Text creation support apparatus

Abstract

To provide a text creation support apparatus which creates more appropriate sentences, based on characteristics of action targets selected from among prediction targets using machine learning results.SOLUTION: A text creation support apparatus 1 includes: a model creation module 11 which creates a model indicating a relation between first explanatory variables and objective variables based on the first explanatory variables and the objective variables of a plurality of learning targets; a scoring module 12 which calculates scores of a plurality of prediction targets by inputting second explanatory variables of the plurality of prediction targets to the model; a selection module 13 which selects a plurality of action targets out of the plurality of prediction targets, based on the scores; a first prompt creation module 14 which creates a first prompt to instruct a generative AI 3 so as to estimate characteristics of the plurality of action targets by referring to data related to each of the plurality of action targets; and a second prompt generation module 15 which creates a second prompt to instruct the generative AI so as to create a sentence in consideration of the characteristics.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a writing assistance device that supports writing using generative AI in order to more effectively take actions based on scores calculated using the results of machine learning. [Background technology]

[0002] A system has been put into practical use and is being improved, which creates a model by performing machine learning on a combination of a first explanatory variable that indicates the attributes or characteristics of the learning object and a target variable that is a known result related to the learning object, and then inputs a second explanatory variable that indicates the attributes or characteristics of the prediction object into this model to predict the results related to the prediction object. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6031165 [Patent Document 2] Patent No. 6158464 Summary of the Invention [Problem to be solved by the invention]

[0004] For example, in the field of marketing, it is now possible to predict the purchases of other customers using the attributes and purchase history data of multiple customers who are the learning targets. By taking actions such as sending information letters to customers who have a high probability of purchasing among the predicted targets, it is possible to achieve high effectiveness at low cost.

[0005] However, even with the improvement in prediction accuracy, the wording of conventional information messages was either uniform or the only way to create messages was to segment the target groups to a certain extent.

[0006] One aspect of the present invention relates to the creation of more appropriate text for each action based on the individual characteristics of the action target selected from the predicted targets using the results of machine learning, thereby enabling the effective execution of that action. [Means for solving the problem]

[0007] A writing support device according to one aspect of the present invention comprises: a model creation module that creates a model indicating a relationship between a first explanatory variable and a response variable based on the first explanatory variable and the response variable of each of a plurality of learning objects; a scoring module that inputs a second explanatory variable of each of a plurality of prediction targets into the model and calculates a score for each of the plurality of prediction targets; a selection module that selects a plurality of behavior targets from the plurality of prediction targets based on the scores; a first prompt creation module that creates a first prompt that instructs a first generation AI to estimate characteristics of each of the plurality of behavioral objects by referring to data related to each of the plurality of behavioral objects; a second prompt creation module that creates a second prompt that instructs a second generation AI to generate a sentence that takes the characteristics into consideration; Includes. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 shows a writing assistance device 1 according to the first embodiment and an external device connected to the writing assistance device 1. As shown in FIG. [Figure 2] FIG. 2 shows the functions and operations of the writing assistance device 1 according to the first embodiment. [Figure 3A] FIG. 3A shows an example of learning target data used for creating a model in a first specific example of the first embodiment. [Figure 3B] FIG. 3B shows an example in which the learning data of FIG. 3A is quantified for model creation. [Figure 3C]FIG. 3C shows an example of prediction target data used for scoring in the first specific example of the first embodiment. [Figure 3D] FIG. 3D shows an example in which the prediction target data in FIG. 3C is quantified for scoring purposes. [Figure 3E] FIG. 3E shows an example of scores calculated in the first specific example of the first embodiment. [Figure 3F] FIG. 3F shows the theoretical effect obtained by selecting an action target in the first specific example of the first embodiment. [Figure 4A] FIG. 4A shows an example of action target data used to create a first prompt in a first specific example of the first embodiment. [Figure 4B] FIG. 4B shows an example in which external data is added to the action target data shown in FIG. 4A. [Figure 4C] FIG. 4C shows an example of a first prompt generated in the first specific example of the first embodiment. [Figure 4D] FIG. 4D shows an example of a response from Production AI3 to the first prompt shown in FIG. 4C. [Figure 4E] FIG. 4E shows a further example of a first prompt. [Figure 4F] FIG. 4F shows an example of a response from Production AI3 to the first prompt shown in FIG. 4E. [Figure 4G] FIG. 4G shows an example of a second prompt generated in the first specific example of the first embodiment. [Figure 4H] FIG. 4H shows an example of a response from Generation AI3 to the second prompt shown in FIG. 4G. [Figure 5A] FIG. 5A shows an example of action target data used to create a first prompt in the second specific example of the first embodiment. [Figure 5B] FIG. 5B shows an example of a first prompt generated in the second specific example of the first embodiment. [Figure 5C] FIG. 5C shows an example of a response from Production AI3 to the first prompt shown in FIG. 5B. [Figure 5D] FIG. 5D shows a further example of a first prompt. [Figure 5E] FIG. 5E shows an example of a response from Production AI3 to the first prompt shown in FIG. 5D. [Figure 5F] FIG. 5F shows an example of a second prompt generated in the second specific example of the first embodiment. [Figure 5G] FIG. 5G shows an example of a response from Generation AI3 to the second prompt shown in FIG. 5F. [Figure 6A] FIG. 6A shows an example of action target data used to create a first prompt in the third specific example of the first embodiment. [Figure 6B] FIG. 6B shows an example of a first prompt generated in the third specific example of the first embodiment. [Figure 6C] FIG. 6C shows an example of a response from Production AI3 to the first prompt shown in FIG. 6B. [Figure 6D] FIG. 6D shows a further example of a first prompt. [Figure 6E] FIG. 6E shows an example of a response from Production AI3 to the first prompt shown in FIG. 6D. [Figure 6F] FIG. 6F shows an example of a second prompt generated in the third specific example of the first embodiment. [Figure 6G] FIG. 6G shows an example of a response from Production AI3 to the second prompt shown in FIG. 6F. [Figure 7] FIG. 7 shows the functions and operations of a writing assistance device 1e according to the second embodiment. [Figure 8A] FIG. 8A shows an example of data relating to each of the learning target and the prediction target used to create the third prompt in the specific example of the second embodiment. [Figure 8B] FIG. 8B shows an example of a third prompt generated in a specific example of the second embodiment. [Figure 8C] FIG. 8C shows an example of a response from Production AI3 to the third prompt shown in FIG. 8B. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Each embodiment described below shows an example of the present invention and does not limit the content of the present invention. Furthermore, not all of the configurations and operations described in each embodiment are necessarily essential as the configurations and operations of the present invention. Note that the same components are given the same reference numerals, and redundant explanations will be omitted.

[0010] <1. First embodiment> <1-1.Configuration> FIG. 1 shows a writing support device 1 according to the first embodiment and external devices connected to the writing support device 1. The writing support device 1 is a computer system including a CPU, memory, etc. (not shown). The writing support device 1 may be configured with a single computer or may be configured with multiple computers connected via a network. The writing support device 1 is connected to external devices such as a database 2 and a generation AI 3.

[0011] The database 2 stores data of the learning target and prediction target, as well as other data. The database 2 is not limited to data stored in a single storage, but may be data stored in multiple storages in a distributed manner. The writing assistance device 1 acquires various data from the database 2, and performs model creation, scoring, selection of action targets, and creation of prompts.

[0012] The generative AI 3 includes a large-scale language model (LLM). The large-scale language model is a language model constructed using large amounts of text data and deep learning technology, and processes tasks such as text generation, translation, question answering, text summarization, and sentiment analysis in response to prompts sent from the writing assistance device 1. The large-scale language model preferably includes an attention mechanism. The attention mechanism extracts important parts from the input and significantly contributes to improving the processing speed and accuracy of the large-scale language model.

[0013] The generation AI 3 may further include an image generation AI. The image generation AI is a system that generates images from prompts written in text, and uses, for example, a diffusion model. Note that the "image" in the following description may also be a video.

[0014] The writing support device 1 sends a prompt to the generation AI 3 and receives an output from the generation AI 3.

[0015] <1-2. Functions and operations> 2 shows the functions and operations of the writing support device 1 according to the first embodiment. The writing support device 1 includes a model creation module 11, a scoring module 12, a selection module 13, a first prompt creation module 14, and a second prompt creation module 15. These modules are implemented by loading programs into a memory included in the writing support device 1 and executing them by a CPU.

[0016] The model creation module 11 acquires the first explanatory variables and the dependent variables of each of the multiple learning objects from the database 2, and creates a model showing the relationship between the first explanatory variables and the dependent variables. A specific example of model creation will be described later with reference to Figures 3A and 3B.

[0017] The scoring module 12 acquires second explanatory variables for each of the multiple prediction targets from the database 2, inputs the second explanatory variables into the model, and calculates a score for each of the multiple prediction targets. Specific examples of scoring will be described later with reference to Figures 3C to 3E.

[0018] The selection module 13 selects a plurality of behavior targets from the plurality of predicted targets based on the scores. A specific example of behavior target selection will be described with reference to FIG. 3F.

[0019] The first prompt creation module 14 acquires data on each of the plurality of behavior targets from the database 2 and creates a first prompt that instructs the generation AI 3 to estimate the characteristics of each of the plurality of behavior targets. A specific example of creating the first prompt will be described later with reference to Figures 4A to 4F.

[0020] The second prompt generation module 15 acquires the characteristics of each of the multiple behavioral targets estimated by the generation AI 3 and generates a second prompt that instructs the generation AI 3 to generate a sentence taking these characteristics into consideration. A specific example of generating a second prompt will be described later with reference to Figures 4G and 4H.

[0021] In this application, the generation AI3 that estimates the characteristics of the target of action in response to the first prompt may be referred to as the first generation AI, and the generation AI3 that generates a sentence in response to the second prompt may be referred to as the second generation AI. The first and second generation AIs may be the same generation AI or different generation AIs.

[0022] <1-3. First concrete example> <1-3-1. Model Creation> 3A shows an example of training data used for model creation in the first specific example of the first embodiment. The training data includes various information about each of a large number of customers identified by their customer IDs, such as gender, address, membership type, and whether or not they have made a purchase.

[0023] 3B shows an example in which the learning data in FIG. 3A is quantified for model creation. To create the model, for example, gender is converted to 0 or 1, address is converted to a value between 1 and 47 depending on the prefecture, and membership type is converted to a value between 1 and 4 depending on the rank, and these are used as first explanatory variables. Also, for example, whether or not a purchase has been made is converted to 0 or 1, and this is used as the objective variable. The model creation module 11 creates a function indicating the relationship between multiple first explanatory variables and the objective variable as a model.

[0024] <1-3-2. Scoring> FIG. 3C shows an example of prediction target data used for scoring in a first specific example of the first embodiment. The prediction target is composed of customers different from the customers that constitute the learning target. The prediction target data includes information such as gender, address, and membership type for each of a large number of customers identified by a customer ID. However, the prediction target data does not include data on whether or not a purchase has been made. Whether or not a prediction target has made a purchase is unknown information. With respect to items other than whether or not a purchase has been made, the items included in the prediction target data are the same as the items included in the learning target data.

[0025] Figure 3D shows an example in which the prediction target data in Figure 3C has been quantified for scoring. As with the learning target data, gender, address, membership type, etc. are each converted into numerical values, which are used as second explanatory variables.

[0026] 3E shows an example of a score calculated in the first specific example of the first embodiment. A score for each of the multiple prediction targets is calculated by inputting the second explanatory variables of each of the multiple prediction targets into the model. This score is a value corresponding to the objective variable in the training data, for example, a value indicating the probability that each of the prediction targets will purchase a product.

[0027] <1-3-3. Selecting the target of action> FIG. 3F shows the theoretical effect obtained by selecting behavior targets in the first specific example of the first embodiment. Assume that the prediction targets are, for example, 460,000 customers. If a message is sent by direct mail to all 460,000 customers, 60,000 customers will purchase the product. Sending messages to all 460,000 customers would be extremely costly. Therefore, if a message is sent to half of the 460,000 customers (230,000), and those 230,000 customers are randomly selected (unanalyzed), 30,000 are expected to purchase the product. In other words, the number of purchasers is proportional to the number of messages sent. On the other hand, if the 230,000 customers are selected in descending order of score (using the analysis results), the number of customers who purchase the product is expected to increase, to, for example, 51,000. The ratio of the number of behavior targets to the number of prediction targets is not limited to half; even if the ratio is other ratios (except 0 and 1), better results can be obtained when using the analysis results than when not using analysis.

[0028] <1-3-4. Creating the first prompt> 4A shows an example of behavior target data used to generate a first prompt in a first specific example of the first embodiment. Fig. 4A shows only data for customer ID "bbbb," who is one of multiple behavior targets selected from prediction targets based on their scores.

[0029] The behavior target data may include items included in the prediction target data. The behavior target data may also include items not included in the prediction target data, for example, purchase data for customer ID "bbbb" may include the date, amount, and product type of a product with a donation attached that customer ID "bbbb" purchased.

[0030] Figure 4B shows an example of adding external data to the behavior target data shown in Figure 4A. The external data shown in Figure 4B is event data including events for each month of the dates included in the behavior target data shown in Figure 4A, the Business Activity Index (CI) and Business Activity Index (DI), and the Nikkei Stock Average N225. The Business Activity Index (CI) is an index that indicates the direction of economic expansion or contraction. The Business Activity Index (DI) is an index that indicates the peaks and troughs of the economy, with the 50 mark as the dividing line. Event data is extracted according to the dates included in the behavior target data from data stored in Database 2, separate from the learning target data and prediction target data.

[0031] FIG. 4C shows an example of a first prompt generated in a first specific example of the first embodiment. The first prompt includes a command statement, auxiliary information, purchase data, and event data. The command statement includes a statement requesting the generation AI 3 to infer characteristics of the target of behavior, such as age and personality. The command statement may be stored in advance as a template. The purchase data and event data are the purchase data and event data shown in FIG. 4B.

[0032] The auxiliary information may include information stored in advance as a template. The auxiliary information may also include information derived from the behavioral target data. For example, if a correlation between the purchase data of customer ID "bbbb" and trends in stock prices by industry is obtained, and there is a high positive correlation between the deposit amount of customer ID "bbbb" and stock prices in the construction industry, the auxiliary information may include an estimate that customer ID "bbbb" works in the construction industry. Furthermore, if a correlation between the purchase data of customer ID "bbbb" and trends in the market value of various assets is obtained, and there is a high positive correlation between the deposit amount of customer ID "bbbb" and the appraised value of a real estate investment trust, the auxiliary information may include an estimate that the assets held by customer ID "bbbb" are mainly real estate.

[0033] Figure 4D shows an example of a response from the generation AI 3 to the first prompt shown in Figure 4C. The generation AI 3 outputs the results of its predictions of the target's age, personality, and other characteristics. The prediction accuracy can be improved by incorporating external data.

[0034] 4E shows a further example of the first prompt. The first prompt includes a command requesting the generation AI3 to guess the advantages and disadvantages of the target of the action.

[0035] Figure 4F shows an example of a response from the generation AI3 to the first prompt shown in Figure 4E. The generation AI3 outputs the results of inferring characteristics, including advantages and disadvantages, of the action target.

[0036] <1-3-5. Creating the second prompt> FIG. 4G shows an example of a second prompt generated in the first specific example of the first embodiment. The second prompt includes a command requesting the user to write a thank-you letter for the product purchase, taking into account the inferred advantages and disadvantages, so as to encourage the user to continue purchasing the product. The command also requests that the user not mention the user's place of employment or assets, which are uncertain information. The command may be stored in advance as a template. Note that when the first prompt and the second prompt are sent to the same generation AI 3, the inference result of the feature generated in response to the first prompt is assumed to be inherited within the generation AI 3. When the first prompt and the second prompt are sent to different generation AIs 3, the inference result of the feature generated in response to the first prompt must be included in the second prompt.

[0037] Figure 4H shows an example of a response from the generation AI3 to the second prompt shown in Figure 4G. The generation AI3 outputs a draft letter to the target of action. By generating a letter based on the predicted strengths and weaknesses, the customer who receives the letter will feel that the sender "understands me," which will increase their liking for the sender and make them more likely to purchase the product.

[0038] The writing support device 1 may further request the generation AI 3 to generate an image to accompany the letter. As an example, when the image generation AI constituting the generation AI 3 operates in response to a prompt in English, the writing support device 1 outputs a prompt to the large-scale language model constituting the generation AI 3, saying, "Please describe in English an image that would be effective to accompany the letter." Then, based on the response from the large-scale language model, the writing support device 1 causes the image generation AI to generate an image. The writing support device 1 can obtain the HTML document linked to the image file from the large-scale language model by generating a prompt to the large-scale language model to generate an HTML document of the letter linked to the image file from the file name of the image and the draft of the letter in Figure 4H.

[0039] <1-3-6. Additional prompt examples> A further prompt may be added containing the following statement: The statement may be stored in advance as a template.

[0040] (1) Before instructing the AI ​​to infer characteristics, an instruction to list the perspectives is provided. For example, before the first prompt shown in Figure 4C, a prompt including the command "List the important perspectives that should be considered when inferring a persona in marketing" is given to the AI ​​generation 3. Then, the command included in the first prompt shown in Figure 4C is changed to "Based on the listed perspectives, infer Mr. A's age, personality, family structure, and lifestyle," and given to the AI ​​generation 3. By having the AI ​​infer based on the listed perspectives, it is possible to reduce oversights and improve the level of thinking.

[0041] (2) The prompts are structured to prompt the AI ​​to consider the characteristics several times and choose the best answer. For example, after the first prompt shown in Figure 4C, a prompt including the command "Create two more personas for A based on the listed viewpoints" is given to the AI ​​generation 3. Then, a prompt including the command "Compare or improve these personas and present the persona that seems most effective" is given to the AI ​​generation 3. This allows the AI ​​generation 3 to discover other aspects of the target of action and further improve accuracy.

[0042] (3) Add a step to refine the generated text in light of the original data. For example, after the second prompt shown in Figure 4G, the generation AI 3 is given a prompt containing the command, "Refer to the supplementary information, the member's purchase data, and the event data for the relevant month, verify whether there are any problems or anything noteworthy, and revise this letter." By having the AI ​​3 review the original data that served as the basis for its inference, it is possible to further improve accuracy.

[0043] <1-4. Second specific example> The second example shows a different example of prompt creation, with model creation, scoring, and action target selection being similar to the first example.

[0044] <1-4-1. Creating the first prompt> FIG. 5A shows an example of behavior target data used to generate a first prompt in a second specific example of the first embodiment. FIG. 5A shows only data for customer ID "cccc," who is one of multiple behavior targets selected from among prediction targets based on their scores. The behavior target data includes, for example, attribute information and purchase history for customer ID "cccc" of a company that manufactures and sells confectionery and nutritional supplements. In the second specific example, it is not necessary to import external data as in FIG. 4B.

[0045] FIG. 5B shows an example of a first prompt generated in a second specific example of the first embodiment. The first prompt includes a command statement, attribute information, and a purchase history. The command statement includes a statement requesting the generation AI 3 to infer characteristics such as the persona and personality of the target of the behavior. The attribute information and purchase history are the attribute information and purchase history shown in FIG. 5A.

[0046] Figure 5C shows an example of a response from the generation AI 3 to the first prompt shown in Figure 5B. The generation AI 3 outputs the results of inferring characteristics such as the persona and personality of the target of the action.

[0047] 5D shows a further example of the first prompt, which includes a command requesting the generation AI3 to infer purchasing motives and product preferences as characteristics of the behavioral target.

[0048] Figure 5E shows an example of a response from the generation AI3 to the first prompt shown in Figure 5D. The generation AI3 outputs the results of inferring characteristics including the purchasing motive and product preference of the behavior target.

[0049] <1-4-2. Creating the second prompt> 5F shows an example of a second prompt generated in the second specific example of the first embodiment. The second prompt includes an instruction requesting the target to write a thank-you letter for the product purchase, based on the inferred motives and preferences, to encourage the target to purchase the product. The instruction also requests that personal information, such as attribute information, not be mentioned.

[0050] Figure 5G shows an example of a response from the generation AI3 to the second prompt shown in Figure 5F. The generation AI3 outputs a draft letter to the target of the action. By generating the text based on the inferred motives and preferences, the customer who receives the letter will be reminded of past purchases, such as "That reminds me, that was delicious," and will likely purchase again.

[0051] The writing support device 1 may further request the generation AI 3 to generate an image to be attached to the letter.

[0052] <1-5.Third Specific Example> The third example shows a different example of prompt creation, with model creation, scoring, and action target selection being similar to the first example.

[0053] <1-5-1. Creating the first prompt> FIG. 6A shows an example of behavior target data used to generate a first prompt in a third specific example of the first embodiment. FIG. 6A shows only data for customer ID "dddd," who is one of multiple behavior targets selected from prediction targets based on their scores. The behavior target data may include, for example, a grocery store shopping record for customer ID "dddd," as well as household composition data. In the third specific example, external data such as that shown in FIG. 4B need not be imported.

[0054] 6B shows an example of a first prompt generated in a third specific example of the first embodiment. The first prompt includes an instruction statement, a shopping record, and a designation of an output format. The instruction statement includes a statement requesting the generation AI3 to predict the dishes for each date in the household of the behavior target. The shopping record is the shopping record shown in FIG. 6A.

[0055] Figure 6C shows an example of a response from Generation AI3 to the first prompt shown in Figure 6B, which outputs a guessed dish and a list of ingredients extracted from the shopping record.

[0056] 6D shows a further example of the first prompt, which includes a statement requesting the generation AI3 to suggest an additional ingredient to add to the inferred dish.

[0057] Figure 6E shows an example of a response from generation AI3 to the first prompt shown in Figure 6D. Generation AI3 outputs a guessed dish, ingredients extracted from the shopping record, and suggested ingredients.

[0058] <1-5-2. Creating the second prompt> 6F shows an example of a second prompt generated in the third specific example of the first embodiment. The second prompt includes an instruction requesting the target to write a recommended menu for the target's next shopping trip based on the suggested ingredients and a guide message promoting the ingredients.

[0059] Figure 6G shows an example of a response from the generation AI3 to the second prompt shown in Figure 6F. The generation AI3 outputs a draft of an information message for the target of action. By generating a message based on the inferred dish and suggested ingredients, the customer who receives the letter will think, "So I can make that dish with ingredients that I and my family like," or "It would be even more delicious if I added one more dish," increasing the likelihood of further purchases.

[0060] The writing support device 1 may further request the generation AI 3 to generate an image to be attached to the letter.

[0061] <1-6. Other examples> In the first to third specific examples, the learning subject and the prediction subject are customers, but the present application is not limited to this. The learning subject and the prediction subject may be, for example, a health checkup recipient. By using "whether or not the learning subject's health checkup recipient has contracted a specific disease" as the objective variable and scoring the prediction subject's health checkup recipient, it is possible to predict people who are likely to contract a specific disease in the future and set them as behavior targets. By having the generation AI 3 estimate the characteristics of the behavior target and having the generation AI 3 generate a vaccination guide message based on those characteristics, it is possible to obtain a more effective guide message that can be used to prevent disease.

[0062] The learning and prediction targets can also be employees. By using "whether or not the learning target employee has been promoted" as the objective variable and scoring the prediction target employee, it is possible to select the target of action according to the likelihood of promotion. By having the generation AI 3 estimate the characteristics of the target employee and then having the generation AI 3 generate advice and training information based on those characteristics, it is possible to obtain advice and training information that is appropriate for that employee.

[0063] When the learning target and prediction target are humans, the following features should be estimated: (1) Psychological motivations (purpose, motivation, needs, worries, background, emotions) (2) Preferences (hobbies, tastes, likes and dislikes, preferred activities, styles, trends) (3) Thoughts and ideas (values, outlook on life, outlook on society, outlook on the world, outlook on family, philosophy of life) (4) Propensities (strengths, weaknesses, personality traits, behavior patterns, lifestyle) (5) Demographics (age, gender, occupation, income, education level, digital skills) (6) Cultural and regional characteristics (regionality, language, cultural values)

[0064] The learning and prediction targets are not limited to people; they can also be objects, such as products. By using "whether the product was a hit" as the objective variable for the learning target product and scoring the prediction target product, it is possible to predict products that are likely to be hits in the future and use them as action targets. By having the generation AI 3 estimate the characteristics of the target product and generate advertising copy based on those characteristics, it is possible to obtain advertising copy for that product that is highly effective.

[0065] The learning target and prediction target may be, for example, a machine. By using "whether or not the machine to be learned has broken down" as the objective variable and scoring the machine to be predicted, it is possible to predict which machines are likely to break down in the future and set them as the target of action. By having the generation AI 3 estimate the characteristics of the machine to be acted upon and then having the generation AI 3 generate a maintenance guide message for the machine owner based on those characteristics, it is possible to obtain a more appealing guide message.

[0066] When the learning target and prediction target are objects, the features to be estimated include the following: (1) Cost performance, durability, design, size, functionality, and convenience (2) Customer satisfaction, brand reliability, history, and safety (3) Exclusivity, scarcity, and eco-friendliness

[0067] <1-7.Effects> According to the first embodiment, the writing support device 1 a model creation module 11 that creates a model indicating a relationship between a first explanatory variable and a response variable based on the first explanatory variable and the response variable of each of a plurality of learning objects; a scoring module 12 that inputs a second explanatory variable of each of the plurality of prediction targets into a model and calculates a score for each of the plurality of prediction targets; a selection module 13 for selecting a plurality of action targets from the plurality of prediction targets based on the scores; a first prompt creation module 14 that creates a first prompt that instructs the generation AI 3 to infer characteristics of each of the plurality of behavioral objects by referring to data relating to each of the plurality of behavioral objects; A second prompt generation module 15 that generates a second prompt that instructs the generation AI 3 to generate a sentence that takes the feature into consideration; Includes.

[0068] This not only improves the probability of achieving the goal by selecting an action target based on the score, but also allows for the generation of appealing sentences by considering the characteristics of each action target. Furthermore, rather than generating sentences directly from data about each action target, the process is divided into a step of estimating characteristics and a step of creating sentences that take characteristics into account. This clarifies the estimations that were used to generate sentences, and individual verification is also possible. Furthermore, the characteristics to be estimated (strengths and weaknesses, purchasing motivations and product preferences, dishes and suggested ingredients, etc.) can be specified in the first prompt, enabling the direction of the sentences to be generated. In this way, the characteristics of the action target (people or objects) can be captured in detail and from multiple angles, allowing for the generation of sentences that are appropriate for each action target.

[0069] 2. Second embodiment <2-1. Functions and operations> 7 shows the functions and operations of a writing support device 1e according to the second embodiment. In addition to the various modules included in the writing support device 1 described with reference to FIG. 2, the writing support device 1e further includes a third prompt creation module 16 and a variable creation module 17. These modules are implemented by loading programs into a memory included in the writing support device 1e and executing them by a CPU.

[0070] The third prompt creation module 16 acquires data on each of the multiple learning targets and prediction targets from the database 2 and creates a third prompt that instructs the generation AI 3 to analyze the data on each of the multiple learning targets and prediction targets. The data on each of the multiple learning targets and prediction targets includes text data. Systems for classifying text data have been known for some time, but classification alone has not been able to fully utilize the information contained in the text data. The writing support device 1e can perform scoring with high accuracy by having the generation AI 3 analyze the text data. A specific example of creating a third prompt will be described below with reference to Figures 8A to 8C.

[0071] In this application, the generated AI3 that performs analysis in response to the third prompt may be referred to as the third generated AI. The third generated AI may be the same as either or both of the first and second generated AIs, or may be a separate generated AI.

[0072] The variable creation module 17 acquires the analysis results of the data on the learning target and the prediction target from the generation AI 3, and creates first and second explanatory variables. The created explanatory variables are stored in the database 2. The first explanatory variable is used for model creation, and the second explanatory variable is used for scoring, and the subsequent processing is the same as in the first embodiment.

[0073] <2-2. Specific examples> <2-2-1. Creating the third prompt> FIG. 8A shows an example of data relating to a learning target and a prediction target used to create a third prompt in a specific example of the second embodiment. In this specific example, the learning target and the prediction target are stores. Of the multiple stores included in FIG. 8A, those whose store names are unknown are unlikely to become prediction targets, but they can still be used as learning targets. For stores that are learning targets, a separate objective variable can be prepared, allowing the model creation module 11 to create a model. The objective variable is, for example, whether or not a problem has occurred at the store.

[0074] FIG. 8A shows records of customer inquiries about each store. The time period is the time period when the inquiry was received by phone, the business type is the business type of the store, and the category is the category of the inquiry. The time period, business type, and category data may be assigned automatically or manually. The text conversion of the inquiry voice may also be performed automatically or manually.

[0075] FIG. 8B shows an example of a third prompt generated in a specific example of the second embodiment. The third prompt includes a command statement, an analysis condition, an input statement, and a designation of an output format. The command statement includes a statement requesting the generation AI 3 to analyze customer sentiment based on the text data of the query voice. The command statement may be stored in advance as a template. The input statement is the text data of the query voice shown in FIG. 8A.

[0076] Figure 8C shows an example of responses from generation AI3 to the third prompt shown in Figure 8B for multiple stores. Generation AI3 outputs positive, negative, and neutral scores.

[0077] <2-2-2. Creating variables> The variable creation module 17 acquires the scores shown in FIG. 8C and creates first and second explanatory variables. The first and second explanatory variables may use the scores shown in FIG. 8C as they are, or may be further processed. As an example of a processed explanatory variable, a value obtained by subtracting a negative score from a positive score may be used. As another example of a processed explanatory variable, a principal component score obtained by performing principal component analysis using values ​​obtained by quantifying the time period, business type, classification, etc. shown in FIG. 8A and the scores shown in FIG. 8C may be used.

[0078] By using the first and second explanatory variables obtained in this way, for example, it is possible to score the possibility of trouble occurring with customers at the store that is the prediction target, and send a warning message to the behavior target with the highest score before such trouble occurs. To create this warning message, the generation AI 3 is made to estimate the characteristics of the store that is the behavior target, and by having the generation AI 3 generate a sentence based on those characteristics, it is possible to obtain an effective warning message for that store.

[0079] While the example described here is one in which the generation AI 3 is provided with text data and used to perform sentiment analysis, the present application is not limited to this. The generation AI 3 may be configured to extract keywords from the text data, and the presence or absence of each extracted keyword may be used as an explanatory variable (0 or 1).

[0080] Although the text data provided to the generation AI 3 is a text version of a voice inquiry, the present invention is not limited to this. For example, the text data may be a questionnaire entry.

[0081] The writing support device 1e may further request the generation AI3 to generate an image to be added to the warning message.

[0082] <2-3. Effects> According to the second embodiment, the writing support device 1e: a third prompt creation module 16 that creates a third prompt that instructs the third generation AI to output an analysis result by referring to text data related to each of the plurality of learning targets and the plurality of prediction targets; a variable creation module 17 that creates a first explanatory variable and a second explanatory variable based on the analysis result; Further includes:

[0083] This allows for improved score accuracy by analyzing text data and creating explanatory variables. Even if the text data is incomplete, the large-scale language model can supplement the information by providing the most probable interpretation, preventing a decline in score accuracy. [Explanation of symbols]

[0084] 1, 1e...Text creation support device, 2...Database, 3...Generative AI, 11...Model creation module, 12...Scoring module, 13...Selection module, 14...First prompt creation module, 15...Second prompt creation module, 16...Third prompt creation module, 17...Variable creation module

Claims

1. a model creation module that creates a model indicating a relationship between a first explanatory variable and a response variable based on the first explanatory variable and the response variable of each of a plurality of learning objects; a scoring module that inputs a second explanatory variable of each of a plurality of prediction targets into the model and calculates a score for each of the plurality of prediction targets; a selection module that selects a plurality of behavior targets from the plurality of prediction targets based on the scores; a first prompt creation module that creates a first prompt that instructs a first generation AI to estimate characteristics of each of the plurality of behavioral objects by referring to data related to each of the plurality of behavioral objects; a second prompt creation module that creates a second prompt that instructs a second generation AI to generate a sentence that takes the characteristics into consideration; A writing support device including:

2. 2. The writing support device according to claim 1, a third prompt creation module that creates a third prompt that instructs a third generation AI to output an analysis result by referring to text data related to each of the plurality of learning targets and the plurality of prediction targets; a variable creation module that creates the first explanatory variable and the second explanatory variable based on the analysis result; The writing support device further includes:

Images