Article generation device

Abstract

To provide a sentence generating device capable of easily creating sentences showing a person's tendencies. [Solution] The system includes an answer extraction unit that extracts answer results for answer items to question items that match a second condition from answer items included in the survey results for respondents who match a first condition, among the answer results of a survey given to multiple respondents, and a sentence generation unit that generates a sentence expressing the tendencies of respondents who match the first condition in accordance with the answer results extracted by the answer extraction unit.

Description

[Technical field]

[0001] The present invention relates to a text generation device. [Background technology]

[0002] Conventionally, in marketing, when analyzing the image of a target person, a persona is created. As a technique for creating this persona, for example, there is a technique that uses information from SNS (for example, see Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2023-89233 A Summary of the Invention [Problem to be solved by the invention]

[0004] However, creating an accurate persona using sentences that indicate a person's tendencies is difficult and time-consuming.

[0005] Therefore, the present disclosure has been made in consideration of the above points, and aims to provide a sentence generation device that can easily create sentences that show a person's tendencies. [Means for solving the problem]

[0006] The first aspect of the text generation device includes an answer extraction unit that extracts answer results of answer items to question items that match a second condition from answer items included in the survey results of respondents who match a first condition among the response results of a survey given to a plurality of respondents, and a text generation unit that generates a text expressing the tendencies of the respondents who match the first condition in accordance with the answer results extracted by the answer extraction unit.

[0007] In addition, in a second aspect of the sentence generation device, the question item that meets the second condition is a question item for a characteristic answer item having a unique answer ratio among the answer items in the questionnaire, and the sentence generation unit generates the sentence using a word corresponding to the characteristic answer item.

[0008] In addition, in a third aspect of the sentence generation device, the characteristic answer item is an answer item for which the difference between the response results to the questionnaire for respondents who meet the first condition and the average of the response results to the questionnaire for respondents who meet a third condition different from the first condition is equal to or greater than a predetermined threshold.

[0009] In addition, in a fourth aspect of the sentence generation device, the second condition is that the question item is manually selected in advance.

[0010] In addition, the sentence generation device of the fifth aspect further includes a display unit that displays the sentence generated by the sentence generation unit and displays question items corresponding to the characteristic answer items used in sentence generation by the sentence generation unit.

[0011] In addition, in a sixth aspect of the sentence generation device, the display unit displays the comparison result of the question items as a graph.

[0012] In addition, the sentence generation device of the seventh aspect further includes an additional reception unit that receives, during or after the display of the sentence by the display unit, a selection of question items to be used in sentence generation by the sentence generation unit from among the question items corresponding to the characteristic answer items, and the sentence generation unit regenerates the sentence using words corresponding to the answer items to the question items selected through the additional reception unit, and the display unit displays the regenerated sentence.

[0013] Moreover, the sentence generation device of an eighth aspect includes an image generation unit that generates a face image of a respondent who meets the first condition from the sentence generated by the sentence generation unit. Effect of the Invention

[0014] According to the present disclosure, it is possible to provide a sentence generation device capable of easily creating sentences showing a person's tendencies. [Brief description of the drawings]

[0015] [Figure 1] 1 is a schematic block diagram of a sentence generation device according to an embodiment of the present disclosure. [Diagram 2] FIG. 11 is an explanatory diagram for explaining an example of a questionnaire according to an embodiment of the present disclosure. [Diagram 3] 1 is a block diagram illustrating an example of a functional configuration of a sentence generation device according to an embodiment of the present disclosure. [Figure 4] FIG. 11 is an explanatory diagram for explaining an example of a prompt according to an embodiment of the present disclosure. [Diagram 5] FIG. 2 is an explanatory diagram for explaining an example of a display by a display unit according to an embodiment of the present disclosure. [Figure 6A] 11 is an explanatory diagram for explaining the scoring of the average of the response results of a target and a comparison subject in the embodiment of the present disclosure. FIG. [Figure 6B] 11 is an explanatory diagram for explaining the scoring of the average of the response results of a target and a comparison subject in the embodiment of the present disclosure. FIG. [Figure 6C] 11 is an explanatory diagram for explaining the scoring of the average of the response results of a target and a comparison subject in the embodiment of the present disclosure. FIG. [Figure 6D] 11 is an explanatory diagram for explaining the scoring of the average of the response results of a target and a comparison subject in the embodiment of the present disclosure. FIG. [Figure 7A] 11 is an explanatory diagram for explaining the scoring of the average of the response results of a target and a comparison subject in the embodiment of the present disclosure. FIG. [Figure 7B] 11 is an explanatory diagram for explaining the scoring of the average of the response results of a target and a comparison subject in the embodiment of the present disclosure. FIG. [Figure 7C] 11 is an explanatory diagram for explaining the scoring of the average of the response results of a target and a comparison subject in the embodiment of the present disclosure. FIG. [Figure 7D] 11 is an explanatory diagram for explaining the scoring of the average of the response results of a target and a comparison subject in the embodiment of the present disclosure. FIG. [Figure 8] 1 is a flowchart illustrating an example of a flow of a sentence generation process performed by a sentence generation device according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In each drawing, the same or equivalent components and parts are given the same reference numerals. In addition, the dimensional ratios of the drawings are exaggerated for the convenience of explanation and may differ from the actual ratios.

[0017] An example of a text generation device 10 according to the present embodiment will be described with reference to FIG. FIG. 1 is a block diagram showing a hardware configuration of a text generation device 10 according to the present embodiment.

[0018] 1, a sentence generation device 10 according to this embodiment includes a CPU (Central Processing Unit) 101, a ROM (Read Only Memory) 102, a RAM (Random Access Memory) 103, a storage unit 104, an input unit 105, and a display device 106. Each component is connected to each other via a bus 107 so as to be able to communicate with each other.

[0019] The CPU 101 is a central processing unit, and executes various programs and controls each part. That is, the CPU 101 reads a program from the ROM 102 or the storage unit 104, and executes the program using the RAM 103 as a working area. The CPU 101 controls each of the above components and performs various arithmetic processing according to the program recorded in the ROM 102 or the storage unit 104. In this embodiment, the program is stored in the ROM 102 or the storage unit 104.

[0020] The ROM 102 stores various programs and various data. The RAM 103 temporarily stores programs or data as a working area. The storage unit 104 is configured with an HDD (Hard Disk Drive) or an SSD (Solid State Drive) and stores various programs including an operating system and various data.

[0021] Here, the information stored in the storage unit 104 in this embodiment includes, for example, a text generation model and an image generation model. The text generation model is used by the text generation unit 112 described later. The text generation model is publicly known, for example, as disclosed in ChatGPT (https: / / openai.com / index / chatgpt / ), and therefore a detailed description thereof will be omitted. Such a text generation model is configured by a large language model (LLM). The image generation model is used by the image generation unit 113 described later. The image generation model is publicly known, for example, as disclosed in DALL-E3 (https: / / openai.com / index / dall-e-3 / ), and therefore a detailed description thereof will be omitted. Note that the text generation model and the image generation model are not limited to being stored in the storage unit 104, and may be stored in another device, for example, a server device connected to the text generation device 10 via a network.

[0022] The information stored in the storage unit 104 includes questionnaire questions and answers.

[0023] In addition, the sentence generation model, the image generation model, and some or all of the questionnaire questions and answer results are not limited to being stored in the sentence generation device 10, but may be stored in another device, such as a server device connected to the sentence generation device 10 via a network.

[0024] Here, an example of a questionnaire used in this embodiment will be described. As the questionnaire, for example, as shown in the figure, various question items such as "occupation", "employment status", "individual annual income", "marital status", "highest level of education", "gender", "topics of interest", and "frequency of cooking" are set. Then, answer candidates for the question items are predefined for each question item, and the respondent can answer the question item by selecting an answer candidate. For example, as shown in FIG. 2, for the question item "occupation", answer candidates are predefined as "clerical work", "technical work", "service work", "executive", "unemployed", "pensioner", and "other". Note that the question items and answer candidates of the questionnaire are not limited to those shown in FIG. 2, and may be other question items and answer candidates. Here, in this embodiment, for example, a "general consumer survey" is used as the questionnaire, but is not limited thereto.

[0025] The input unit 105 includes a pointing device such as a mouse and a keyboard, and is used for various inputs. The input unit 105 is used for inputting settings of a first condition and a third condition, which will be described later.

[0026] The display device 106 is, for example, a liquid crystal display. The display device 106 displays various information under the control of the CPU 101.

[0027] Next, the functional configuration realized by the sentence generation device 10 will be described. FIG. 3 is a block diagram showing an example of a functional configuration of the CPU 101 of the sentence generation device 10. As shown in FIG.

[0028] 3, the writing generation device 10 has, as its functional configuration, a reception unit 110, an answer extraction unit 111, a writing generation unit 112, an image generation unit 113, a display unit 114, and an additional reception unit 115. Each functional configuration is realized by the CPU 101 reading and executing a program stored in the ROM 102 or the storage unit 104. Here, the writing generation device 10 of the present embodiment is a device that executes a process of concretely forming, as a persona, an image of a target person to be targeted in marketing or the like, through writing.

[0029] (Reception unit 110) The reception unit 110 receives the first condition and the third condition input by the user. Here, the first condition is a condition that can be arbitrarily set by the user of the text generation device 10, and is a condition for determining a target persona for generating a persona. For example, "a woman in her 60s who is highly health conscious" is an example. Then, the reception unit 110 extracts "a woman in her 60s who is highly health conscious" from the answer to one question item or from the answers to multiple question items in order to extract a respondent who matches the first condition "a woman in her 60s who is highly health conscious" from among the respondents to the questionnaire. For example, "health", "60s", and "female" are extracted as keywords from the first condition "a woman in her 60s who is highly health conscious", and respondents who selected "health" in the question item "topic of interest" or whose answer to the question item includes the keyword "health" and selected "60s" in the question item "age" and "female" in the question item "gender" are extracted. The extraction of respondents is not limited to the above-mentioned manner, and may be performed in other manners. For example, respondents to a questionnaire may be clustered in advance based on the answers to questions (for example, a group that is particular about health, a group that likes traveling, etc.), and respondents may be extracted based on the results of the clustering.

[0030] The third condition is a condition that can be arbitrarily set by the user of the text generation device 10, and is a condition for determining a comparison target person image to be compared with a target person image for generating a persona. For example, it includes "all respondents to the questionnaire", "women in their 60s" and "women" that include all of the first condition "women in their 60s who are highly health conscious" and exclude some of the first condition, "respondents other than those in the first condition" that are not included in the first condition, and "respondents who gave predetermined answers to specific question items" that are unrelated to the first condition. The reception unit 110 then extracts respondents who satisfy this third condition from among the respondents to the questionnaire. Hereinafter, respondents who satisfy the first condition are also called "targets", and respondents who satisfy the third condition are also called "comparison targets".

[0031] In addition, the receiving unit 110 is not limited to receiving the first condition and the third condition and extracting respondents who meet the conditions, but may also receive information on respondents extracted by another device of the text generation device 10.

[0032] The reception unit 110 also receives candidate question items manually selected in advance by the user from among the questionnaire question items. Here, the candidate question items are candidate question items used when the sentence generation unit 112 generates a target person image. The candidate question items are not limited to those input by the user, and may be frequently used question items that are stored in advance as candidates. The reception unit 110 may also extract question items related to the first condition, such as question items including the keywords "health," "60s," and "female" extracted from the first condition "health-conscious woman in her 60s" input by the user.

[0033] (Answer extraction part 111) The answer extraction unit 111 extracts candidates for the question items received by the reception unit 110 from the answer results of the questionnaire given to a plurality of respondents, and extracts answer results for the question items that match the second condition from the answer items included in the questionnaire results for the respondents that match the first condition. Here, the question items that match the second condition are question items for characteristic answer items whose answer ratio is unique among the answer items of the questionnaire. In this embodiment, from among the question candidates received by the reception unit 110, answer items for which the difference between the answer results of the questionnaire for the respondents that match the first condition and the average answer results of the questionnaire for the respondents that match a third condition different from the first condition is equal to or greater than a predetermined threshold are set as characteristic answer items.

[0034] Specifically, the answer extraction unit 111 first extracts the answer results of the target and the comparison target for the candidate question items received by the reception unit 110 from the answer items included in the survey result of the target. Then, the answer extraction unit 111 scores the average of the extracted answer results of the target and the comparison target, and extracts answer items whose score difference is equal to or greater than a predetermined threshold as characteristic answer items to be used by the sentence generation unit 112 when generating a person image of the target. Here, the predetermined threshold can be arbitrarily set by the user. The scores will be described later with reference to FIGS. 6A, 6B, 6C, 6D, 7A, 7B, 7C, and 7D.

[0035] (Sentence generation unit 112) The sentence generating unit 112 generates sentences expressing the tendency of the respondent who meets the first condition according to the answer results extracted by the answer extracting unit 111. Specifically, the sentence generating unit 112 generates sentences expressing the tendency of the respondent who meets the first condition by using words corresponding to the characteristic answer items extracted by the answer extracting unit 111. That is, as shown in FIG. 4, the sentence generating unit 112 generates a prompt by adding the fixed phrases "instruction", "constraint condition", and "output format" to the characteristic answer items extracted by the answer extracting unit 111, and inputs the prompts to the above-mentioned sentence generation model. In the example shown in FIG. 4, the characteristic answer items are written in the "# target" column. Here, the "# target" column includes answer items that are not characteristic answer items but are predetermined by the user and are always used. For example, "individual annual income", "gender", "married couple employment status", etc. are set to be included in the "# target" column even if they do not correspond to the characteristic answer items. That is, the sentence generating unit 112 generates sentences by using words corresponding to the characteristic answer items and the always used answer items. In this embodiment, the image of a person represented by a sentence generated by the sentence generation unit 112 is a persona that is the target of marketing. Note that the prompt is not limited to being used as is as generated by the sentence generation unit 112, and the user may be able to modify it as appropriate. The sentence generated by the sentence generation unit 112 may be able to be modified as appropriate by the user. The sentence generation unit 112 may also generate a prompt that creates a headline for the generated sentence with a number of characters specified by the user, thereby making it possible to generate a headline as well as generate a sentence.

[0036] (Image generation unit 113) The image generation unit 113 generates a face image of a respondent who meets the first condition from the sentence generated by the sentence generation unit 112. Specifically, the image generation unit 113 inputs the sentence generated by the sentence generation unit 112 into the above-mentioned image generation model. As a result, the image generation unit 113 generates a face image representing the portrait of the target.

[0037] (Display section 114) The display unit 114 displays the sentences generated by the sentence generation unit 112 on the display device 106, and also displays question items corresponding to the characteristic answer items used in generating the sentences by the sentence generation unit 112. As described above, the user may be allowed to appropriately modify the displayed sentences.

[0038] Thereafter, the display unit 114 displays on the display device 106 the face image generated by the image generation unit 113 and a graph of the comparison result between the question item and the question item. Specifically, as shown in FIG. 5, the display unit 114 displays the sentence in the sentence display unit S on the upper left side of the display area of ​​the display device 106, the face image in the face image display unit T on the upper right side, and a graph of the comparison result between the question item and the question item in the graph display unit U on the lower side. The display unit 114 also displays the heading of the sentence displayed in the sentence display unit S in the heading display unit V on the upper side of the sentence display unit S. Here, the display of the graph of the comparison result of the question items is specifically displayed by dividing the data of the answer result for the question items such as "married or not", "employment type", and "highest level of education" used when the sentence generation unit 112 generates a sentence and inputted to the sentence generation model into a target and a comparison target. Note that the display of the graph of all the question items is not limited to the case of displaying the graph of all the question items, and only the question items whose scores are higher than a predetermined threshold value may be displayed. The display order of the graphs may be the order in which the sentence generation unit 112 used them to generate sentences, the order in which sentences related to the question items are included in the generated sentences, or the order of the largest difference in scores. The display unit 114 is not limited to displaying sentences generated by the sentence generation unit 112 on the display device 106, and may execute a process of outputting sentences generated by the sentence generation unit 112 to another device connected to the sentence generation device 10 via a network.

[0039] In addition, the display unit 114 is not limited to displaying on the display device 106 the sentence generated by the sentence generation unit 112 and question items corresponding to the feature answer items used in generating the sentence by the sentence generation unit 112, and then displaying the facial image generated by the image generation unit 113 and a graph of the comparison results of the question items, but may display the sentence, the question items, the facial image, and the graph at the same time.

[0040] (Additional Reception Section 115) The additional reception unit 115 receives selection of question items to be used in sentence generation by the sentence generation unit 112 from among question items corresponding to the characteristic answer items during or after the display unit 114 displays the sentence. The sentence generation unit 112 then regenerates a prompt using words corresponding to answer items to the question items selected through the additional reception unit 115, and regenerates a sentence using the prompt. The display unit 114 then displays the regenerated sentence on the display device 106.

[0041] Next, the calculation of the average score of the answer results of the target and the comparison subject will be described with reference to Figures 6A, 6B, 6C, 6D, 7A, 7B, 7C, and 7D.

[0042] FIG. 6A is an example showing the answer result of a questionnaire before processing for a question item for which multiple answers can be input, for example, "hobby". For example, a person with a respondent ID "1111" is not included in the target (the "target" column is blank), is included in the comparison target (the "comparison target" column is "○"), and has a weight back of "0.9", which indicates that the person has answered "baseball" and "soccer" to the question item "hobby". Also, a person with a respondent ID "2222" is included in the target (the "target" column is "○"), is included in the comparison target (the "comparison target" column is "○"), and has a weight back of "1", which indicates that the person has answered "baseball", "driving", and "hot spring" to the question item "hobby". In the figure, "1" is input for the answer candidate selected by the respondent, and "0" is input for the answer candidate not selected by the respondent. Here, the weight back is a value that can be set by the user, and is a value that corrects the answer result of the questionnaire according to the bias of the population.

[0043] Figure 6B is a diagram in which the "0" indicating the unselected answer candidates in Figure 6A is converted to "-1", because a weighting cannot be applied to "0".

[0044] Figure 6C is a diagram where the results of Figure 6B are tabulated with weighting. That is, the numerical values ​​in the answer column of Figure 6B ("1" or "-1") are multiplied by the weighting, and the numerical values ​​are substituted for the numerical values ​​in the answer column and tabulated. For example, for respondent ID "1111", baseball "1", soccer "1", driving "-1", and hot springs "-1" in Figure 6B are multiplied by the weighting "0.9", respectively, and are tabulated as baseball "0.9", soccer "0.9", driving "-0.9", and hot springs "-0.9".

[0045] FIG. 6D is a diagram showing the scores tabulated in FIG. 6C. Specifically, for the "target average", the average of the answer columns of respondent IDs "2222" and "3333" marked with "○" in the "target" column in FIG. 6C is tabulated. For example, for "baseball", the average of "1" of respondent ID "2222" and "1.2" of respondent ID "3333" is tabulated as "1.1". For "soccer", the average of "-1" of respondent ID "2222" and "-1.2" of respondent ID "3333" is tabulated as "-1.1". For "comparison target average", the average of the answer columns of respondent IDs "1111", "2222", "3333", "4444", and "5555" marked with "○" in the "comparison target" column in FIG. 6C is tabulated. For example, for "baseball," the average of "0.9" for respondent ID "1111," "1" for respondent ID "2222," "1.2" for respondent ID "3333," "-1.3" for respondent ID "4444," and "-0.8" for respondent ID "5555" is calculated as "0.2." For "soccer," the average of "0.9" for respondent ID "1111," "-1" for respondent ID "2222," "-1.2" for respondent ID "3333," "1.3" for respondent ID "4444," and "0.8" for respondent ID "5555" is calculated as "0.16." For "difference," the difference between the calculated "target average" and the "comparison average" is calculated. For example, for "baseball", the difference between the target average "1.1" and the comparison target average "0.2" is calculated as "0.9". For "soccer", the difference between the target average "-1.1" and the comparison target average "0.16" is calculated as "-1.26". The answer extraction unit 111 extracts answer items whose "difference" is equal to or greater than a predetermined threshold as characteristic answer items to be used by the sentence generation unit 112 when generating a target person image. Here, when the difference is a "+ (plus)" value, the characteristic answer item part of the sentence generated by the sentence generation unit 112 is a positive description (e.g., "does ~", "prefers ~", "is high", etc.), and when the difference is a "- (minus)" value, the characteristic answer item part of the sentence generated by the sentence generation unit 112 is a negative description (e.g., "does not ~", "avoids ~", "is low", etc.).In addition, when the content of the question item is a negative description (for example, "I don't like to do ~"), if the difference is a "+ (plus)" value, the feature answer item part of the sentence generated by the sentence generating unit 112 will be a negative description (for example, "I don't do ~", "Avoid ~", "Low ~", etc.), and when the difference is a "- (minus)" value, the feature answer item part of the sentence generated by the sentence generating unit 112 will be a positive description (for example, "I do ~", "I like ~", "High ~", etc.). In addition, when there are multiple answer items whose differences are equal to or greater than a predetermined threshold, the sentence generating unit 112 may generate a sentence using all the feature answer items whose differences are equal to or greater than the predetermined threshold, or the sentence generating unit 112 may generate a sentence using only the feature answer item whose difference value is the largest, or the sentence generating unit 112 may generate a sentence using only the difference of the magnitude of the rank predetermined by the user's setting.

[0046] FIG. 7A shows an example of the response results of a questionnaire before processing for a question item for which only one answer can be input, such as "I like luxury brands," in the case where the answer candidates have different levels of degree, such as "1: Very applicable," "2: Somewhat applicable," "3: Not very applicable," and "4: Not applicable at all." For example, a person with a respondent ID of "1111" is not included in the target (the "target" column is blank), is included in the comparison target (the comparison target column is "○"), and has a weighting of "0.9," and has answered the question item "I like luxury brands" as "Very applicable." Also, a person with a respondent ID of "2222" is not included in the target (the "target" column is blank), is included in the comparison target (the comparison target column is "○"), and has a weighting of "1," and has answered the question item "I like luxury brands" as "Somewhat applicable." In FIG. 7A, a "1" is entered for the answer candidate selected by the respondent, and a "0" is entered for the answer candidate not selected by the respondent.

[0047] Figure 7B is a diagram in which the "0" indicating the unselected answer candidates in Figure 7A is converted to "-1", because a weighting cannot be applied to "0".

[0048] Figure 7C is a diagram where the results of Figure 7B are tabulated with weighting. That is, the numerical values ​​in the answer column of Figure 7B ("1" or "-1") are multiplied by the weighting, and the numerical values ​​are substituted for the numerical values ​​in the answer column and tabulated. For example, for respondent ID "1111," the values ​​in Figure 7B for "very applicable" ("1"), "somewhat applicable" ("-1"), "not really applicable" ("-1"), and "not at all applicable" ("-1") are each multiplied by a weighting of "0.9," and tabulated as "very applicable" ("0.9"), "somewhat applicable" ("-0.9"), "not really applicable" ("-0.9"), and "not at all applicable" ("-0.9").

[0049] FIG. 7D is a diagram showing the scores tabulated in FIG. 7C. Specifically, for the "target average", the average of the answer columns of respondent IDs "2222", "3333", "6666", "7777", and "8888" with "○" in the "target" column in FIG. 7C is tabulated. For example, for "very applicable", the average of "-1" for respondent ID "2222", "-1.2" for respondent ID "3333", "-0.9" for respondent ID "6666", "-1" for respondent ID "7777", and "-1.2" for respondent ID "8888" is tabulated as "-1.06". For "somewhat applicable," the average of "1" for respondent ID "2222," "-1.2" for respondent ID "3333," "0.9" for respondent ID "6666," "-1" for respondent ID "7777," and "-1.2" for respondent ID "8888" is calculated as "-0.3." For "average of comparison," the average of the response columns of respondent IDs "1111," "2222," "3333," "4444," "5555," "6666," "7777," "8888," and "9999" with "○" in the "comparison" column in FIG. 7C is calculated. For example, for "Very applicable," the average of "0.9" for respondent ID "1111," "-1" for respondent ID "2222," "-1.2" for respondent ID "3333," "-1.3" for respondent ID "4444," "0.8" for respondent ID "5555," "-0.9" for respondent ID "6666," "-1" for respondent ID "7777," "-1.2" for respondent ID "8888," and "1.3" for respondent ID "9999" is calculated as "-0.4." For "somewhat applicable," the average of "-0.9" for respondent ID "1111," "1" for respondent ID "2222," "-1.2" for respondent ID "3333," "-1.3" for respondent ID "4444," "-0.8" for respondent ID "5555," "0.9" for respondent ID "6666," "-1" for respondent ID "7777," "-1.2" for respondent ID "8888," and "-1.3" for respondent ID "9999" is calculated as "-0.64." For "difference," the difference between the calculated "target average" and the "comparison target average" is calculated.For example, for "very applicable", the difference between the target average "-1.06" and the comparison target average "-0.4" is calculated as "-0.66". For "somewhat applicable", the difference between the target average "-0.3" and the comparison target average "-0.64" is calculated as "0.34". The answer extraction unit 111 extracts answer items whose "difference" is equal to or greater than a predetermined threshold as characteristic answer items to be used by the sentence generation unit 112 when generating a target person image. Here, when the difference is a "+ (plus)" value, the characteristic answer item part of the sentence generated by the sentence generation unit 112 is a positive description (e.g., "does ~", "prefers ~", "is high", etc.), and when the difference is a "- (minus)" value, the characteristic answer item part of the sentence generated by the sentence generation unit 112 is a negative description (e.g., "does not ~", "avoids ~", "is low", etc.). In addition, when the content of the question item is a negative description (for example, "I don't like to do ~"), if the difference is a "+ (plus)" value, the feature answer item part of the sentence generated by the sentence generating unit 112 will be a negative description (for example, "I don't do ~", "Avoid ~", "Low ~", etc.), and when the difference is a "- (minus)" value, the feature answer item part of the sentence generated by the sentence generating unit 112 will be a positive description (for example, "I do ~", "I like ~", "High ~", etc.). In addition, when there are multiple answer items whose differences are equal to or greater than a predetermined threshold, the sentence generating unit 112 may generate a sentence using all the feature answer items whose differences are equal to or greater than the predetermined threshold, or the sentence generating unit 112 may generate a sentence using only the feature answer item whose difference value is the largest, or the sentence generating unit 112 may generate a sentence using only the difference of the magnitude of the rank predetermined by the user's setting.

[0050] Next, the operation of the writing generation device 10 will be described with reference to Fig. 8. Fig. 8 is a flowchart showing an example of the flow of writing generation processing by the writing generation device 10. The processing is performed by the CPU 101 reading a program from the ROM 102 or the storage unit 104, developing and executing the program. Also, as a premise of this flowchart, the writing generation device 10 stores questionnaire question items and answer results in the storage unit 104.

[0051] In step S100, the CPU 101 (acceptor 110) of the text generation device 10 accepts a first condition that is a condition for determining a target character image for generating a persona, and a third condition that is a condition for determining a comparison target character image to be compared with the target character image for generating a persona. Then, the process proceeds to the next step S101.

[0052] In step S101, the CPU 101 (acceptor unit 110) of the text generation device 10 extracts respondents who satisfy the first condition and those who satisfy the third condition from the respondents to the questionnaire. Then, the process proceeds to the next step S102.

[0053] In step S102, the CPU 101 (acceptor 110) of the text generation device 10 accepts candidate question items manually selected in advance by the user from among the question items of the questionnaire, and then the process proceeds to the next step S103.

[0054] In step S103, the CPU 101 (answer extraction unit 111) of the text generation device 10 converts the average of the questionnaire answer results for the target and the comparison subject from the candidate question items received in step S102 into scores, calculates the difference, and extracts characteristic answer items whose difference is equal to or exceeds a predetermined threshold value. Then, the process proceeds to the next step S104.

[0055] In step S104, the CPU 101 (sentence generation unit 112) of the sentence generation device 10 generates a prompt using the characteristic answer items extracted in step S103 and a predetermined fixed phrase, and then proceeds to the next step S105.

[0056] In step S105, the CPU 101 (sentence generation unit 112) of the sentence generation device 10 inputs the prompt generated in step S104 into the sentence generation model, and then proceeds to the next step S106.

[0057] In step S106, the CPU 101 (sentence generation unit 112) of the sentence generation device 10 generates sentences expressing the tendency of the target using the sentence generation model, and then proceeds to the next step S107.

[0058] In step S107, the CPU 101 (image generating unit 113) of the text generating device 10 generates a face image representing the target person's image based on the text generated in step S106. Here, the face image may be generated when an input for executing generation of a face image by a user is received, or the CPU 101 (image generating unit 113) may generate the face image regardless of the user's input. Then, the process proceeds to the next step S108.

[0059] In step S108, the CPU 101 (display unit 114) of the sentence generation device 10 displays the sentence generated in step S106, a graph of the comparison result of the question items corresponding to the characteristic answer items used in sentence generation by the sentence generation unit 112, and the face image generated in step S107 on the display device 106 (see FIG. 5). Then, the process ends.

[0060] Although not shown, during or after display of the sentence by the display unit 114, a selection of question items to be used in sentence generation by the sentence generation unit 112 from among question items corresponding to the characteristic answer items may be accepted. In this case, the sentence generation unit 112 regenerates the sentence using words corresponding to the answer items to the question items selected through the addition acceptance unit 115, and the display unit 114 displays the regenerated sentence on the display device 106. The image generation unit 113 generates a face image representing the portrait of the target based on the regenerated sentence.

[0061] According to this embodiment, it is possible to easily create text that shows the tendencies of people who are the target of marketing. That is, when analyzing the image of a target person in marketing, a persona is created, but since it is done by relying on the experience and intuition of the user, there are problems in that it is a heavy burden and there is a large variation in accuracy. In this embodiment, it is possible to solve such problems.

[0062] The present disclosure is not limited to the above-described embodiment, and various modifications other than those described above are possible without departing from the spirit of the present disclosure.

[0063] The characteristic answer items are not limited to the above-mentioned case of comparing the scores and extracting from answer items whose difference is equal to or greater than a predetermined threshold, but may be extracted by other methods or by combining a plurality of methods. For example, when the answer items are expressed numerically, such as annual income or age, the average numerical values ​​of the target and comparison subject may be compared, and characteristic answer items may be extracted from answer items whose difference is equal to or greater than a predetermined threshold.

[0064] In addition, extraction of characteristic answer items may be performed using machine learning. Specifically, the answer numbers of the answer candidates of all question items in the questionnaire are used as explanatory variables (for example, values ​​from 1 to 7 (see FIG. 2)), and a model is constructed as a "discrimination problem" in which the respondent who selected the answer candidate is the target or the comparison subject as the objective variable. Then, explanatory variables (question items) derived when constructing such a model may be extracted as characteristic answer items that have a high degree of importance.

[0065] In addition, a model may be constructed that learns the target response results, the comparison target response results, and the characteristic response items, and the characteristic response items may be extracted by inputting the current target response results and the comparison target questionnaire and response data.

[0066] The present disclosure can also be applied to a program. In the above embodiment, the program is pre-stored (installed) in the ROM 102 or the storage unit 104, but the present disclosure is not limited to this. The program may be provided in a form recorded on a recording medium such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), or a USB (Universal Serial Bus) memory. The program may also be downloaded from an external device via a network.

[0067] The program of the present disclosure can be provided as a program product. The program product includes any type of product for providing the program. For example, the program product includes a program provided through a network such as the Internet, and a non-transitory computer-readable recording medium such as a CD-ROM or DVD on which the program is stored.

[0068] In addition, the following supplementary notes are further disclosed regarding one embodiment of the technology disclosed in the present application described above.

[0069] (Appendix 1) an answer extraction unit that extracts answer results of answer items to question items that meet a second condition from answer items included in the survey results of respondents that meet a first condition among the survey response results given to a plurality of respondents; a sentence generating unit that generates a sentence expressing a tendency of an answerer that meets the first condition according to the answer results extracted by the answer extracting unit; A sentence generation device comprising:

[0070] (Appendix 2) The question item that meets the second condition is a question item for a characteristic answer item having a unique answer ratio among the answer items of the questionnaire, The sentence generation device according to claim 1, wherein the sentence generation unit generates the sentences using words corresponding to the characteristic answer items.

[0071] (Appendix 3) The text generation device described in Appendix 2, wherein the characteristic answer item is an answer item for which the difference between the answer results of the questionnaire for respondents who meet the first condition and the average of the answer results of the questionnaire for respondents who meet a third condition different from the first condition is equal to or greater than a predetermined threshold.

[0072] (Appendix 4) 4. The sentence generation device according to claim 1, wherein the second condition is that the question item is manually selected in advance.

[0073] (Appendix 5) The sentence generation device according to claim 2 or 3, further comprising a display unit that displays the sentence generated by the sentence generation unit and displays question items corresponding to the characteristic answer items used in generating the sentence by the sentence generation unit.

[0074] (Appendix 6) The sentence generation device according to claim 5, wherein the display unit displays a comparison result of the question items as a graph.

[0075] (Appendix 7) The display unit may further include an additional reception unit that receives selection of question items to be used in the sentence generation by the sentence generation unit from among question items corresponding to the characteristic answer items during or after the display of the sentence by the display unit, the sentence generation unit regenerates a sentence using words corresponding to an answer item to the question item selected through the additional reception unit; 7. The text generation device according to claim 5, wherein the display unit displays the regenerated text.

[0076] (Appendix 8) A text generation device as described in any one of Supplementary Note 1 to Supplementary Note 7, comprising an image generation unit that generates a facial image of a respondent who meets the first condition from the text generated by the text generation unit. [Explanation of symbols]

[0077] 10 Sentence generator 111 Answer extraction part 112 Sentence generation section 113 Image Generation Unit 114 Display section 115 Additional Reception Department

Claims

1. A computer-implemented method for detecting a temperature rise in a computer-implemented manner comprising: The CPU includes: a reception unit that receives, from among the response results of a questionnaire given to a plurality of respondents stored in a storage unit, a first condition for determining a target person image for generating a persona and a third condition for determining a comparison target person image for comparison with the target person image for generating the persona; an answer extraction unit that extracts answer results of answer items for question items for characteristic answer items that meet a second condition that the answer ratio is unique among the answer items of the questionnaire from answer items included in the questionnaire results for the respondents that meet the first condition received by the reception unit; a sentence generation unit that generates a sentence expressing a tendency of the persona that matches the first condition as a sentence using a sentence generation model configured by a large-scale language model according to the answer result extracted by the answer extraction unit, the sentence using words corresponding to the characteristic answer item; It functions as The characteristic answer item is an answer item for which the difference between the answer result of the questionnaire for respondents who meet the first condition and the answer result of the questionnaire for respondents who meet the third condition is equal to or greater than a predetermined threshold.

2. 2. The text generation device according to claim 1, wherein the characteristic answer item is an answer item for which a difference between the answer results of the questionnaire for respondents who meet the first condition and an average of the answer results of the questionnaire for respondents who meet a third condition different from the first condition is equal to or greater than a predetermined threshold.

3. The sentence generation device according to claim 1 , wherein the second condition is that the question item is manually selected in advance.

4. 2. The text generation device according to claim 1, further comprising a display unit that displays the text generated by the text generation unit and also displays question items corresponding to the characteristic answer items used in text generation by the text generation unit.

5. The text generation device according to claim 4 , wherein the display unit displays the comparison result of the question items as a graph.

6. The display unit may further include an additional reception unit that receives selection of question items to be used in the sentence generation by the sentence generation unit from among question items corresponding to the characteristic answer items during or after the display of the sentence by the display unit, the sentence generation unit regenerates a sentence using words corresponding to an answer item to the question item selected through the additional reception unit; The text generation device according to claim 4 or 5, wherein the display unit displays the regenerated text.

7. 2. The text generation device according to claim 1, further comprising an image generation unit that generates a face image of a respondent who meets the first condition from the text generated by the text generation unit.

Images