect of the present disclosure, the program causes a survey management server to perform the following steps: extracting characteristics of the target by comparing attribute information of a target identified by a first condition with attribute information of a comparison target identified by a second condition from multiple monitors; generating a prompt to be input into a generative AI model based on the extracted characteristics of the target; and acquiring profile data representing the target generated by the generative AI model by inputting the generated prompt into the generative AI model. [Effects of the Invention]

[0009] According to the above aspect, it is possible to easily obtain profile data that indicates the tendency of a target that meets a specific condition. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is a diagram illustrating an example of a questionnaire system according to an embodiment. [Figure 2] 10 is a flowchart illustrating an example of processing by a questionnaire management server according to the embodiment. [Figure 3] FIG. 2 is a diagram for explaining an example of processing by a questionnaire management server and a generation AI model according to an embodiment. [Figure 4] FIG. 10 is a first diagram showing an example of a profile analysis result according to the embodiment. [Figure 5] FIG. 10 is a second diagram showing an example of a profile analysis result according to the embodiment. [Figure 6] 10 is a flowchart illustrating an example of processing by a questionnaire management server according to a modified example of an embodiment. [Figure 7] FIG. 1 is a schematic block diagram illustrating a configuration of a computer according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0011] <Embodiment> (System Configuration) 1 is a diagram showing an example of a survey system according to an embodiment. As shown in the figure, the survey system 1 includes respondent terminals 10a, 10b, ..., a client terminal 20, a survey management server 30, a big data DB (database) 40, and an external server 50. The respondent terminals 10a, 10b, etc. and the survey management server 30 are communicatively connected. The client terminals 20 and the survey management server 30 are communicatively connected. The big data DB 40 and the survey management server 30 are communicatively connected. The external server 50 and the survey management server 30 are communicatively connected.

[0012] The respondent terminals 10a, 10b, etc. are terminal devices such as smartphones and PCs (personal computers) that are used by respondents to a survey (hereinafter, respondents to a survey and candidate respondents may be referred to as monitors) when answering a survey. Respondents to a survey are selected for each survey, and the number of respondents varies for each survey. Therefore, the respondent terminals 10a, 10b, etc. may change for each survey, but for convenience, they will be referred to as respondent terminals 10a, 10b, etc. in this document. Although two respondent terminals 10a, etc. are shown in FIG. 1, three or more respondent terminals 10a, etc. may be present.

[0013] The client terminal 20 is a terminal device such as a smartphone or a personal computer (PC) used by a company or the like that requests the implementation and analysis of a survey. For example, a company developing a product targeting a certain demographic of users may create questions that can be used for product development and sales by specifying the target users' age, gender, place of residence, hobbies, and preferences. The client terminal 20 transmits survey implementation request information, including the survey target conditions and question content, to the survey management server 30 to request the implementation of the survey. The client terminal 20 also receives survey implementation results from the survey management server 30. While FIG. 1 illustrates one client terminal 20, multiple client terminals 20 may be present since various companies request the implementation of surveys. Note that a company may request the survey management server 30 to analyze only the surveys it has conducted.

[0014] The survey management server 30 is a server device operated by a research company that conducts and analyzes surveys at the request of companies and other organizations. For example, the research company has users who are willing to answer various surveys as members. The survey management server 30 selects monitors from the members for each survey and conducts the survey targeting the selected monitors. For example, members who answer a survey may be awarded points that can be used for online shopping, etc. The survey management server 30 accepts survey requests, conducts and compiles surveys, outputs survey results, and analyzes the profile of the analysis subject (target) based on the survey results. In this case, the survey management server 30 analyzes the characteristics of the target monitors and generates summaries of the analysis results using a generative AI (artificial intelligence) model. The survey management server 30 includes an input receiving unit 31, a survey processing unit 32, an extraction unit 33, a prompt generation unit 34, a profile data acquisition unit 35, an output unit 36, and a memory unit 37.

[0015] The input accepting unit 31 accepts requests to conduct a survey and analyze the target's profile. The input accepting unit 31 receives survey implementation request information from the client terminal 20, including the conditions of the survey subjects, question content, survey implementation period, delivery date, etc. The input accepting unit 31 may also receive analysis request information from the client terminal 20, including the conditions (first conditions) of the monitor who will be the subject (target) of profile analysis, and the conditions (second conditions) of the monitor who will be the comparison target. Note that when only a request for analysis of the survey results is accepted from a company, the input accepting unit 31 may also receive from the client terminal 20 the survey results, including the questions of the survey conducted by the company and the answers to those questions.

[0016] The survey processing unit 32 creates survey information based on the question content included in the implementation request information, extracts monitors based on the survey target conditions included in the implementation request information, distributes the survey information to the extracted monitors, collects responses to the survey, and registers them in the survey DB 372.

[0017] The extraction unit 33 compares the attribute information of the target specified by the first condition with the attribute information of the comparison target specified by the second condition from the multiple monitors, and extracts the characteristics of the target.

[0018] The prompt generation unit 34 generates a prompt to be input to the generative AI model 53 based on the target features extracted by the extraction unit 33. As shown in FIG. 1 , the generative AI model 53 includes a natural language model 51 that generates text based on an input prompt and an image generation AI model 52 that generates an image based on the input prompt. The prompt generation unit 34 generates a prompt to be input to the natural language model 51 and a prompt to be input to the image generation AI model. Note that in this embodiment, an example will be described in which the generative AI model 53 includes the natural language model 51 and the image generation AI model, but this is not limited to this. In other embodiments, the generative AI model 53 may be a single model that generates both text and an image based on an input prompt. In this case, the prompt generation unit 34 generates a prompt to generate both a profile statement and a profile image.

[0019] The profile data acquisition unit 35 acquires profile data representing the target generated by the generative AI model by inputting the prompt generated by the prompt generation unit 34 into the generative AI model 53. The profile data includes a profile sentence that expresses the characteristics of the target in text and a profile image that expresses the characteristics of the target in images. In the example of FIG. 1, the profile data acquisition unit 35 acquires the profile sentence generated by the natural language model 51 based on the prompt. Furthermore, when the prompt generation unit 34 generates a prompt for a profile image and inputs it to the image generation AI model 52, the profile data acquisition unit 35 acquires the profile image generated by the image generation AI model 52 based on this prompt.

[0020] The output unit 36 ​​outputs the survey results and the profile analysis results of the target to the client terminal 20.

[0021] The memory unit 37 stores various data. For example, the memory unit 37 stores a monitor DB 371 and a survey DB 372. The monitor DB 371 stores basic attribute information of members. For example, the monitor DB 371 stores demographic information such as age and gender, marketing information such as hobbies, preferences, and tendencies, points earned by members, and member rank, all linked to a monitor ID. The survey DB 372 stores responses by monitor for each survey. For example, the survey DB 372 stores monitor IDs and responses to each question in the survey, all linked to a survey ID. Note that in the configuration example of FIG. 1, the monitor DB 371 and the survey DB 372 are stored in the memory unit 37 of the survey management server 30, but these DBs may be stored in a storage device external to the survey management server 30 (for example, the big data DB 40).

[0022] The big data DB 40 stores survey results, such as large-scale surveys conducted independently by research companies on members, linked to monitor IDs. For example, the big data DB 40 stores lifestyle consciousness data, which is the result of a survey on attitudes and values ​​in various categories, such as lifestyle, shopping, and work, hobbies, and how time and money are spent. Large-scale surveys are not necessarily conducted for all members; some surveys are conducted only for a certain subset of members (e.g., those with a certain number of points or more). Furthermore, some members may not respond to a large-scale survey even if they are selected as targets. Therefore, the response results registered in the big data DB 40 vary from member to member. Since it is desirable to store as much data as possible about members' thoughts, attitudes, and behaviors in the big data DB 40, the survey management server 30 may register the results of surveys conducted at the request of a company with the company's approval. Various basic attribute information of monitors (members) may be registered in the big data DB 40 rather than in the monitor DB 371. The big data DB 40 may be registered in the storage unit 37 of the survey management server 30. In this embodiment, the basic attribute information (demographic information, marketing information, etc.) included in the monitor DB 371 and the lifestyle consciousness data included in the big data DB 40 are also collectively referred to simply as "attribute information."

[0023] The external server 50 has a generative AI model 53. In this embodiment, as illustrated in FIG. 1 , the external server 50 has, as the generative AI model 53, a natural language model 51 and an image generation AI model 52. The natural language model 51 generates and outputs text corresponding to an input prompt. The image generation AI model 52 generates and outputs an image corresponding to an input prompt. The natural language model 51 and the image generation AI model 52 may each be a general-purpose model that is used for various purposes, or may be a model obtained by independently retraining (fine-tuning) a general-purpose model. Note that, in this embodiment, an example will be described in which the generative AI model 53 has the natural language model 51 and the image generation AI model 52, but this is not limiting. In other embodiments, the generative AI model 53 may have only one model that generates both text and an image based on an input prompt. A generative AI model 53 having only one model may also be a general-purpose model or a model obtained by retraining a general-purpose model. In addition, the generation AI model 53 (the natural language model 51 and the image generation AI model 52) may be built in the questionnaire management server 30.

[0024] (Processing example) 2 is a flowchart showing an example of processing by the questionnaire management server according to the embodiment. Referring to FIG. 2, an example of processing by the questionnaire management server 30 will be described in detail.

[0025] First, the input accepting unit 31 accepts a request to conduct and analyze a survey (step S101). For example, the client terminal 20 transmits survey implementation request information including the conditions of the monitors who will be surveyed, the question content, the deadline, etc., and the input accepting unit 31 receives the survey implementation request information. In addition to accepting a survey implementation and analysis request from the client terminal 20, an operator who has received a request from a company may input the survey implementation request information and analysis request information to the survey management server 30. When the input accepting unit 31 acquires the survey implementation request information and analysis request information sent from the client terminal 20 or input directly, acceptance of the survey implementation and analysis request information is completed.

[0026] Next, the survey processing unit 32 conducts the survey based on the survey request information (step S102). The survey processing unit 32 creates survey information including a survey ID unique to the survey to be registered in the survey DB 372 and question information to be distributed to respondents. The question information may be automatically generated by an application or the like, or may be generated by an operator referencing the survey request information and inputting the question content as text. The question information is, for example, an HTML-formatted web page that displays questions in text and has an area where respondents can select answers to the questions using radio buttons or check boxes, and an area where they can enter answers as text. This web page is configured so that when respondents enter their answers and press a send button, the answers are sent to the survey management server 30. The survey processing unit 32 references the respondents' attribute information (basic attribute information stored in the survey DB 371, lifestyle consciousness data stored in the big data DB 40, etc.) to extract respondents who meet the survey target conditions of the survey request information. The survey processing unit 32 distributes the created question information to the extracted respondents.

[0027] The monitors use the respondent terminal 10 to check the distributed questions, input answers, and operate the respondent terminal 10 to send the input answers to the survey management server 30. In the survey management server 30, the survey processing unit 32 acquires the answers to the distributed question information. Upon acquiring the answers, the survey processing unit 32 registers the survey answers (survey results) in the survey DB 372 (step S103). The survey processing unit 32 links the answers to the monitor IDs and registers the answers of the monitors assigned with the monitor IDs in the survey DB 372.

[0028] Next, based on the registered survey results, a profile analysis is performed on target respondents among the survey respondents. First, the extraction unit 33 sets conditions (first and second conditions) for identifying targets and comparison targets (step S104). The first and second conditions are, for example, specified in advance by the company when requesting the implementation and analysis of the survey in step S101. For example, the first condition is information indicating that the respondent is a first-time purchaser (a new purchaser), and the second condition is information indicating that the respondent has purchased the brand's products multiple times (a repeat customer). One means of identifying information indicating a new purchaser or a repeat customer may be responses to purchase-related questions in the survey. For example, the first condition is a respondent who selected answer a to a purchase-related question (such as "first-time purchase" or "less than x purchases"), and the second condition is a respondent who selected answer b to the same question (such as "not a first-time purchase" or "more than x purchases"). In other embodiments, the first and second conditions may be information other than a new purchaser or a repeat customer. For example, the first condition may be a respondent who selected answer a (e.g., a positive answer) to question 1 in a questionnaire, and the second condition may be a respondent who selected a different answer b (e.g., a negative answer) to the same question 1. Furthermore, for example, the first condition may be all respondents to the questionnaire, and the second condition may be monitors who were not eligible for the questionnaire (did not respond) and whose lifestyle consciousness data is registered in the big data DB 40. The first and second conditions are not limited to these. Furthermore, each of the first and second conditions may be a combination of multiple conditions (e.g., a respondent who selected answer a to question 1 and answer b to question 2).

[0029] The extraction unit 33 also determines whether profile analysis is possible based on the number of targets and comparison targets identified based on the set first and second conditions (step S105). For example, if both the number n1 of targets identified from the first condition and the number n2 of comparison targets from the second condition are equal to or less than the minimum value (e.g., n1≦2 and n2≦2), the extraction unit 33 determines that analysis is impossible (step S105; NO). In this case, the extraction unit 33 transmits the determination result that analysis is impossible to the output unit 36, and the output unit 36 ​​outputs an analysis result to the client terminal 20 urging the client to change the conditions because sufficient profile analysis is not possible under the current conditions (step S110). The requesting company operates the client terminal 20 to change the first and second conditions and transmit them to the questionnaire management server 30. Thereafter, the process returns to step S104, and the extraction unit 33 resets the first and second conditions newly acquired through the input receiving unit 31. The output unit 36 ​​may output a result indicating that analysis is not possible on a display device (not shown) of the questionnaire management server 30, and the operator may refer to this result and request a change to the conditions via telephone or client terminal 20.

[0030] On the other hand, if the number of targets n1 and the number of comparison targets n2 are greater than the minimum value, it is determined that profile analysis is possible (step S105; YES). In this case, the extraction unit 33 compares differences in attribute information and extracts features of the targets (step S106). The attribute information includes basic attribute information contained in the monitor DB 371, lifestyle consciousness data contained in the big data DB 40, etc.

[0031] For example, the extraction unit 33 compares basic attribute information (demographic information such as age and gender, and marketing information such as hobbies and preferences) between the target and the comparison target, and calculates the difference between the target and the comparison target. For example, the extraction unit 33 calculates the difference between the average age of the target and the comparison target. The extraction unit 33 calculates the difference between the gender composition ratio of the target (female 0%, male 0%) and the gender composition ratio of the comparison target (female ±×%, male ±×). Similarly, the extraction unit 33 calculates the difference between the residential composition ratio of the target (Kanto 0%, Kansai 0%, ...) and the residential composition ratio of the comparison target (Kanto ±×%, Kansai ±×%, ...). The extraction unit 33 also calculates the difference between the hobby and preference composition ratio of the target (hobby A 0%, hobby B 0%, ...) and the hobby and preference composition ratio of the comparison target (hobby A ±×%, hobby B ±×%, ...). Differences are calculated in a similar manner for other basic attribute information. The extraction unit 33 also compares the proportions of lifestyle consciousness data (answers to each question) of the target and the comparison subject and calculates the difference. For example, the extraction unit 33 calculates the difference (±×%) between the proportion of respondents who selected answer a to question 1 among the target and the proportion of respondents who selected answer a to question 1 among the comparison subject. Note that question 1 is not a question from a survey conducted at the request of a client, but a question from a survey conducted to obtain lifestyle consciousness data. The extraction unit 33 similarly calculates the difference between the proportion of respondents in the target and the proportion of respondents in the comparison subject for all answers to each question. At this time, the extraction unit 33 extracts, from among multiple items of attribute information, items that have a significant difference between the target and the comparison subject as items that represent the characteristics of the target (such as the age group of monitors constituting the target and lifestyle consciousness trends). Specifically, the extraction unit 33 extracts, from the attribute information, items whose difference between the target and the comparison subject is equal to or greater than a threshold (for example, 5 points or more).

[0032] The extraction unit 33 extracts a predetermined number (e.g., 20 items) of items from the basic attribute information (demographic information, etc.) and the lifestyle consciousness data in descending order of difference as items representing the overall characteristics of the target. Furthermore, the lifestyle consciousness data may be divided into a plurality of categories A, B, C, etc. In this case, the extraction unit 33 may extract a predetermined number (e.g., 3 items) of items in descending order of difference for each of categories A, B, C, etc. as features representing the target's characteristics (attitudes) for each category. In this case, if there is no item in a certain category with a difference equal to or greater than a threshold, the extraction unit 33 may determine that this category has "no characteristic awareness."

[0033] Next, the prompt generation unit 34 generates a prompt based on the characteristics of the target and inputs it to the generation AI model 53 (the natural language model 51 and the image generation AI model 52) (step S107). Note that here, the prompt generated by the prompt generation unit 34 is a prompt that the generation AI model 53 uses to generate profile data representing the target. Specifically, the prompt generation unit 34 generates a prompt including the characteristics of the target extracted by the extraction unit 33. More specifically, the prompt generation unit 34 generates a prompt that the natural language model 51 uses to generate a profile sentence summarizing items that represent the overall characteristics of the target, and a category-specific profile sentence summarizing features that represent the characteristics (awareness tendencies) of the target for each category of the lifestyle consciousness data. In addition, the prompt generation unit 34 generates a prompt that the image generation AI model 52 uses to generate an image representing a person portrait that summarizes items that represent the overall characteristics of the target.

[0034] Prompts to be input to each model must be appropriately generated in order to obtain accurate profile sentences and profile images from the natural language model 51 and the image generation AI model 52. For this reason, the prompt generation unit 34 generates prompts to be input to each model, for example, by the method described below.

[0035] FIG. 3 is a diagram illustrating an example of processing by the questionnaire management server and the generation AI model according to the embodiment. First, an example of generating a prompt to be input to the natural language model 51 will be described with reference to FIG. 3. The prompt generation unit 34 has in advance text data P1, which is the instruction content to the natural language model 51 in the prompt. The text data P1 is, for example, a fixed phrase such as, "Please summarize the following three characteristics in Japanese using no more than 50 characters." The prompt generation unit 34 also extracts data (feature data P2) representing the target's characteristics extracted by the extraction unit 33 from the attribute information, as information required for the natural language model 51 to generate a profile sentence. Note that the prompt generation unit 34 may prepare structured data P3 in advance by structuring the attribute information, and extract the feature data P2 from the structured data P3. For example, assume that the extraction unit 33 extracts items (answers) a1, a3, and a6 as the target's characteristics for category A of the lifestyle consciousness data. In this case, the prompt generation unit 34 combines the text data P1 with the feature data P2, which is information (text) explaining the items (answers) a1, a3, and a6, to form a prompt sentence that can be accepted by the natural language model 51. In other words, the prompt generation unit 34 generates a prompt that includes pre-stored text data and the extracted features of the target. For example, the prompt generation unit 34 generates a prompt such as, "Please summarize the following three features in Japanese using no more than 50 characters: 'lifestyle consciousness data item a1,' 'lifestyle consciousness data item a3,' and 'lifestyle consciousness data item a6.'" as a prompt for generating a profile sentence for category A. The prompt generation unit 34 inputs the generated prompt for each category, along with attribute information of the target and comparison subject, into the natural language model 51. Then, the natural language model 51 generates a profile sentence that represents the features of the target in each category (e.g., category A of lifestyle consciousness) based on the instructions included in the prompt and the attribute information, which is the feature data P2 of the target. Furthermore, the prompt generation unit 34 generates a prompt for summarizing items that represent the overall features of the target.For example, suppose the extraction unit 33 extracts "average age" and "gender" from the basic attribute information and "item a1 of category A," "item b2 of category B," "item c1 of category C," etc. from the lifestyle consciousness data as items representing the characteristics of the target. In this case, the prompt generation unit 34 combines text data P1 with feature data P2, which is information explaining the extracted items, to form a prompt sentence that can be accepted by the natural language model 51. For example, the prompt generation unit 34 generates a prompt such as "Please summarize the following characteristics in Japanese using 50 characters or less: 'average age,' 'gender,' 'item a1 of category A of lifestyle consciousness data,' 'item b2 of category B of lifestyle consciousness data,' 'item c1 of category C of lifestyle consciousness data,' etc.," as a prompt for generating a profile sentence representing the overall characteristics of the target. Note that the prompt generation unit 34 does not generate a prompt for the category "no characteristic consciousness."

[0036] Similarly, the prompt generation unit 34 generates a prompt for generating an image representing a person image that summarizes items that represent the overall characteristics of the target. The prompt for generating this image representing a person image may be generated, for example, using the same items as the prompt for summarizing items that represent the overall characteristics of the target. The prompt generation unit 34 combines the text data P1 and the feature data P2 to form a prompt sentence that can be accepted by the image generation AI model 52. For example, the prompt generation unit 34 generates a prompt such as "Please generate a person image that summarizes the following characteristics: 'average age', 'gender', 'item a1 in category A of lifestyle consciousness data', 'item b2 in category B of lifestyle consciousness data', 'item c1 in category C of lifestyle consciousness data', ..." as a prompt for generating a profile image that represents the overall characteristics of the target.

[0037] Furthermore, the prompt generation unit 34 may generate prompts conforming to a predetermined "pattern" so that the natural language model 51 and the image generation AI model 52 can generate profile sentences and profile images with higher accuracy. Specifically, the prompt generation unit 34 generates prompts by attaching labels consisting of symbols and text indicating the meaning of information, such as "#role:," "#task:," or "#example:," to the beginning of instructions (text data P1) or feature data P2. This enables the natural language model 51 and the image generation AI model 52 to more accurately interpret the roles and instructions (tasks) required of each model, thereby improving the accuracy of the output profile sentences and profile images. The prompt generation unit 34 may also include a prompt history P4 as "#example:." The prompt generation unit 34 stores and accumulates the prompt history P4, which includes previously generated prompts and profile data generated by these prompts, in the storage unit 37, for example. The prompt generation unit 34 retrieves the prompt history P4 from the storage unit 37 and adds it to the prompt as "#example:." In this way, by providing at least one of a previously given prompt and generated profile data as an example, it is possible to generate profile data that is close to the desired content for each model.

[0038] Next, the profile data acquisition unit 35 inputs each prompt generated by the prompt generation unit 34 into the natural language model 51 and the image generation AI model 52, respectively, to acquire the target's profile sentence generated by the natural language model 51 and the profile image generated by the image generation AI model 52 (step S108). As shown in the example of FIG. 3, when the prompt generation unit 34 inputs a prompt into the input field, the natural language model 51 and the image generation AI model 52 generate profile sentences (profile sentences for each category and profile sentences representing overall characteristics) and profile images based on the input prompt and output them to the output field. The profile data acquisition unit 35 acquires the profile sentences and profile images output to the output field of the arc. The profile sentences and profile images are also collectively referred to as profile data.

[0039] Next, the output unit 36 ​​creates a profile summary D1 and a profile sheet D2 for the target based on the profile data generated by the generation AI model 53. As shown in the example of Fig. 3, the output unit 36 ​​inputs the profile sentence generated by the natural language model 51 and the profile image generated by the image generation AI 52 into predetermined locations in the profile summary D1 and the profile sheet D2, respectively. Once these have been created, the output unit 36 ​​outputs the survey results registered in the survey DB 372 and the created profile analysis results (profile summary and profile sheet) to the client terminal 20 (step S109).

[0040] (Example of profile analysis results) Fig. 4 is a first diagram showing an example of a profile analysis result according to an embodiment. Fig. 5 is a second diagram showing an example of a profile analysis result according to an embodiment. Fig. 4 shows an example of a profile summary D1, and Fig. 5 shows an example of a profile sheet D2.

[0041] As shown in FIG. 4, the profile summary D1 is data showing an overview of the target's profile and includes target information D11, comparison target information D12, profile summary D13, profile image D14, basic attribute information D15, lifestyle consciousness information D16, etc. The target information D11 includes the first condition used to identify the target and the number n1 of targets extracted under the first condition. The comparison target information D12 includes the second condition used to identify the comparison target and the number n2 of comparison targets extracted under the second condition. The profile summary D13 is a profile statement (summary) that expresses the overall characteristics of the target generated by the natural language model 51. The profile image D14 is an image that expresses the target's personality generated by the image generation AI model 52. The basic attribute information D15 is information that aggregates basic attribute information such as the target's demographic information by item, and includes the group to which the largest number of targets out of the total number n1 of targets belongs, the average age of the targets, etc. In the example of Figure 4, the item "gender" shows that the majority of targets belong to the "female" group. The lifestyle consciousness information D16 is a list of profile sentences (summary sentences) that express the characteristics of each of the target categories A, B, C, etc., generated by the natural language model 51.

[0042] As shown in FIG. 5, the profile sheet D2 includes target information D21, comparison target information D22, profile summary D23, profile image D24, basic attribute information D25, and lifestyle and attitude information D26. The target information D21, comparison target information D22, profile summary D23, and profile image D24 are identical to the corresponding items D11 to D24 in the profile summary D1. The basic attribute information D25 includes item information D251, which is identical to the basic attribute information D15 in the profile summary D1, and summary and comparison information D252 for each item. The summary and comparison information D252 indicates the difference between the composition ratio of the most numerous group in the total number of targets n1 and the composition ratio of the most numerous group in the comparison target. For example, for the "residence" item, the "Kanto" group is the most numerous, accounting for 45% of the total number of targets n1, and the difference from the composition ratio of the comparison target belonging to the "Kanto" group is +4.0%. The lifestyle consciousness information D26 represents items by category of lifestyle consciousness data extracted by the extraction unit 33 as characteristics of the target. The lifestyle consciousness information D26 includes extracted answer contents D261 and tally and comparison information D262 for each answer content. The tally and comparison information D262 indicates the difference between the percentage of targets who selected the extracted answer and the percentage of comparison subjects who selected the same answer. The answer contents D261 and tally and comparison information D262 are displayed in descending order of difference. In the example of FIG. 5, for category A, the difference in the percentage of targets who selected answer A6 was the largest between the target and comparison subjects, with 84% of targets selecting answer A6, a difference of +10.6% from the comparison subjects. For example, in category E, if there is no item whose difference from the comparison target is equal to or greater than the threshold and the extraction unit 33 sets "no distinctive consciousness," information D263 indicating no distinctive consciousness is displayed in the life consciousness information D26, instead of the answer content D261 and the summary and comparison information D262, as shown in FIG. 5.

[0043] (Variation) FIG. 6 is a flowchart showing an example of processing by a survey management server according to a modified example of the embodiment. While FIG. 2 illustrates an example in which the survey management server 30 creates and conducts a survey, the present invention is not limited to this. For example, a survey may be created and conducted by a company, with only the analysis of the survey results being performed by a research company (survey management server 30). An example of processing in this case will be described with reference to FIG. 6. First, in step S201, the input receiving unit 31 receives an upload of the survey results conducted by the company along with a request for survey analysis. Step S102 in FIG. 2 is omitted. Next, in step S202, the survey processing unit 32 registers the uploaded survey results in the survey DB 372. The processing in step S202 is the same as the processing in step S103 in FIG. 2. The subsequent processing in steps S203 to S209 is the same as steps S104 to S110 in FIG. 2, respectively. In addition, the requesting company can cause the questionnaire management server 30 to repeatedly execute steps S203 to S209 while changing the first and second conditions through the client terminal 20, thereby conducting profile analysis of the target according to various conditions.

[0044] 1 and 2 illustrate an example in which the prompt generation unit 34 generates a prompt for generating an image representing a person's portrait that summarizes items that represent the target's overall characteristics, inputs the generated prompt together with attribute information to the image generation AI model 52, and the image generation AI model 52 generates an image corresponding to the input prompt. However, the present invention is not limited to this. For example, the storage unit 37 may store multiple candidate images representing the target's portrait that have been created in advance by a research company, and the profile data acquisition unit 35 may use the candidate images stored in the storage unit 37 instead of the image generated by the image generation AI model 52. For example, each candidate image may be associated with multiple keywords that indicate attribute information. The keywords are, for example, keywords such as "female (gender)" or "Kanto (place of residence)" that indicate basic attribute information, or keywords included in the responses to the lifestyle consciousness data (such as "daily small luxuries" or "old traditions"). In this case, in step S107, the prompt generation unit 34 does not generate a prompt to be input to the image generation AI model 52. In addition, in step S108, the profile data acquisition unit 35 acquires, from among the multiple candidate images, a candidate image that includes the most keywords that match the characteristics extracted by the extraction unit 33 as the target's profile image. Note that the profile data acquisition unit 35 may, for example, require that the candidate image include a keyword that indicates a characteristic (such as "female") in the target's basic attribute information, and acquire, from among the candidate images extracted based on the required condition, the candidate image that includes the most keywords that indicate the characteristic in the target's lifestyle consciousness data as the target's profile image. This makes it possible to prevent a profile image that deviates from the target's image from being selected.

[0045] (effect) As described above, according to this embodiment, the survey management server 30 includes an extraction unit 33 that compares attribute information of a target identified by a first condition with attribute information of a comparison target identified by a second condition from multiple monitors to extract characteristics of the target, a prompt generation unit 34 that inputs the target's attribute information and a prompt generated based on the target's characteristics to a natural language model 51 that generates a profile sentence representing the target, and a profile data acquisition unit 35 that acquires the profile sentence generated by the natural language model 51 based on the prompt.

[0046] In this way, the questionnaire management server 30 can easily obtain, using the natural language model 51, profile data that indicates the tendency of a target that meets a specific condition.

[0047] In addition, the prompt generation unit 34 inputs the target's attribute information and a prompt generated based on the target's characteristics to the image generation AI model 52, which generates a profile image representing the target, and the profile data acquisition unit 35 acquires the profile image generated by the image generation AI model 52 based on the prompt.

[0048] In this way, the questionnaire management server 30 can easily obtain an image of the target person's portrait using the image generation AI model 52.

[0049] Furthermore, the extraction unit 33 compares the difference between the target and the comparison subject for each of a plurality of items of attribute information, and extracts an item for which the difference is greater than a predetermined threshold as a feature of the target.

[0050] In this way, the questionnaire management server 30 can appropriately extract attribute information that has a significant difference between the target and the comparison subject as a feature that represents the attribute of the target.

[0051] Furthermore, the extraction unit 33 extracts a predetermined number of items with larger differences from among the items with differences larger than a predetermined threshold value as target features.

[0052] In this way, the questionnaire management server 30 can appropriately narrow down attribute information that has a significant difference between the target and the comparison subject, thereby preventing unnecessary features with small differences from being included in the prompts generated by the prompt generation unit 34, and preventing important features with larger differences from being included in the profile sentences.

[0053] In addition, the extraction unit 33 extracts a predetermined number of items with larger differences from among the items with differences larger than a predetermined standard for each category of attribute information as features for each category of the target, and sets a category in which there are no items with differences larger than a predetermined standard to have no features.

[0054] In this way, the questionnaire management server 30 can exclude prompt generation for categories that are only slightly different from the comparison target by determining that they have no features. This prevents the prompts generated by the prompt generation unit 34 from including features that are only slightly different from the comparison target, resulting in inaccurate target profile statements.

[0055] Furthermore, if the number of targets and comparison objects is equal to or less than a predetermined number, the extraction unit 33 outputs a result indicating that analysis is not possible.

[0056] In this way, the questionnaire management server 30 can prevent outputting inaccurate analysis results when the number of targets and comparison subjects is insufficient.

[0057] (Computer Configuration) 7 is a schematic block diagram showing the configuration of a computer according to an embodiment. The computer 900 includes a processor 901, a main storage device 902, an auxiliary storage device 903, and an interface 904. The above-described survey management server 30 is implemented in the computer 900. The operations of the above-described processing units are stored in the auxiliary storage device 903 in the form of a program. The processor 901 reads the program from the auxiliary storage device 903, loads it into the main storage device 902, and executes the above-described processing in accordance with the program. The processor 901 also allocates a storage area in the main storage device 902 to be used in the above-described processing in accordance with the program.

[0058] The program may be for realizing some of the functions to be performed by the computer 900. For example, the program may be combined with other programs already stored in the auxiliary storage device 903 or other programs implemented in other devices to perform the functions. In other embodiments, the computer may include a custom LSI (Large Scale Integrated Circuit) such as a PLD (Programmable Logic Device) in addition to or instead of the above configuration. Examples of PLDs include PAL (Programmable Array Logic), GAL (Generic Array Logic), CPLD (Complex Programmable Logic Device), and FPGA (Field Programmable Gate Array). In this case, some or all of the functions realized by the processor may be realized by the integrated circuit.

[0059] Examples of the auxiliary storage device 903 include a hard disk drive (HDD), a solid state drive (SSD), a magnetic disk, a magneto-optical disk, a compact disc read-only memory (CD-ROM), a digital versatile disc read-only memory (DVD-ROM), and a semiconductor memory. The auxiliary storage device 903 may be an internal medium directly connected to the bus of the computer 900, or an external medium (external storage device 910) connected to the computer 900 via the interface 904 or a communication line. Furthermore, when this program is distributed to the computer 900 via a communication line, the computer 900 that receives the program may load the program into the main storage device 902 and execute the above-described processing. In at least one embodiment, the auxiliary storage device 903 is a non-transitory tangible storage medium.

[0060] (Appendix 1) A survey management server comprising: an extraction unit that extracts characteristics of a target from multiple monitors by comparing attribute information of the target identified by a first condition with attribute information of a comparison target identified by a second condition; a prompt generation unit that generates a prompt to be input into a generative AI model based on the characteristics of the target extracted by the extraction unit; and a profile data acquisition unit that acquires profile data representing the target generated by the generative AI model by inputting the prompt generated by the prompt generation unit into the generative AI model.

[0061] (Appendix 2) The survey management server according to claim 1, further comprising an output unit that creates a profile analysis result of the target using the profile data acquired by the profile data acquisition unit.

[0062] (Appendix 3) A survey management server as described in Appendix 1 or 2, wherein the generative AI model includes a natural language model that generates text based on an input prompt, the prompt generation unit generates a prompt to be input to the natural language model based on characteristics of the target, and the profile data acquisition unit acquires, as the profile data, a profile sentence representing the target generated by the natural language model.

[0063] (Appendix 4) A survey management server as described in any one of Appendices 1 to 3, wherein the generation AI model includes an image generation AI model that generates an image based on an input prompt, the prompt generation unit generates a prompt to be input into the image generation AI model based on the characteristics of the target, and the profile data acquisition unit acquires a profile image representing the target generated by the image generation AI as the profile data.

[0064] (Appendix 5) The survey management server according to any one of Appendix 1 to Appendix 4, wherein the extraction unit compares the difference between the target and the comparison subject for each of the attribute information of multiple items, and extracts items for which the difference is greater than a predetermined threshold as characteristics of the target.

[0065] (Appendix 6) The survey management server according to claim 5, wherein the extraction unit extracts a predetermined number of items with larger differences from among the items with the differences greater than a predetermined threshold as characteristics of the target.

[0066] (Appendix 7) The survey management server described in Appendix 5 or 6, wherein the extraction unit extracts, for each category of the attribute information, a predetermined number of items with the largest difference from among the items with the difference greater than a predetermined threshold as features for each category of the target, and sets categories in which there are no items with the difference greater than the predetermined standard to have no features.

[0067] (Appendix 8) A survey management server described in any one of Appendices 1 to 7, wherein the extraction unit outputs a result indicating that analysis is not possible if the number of the target and the number of the comparison objects are both below a predetermined number.

[0068] (Appendix 9) The survey management server according to any one of appendices 1 to 8, wherein the prompt generation unit generates the prompt by combining characteristics of the target with predetermined text data.

[0069] (Appendix 10) The survey management server according to claim 9, wherein the prompt generation unit generates the prompt by further combining explanatory text with a prompt history including previously generated prompts.

[0070] (Appendix 11) A survey management server as described in any one of Appendices 1 to 10, wherein the extraction unit identifies the target respondents from among the respondents to the survey based on a first condition and identifies the comparison respondents based on a second condition, the first condition being a respondent who gave a specific answer to a specific question in the survey, and the second condition being a respondent who gave an answer in the survey that is different from that of the target.

[0071] (Appendix 12) The survey management server according to claim 11, wherein the extraction unit accepts the first condition and the second condition setting.

[0072] (Appendix 13) A questionnaire analysis method comprising the steps of: extracting characteristics of a target from multiple monitors by comparing attribute information of the target identified by a first condition with attribute information of a comparison target identified by a second condition; generating a prompt to be input into a generative AI model based on the extracted characteristics of the target; and acquiring profile data representing the target generated by the generative AI model by inputting the generated prompt into the generative AI model.

[0073] (Appendix 14) A program that causes a survey management server to execute the following steps: extracting characteristics of a target from multiple monitors by comparing attribute information of the target identified by a first condition with attribute information of a comparison target identified by a second condition; generating a prompt to be input into a generative AI model based on the extracted characteristics of the target; and acquiring profile data representing the target generated by the generative AI model by inputting the generated prompt into the generative AI model. [Explanation of symbols]

[0074] 1. Survey system 10, 10a, 10b Respondent terminal 20 Client Terminals 30 Survey management server 31 Input reception section 32 Questionnaire processing section 33 Extraction part 34 Prompt Generation Unit 35 Profile data acquisition unit 36 Output section 37 Memory section 371 Monitor DB 372 Survey DB 50 external servers 51 Natural Language Model 52 Image generation AI models 53 Generative AI Models 40 Big Data DB

Claims

1. an extraction unit that compares attribute information of a target identified by a first condition with attribute information of a comparison target identified by a second condition from a plurality of monitors, and extracts features of the target; A prompt generation unit that generates a prompt to be input to a generative AI model based on the target features extracted by the extraction unit; a profile data acquisition unit that acquires profile data representing the target generated by the generative AI model by inputting the prompt generated by the prompt generation unit into the generative AI model; A questionnaire management server comprising:

2. an output unit that generates a profile analysis result of the target using the profile data acquired by the profile data acquisition unit; The questionnaire management server according to claim 1 .

3. the generative AI model includes a natural language model that generates text based on an input prompt; the prompt generator generates a prompt to be input to the natural language model based on the characteristics of the target; the profile data acquisition unit acquires, as the profile data, a profile sentence representing the target generated by the natural language model; The questionnaire management server according to claim 1 .

4. The generative AI model includes an image generation AI model that generates an image based on an input prompt; The prompt generation unit generates a prompt to be input to an image generation AI model based on the characteristics of the target; The profile data acquisition unit acquires, as the profile data, a profile image representing the target generated by the image generation AI model. The questionnaire management server according to claim 1 .

5. The extraction unit compares the difference between the target and the comparison subject for each of the plurality of items of attribute information, and extracts an item for which the difference is greater than a predetermined threshold as a feature of the target. The questionnaire management server according to claim 1 .

6. the extraction unit extracts a predetermined number of items in ascending order of difference from among the items in which the difference is greater than a predetermined threshold as the features of the target; The questionnaire management server according to claim 5 .

7. The extraction unit extracting, for each category of the attribute information, a predetermined number of items with larger differences from among the items with the differences larger than a predetermined threshold value as characteristics for each category of the target; A category in which there is no item in which the difference is greater than the predetermined standard is set to have no feature. The questionnaire management server according to claim 5 .

8. the extraction unit outputs a result indicating that analysis is not possible when the number of the target and the number of the comparison object are both equal to or less than a predetermined number. The questionnaire management server according to any one of claims 1 to 7.

9. the prompt generation unit generates the prompt by combining the characteristics of the target with predetermined text data. The questionnaire management server according to any one of claims 1 to 7.

10. the prompt generation unit generates the prompt by further combining explanatory text with a prompt history including previously generated prompts as examples. The questionnaire management server according to claim 9.

11. the extraction unit identifies the target respondents from among the respondents to the questionnaire based on a first condition, and identifies the comparison respondents based on a second condition; The first condition and the second condition are each a respondent who has given a specific answer to a specific question in the questionnaire. The questionnaire management server according to any one of claims 1 to 7.

12. the extraction unit accepts the first condition and the second condition setting; The questionnaire management server according to claim 11.

13. A step of comparing attribute information of a target specified by a first condition with attribute information of a comparison target specified by a second condition from a plurality of monitors to extract features of the target; generating prompts for input to a generative AI model based on the extracted features of the target; obtaining profile data representing the target generated by the generative AI model by inputting the generated prompt into the generative AI model; A survey analysis method having the above.

14. A step of comparing attribute information of a target specified by a first condition with attribute information of a comparison target specified by a second condition from a plurality of monitors to extract features of the target; generating prompts for input to a generative AI model based on the extracted features of the target; obtaining profile data representing the target generated by the generative AI model by inputting the generated prompt into the generative AI model; A program that causes the survey management server to execute the above.

Images