Recipe suggestion device and method

Abstract

We provide technology that enables the suggestion of food and beverage recipes that meet various user requirements, such as taste. [Solution] The recipe suggestion device includes means for generating recipe suggestion information that includes at least recipe information and evaluation information for suggested recipes corresponding to user request information, using a trained model including a large-scale language model, and means for displaying the recipe suggestion information on a display unit. For generating the recipe suggestion information, the means extracts information corresponding to the user request information from a recipe database that stores the recipe name, recipe information, and evaluation information for each of a plurality of recipes. One or more evaluation pieces of information stored in the recipe database include sensory change time series data that quantitatively shows the changes in sensory perception over time when consuming food and beverages made with the target recipe, and evaluation expressions related to food and beverages made with the target recipe (for example, multifaceted evaluation expressions such as sensory, affective, and physiological evaluations).

Description

Technical Field

[0001] The present invention relates to a technology for proposing food and beverage recipes.

Background Art

[0002] In recent years, generative AI (Artificial Intelligence) technology via large language models such as GPT (Generative Pre-trained Transformer) has been rapidly advancing and spreading, and various recipe proposal methods using such generative AI technology have been devised. For example, Patent Document 1 below discloses a system including means for communicating the remaining contents of a refrigerator on a messenger app group chat, means for receiving requests such as the preferences of participating members and diet requirements, and means for a generative AI to propose a menu and recipes based on this information.

[0003] In addition, various technologies for quantifying the deliciousness of food and beverages have been proposed. Patent Documents 2 and 3 below disclose methods for evaluating the temporal variation of sensations during eating. Specifically, Patent Document 2 discloses a method for acquiring time-series data regarding the change in intensity of sensations or emotions during the ingestion of food and beverages, acquiring data regarding the timing of swallowing the food and beverages, and calculating analysis parameters regarding the change in intensity of sensations or emotions based on the acquired time-series data and the data regarding the timing of swallowing. Patent Document 3 discloses a method for acquiring time-series data regarding the change in intensity of sensations or emotions during the ingestion of food and beverages for each of a plurality of different sensations or emotions, and calculating an index regarding the cross-correlation between the plurality of time-series data based on the plurality of time-series data acquired for the plurality of different sensations or emotions.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

[0005] Although various recipe suggestion methods using generative AI technology have been presented, such as the method disclosed in Patent Document 1 mentioned above, the reality is that the suggested recipes do not always match the user's needs. For example, when the user's requests are vague or when the amount of information available is small, it is difficult to suggest an appropriate recipe.

[0006] This invention has been made in view of these circumstances and provides a technology that enables the proposal of food and beverage recipes that meet user needs. [Means for solving the problem]

[0007] According to the present invention, a request acquisition means for acquiring user request information regarding food and beverages, and a trained model including a large-scale language model A recipe database that stores the recipe name, recipe information, and evaluation information for each of multiple recipes. The system comprises a generation means for generating recipe suggestion information that includes at least recipe information and evaluation information related to suggested recipes in accordance with the user request information, and a display processing means for displaying the recipe suggestion information on a display unit. Huh, before The one or more evaluation pieces of information stored in the recipe database are time-series data of sensory changes that quantitatively show the changes in sensory perception over time when consuming food and beverages prepared using the target recipe. Includes , The generation means extracts the time-series data of sensory changes corresponding to the user request information from the recipe database, and obtains the recipe suggestion information by providing the large-scale language model with prompts generated using at least the extracted time-series data of sensory changes. A recipe suggestion device is provided.

[0008] Furthermore, according to the present invention, a recipe suggestion method performed by a computer, wherein the computer includes the steps of acquiring user request information regarding food and beverages, and a trained model including a large-scale language model A recipe database that stores the recipe name, recipe information, and evaluation information for each of multiple recipes.Using the above, the following steps are performed: a generation step that generates recipe suggestion information which includes at least recipe information and evaluation information for suggested recipes that correspond to the user request information, and a display processing step that displays the recipe suggestion information on the display unit. And, before The one or more evaluation pieces of information stored in the recipe database are time-series data of sensory changes that quantitatively show the changes in sensory perception over time when consuming food and beverages prepared using the target recipe. Includes , In the generation step, the time-series data of sensory changes corresponding to the user request information is extracted from the recipe database, and the recipe suggestion information is obtained by providing the large-scale language model with prompts generated using at least the extracted time-series data of sensory changes. Recipe suggestion methods are provided.

[0009] Furthermore, according to the present invention, a computer program stored in one or more memories and executed by one or more processors can also be provided, which causes one or more computers equipped with the one or more memories and the one or more processors to execute the recipe suggestion method, and a recording medium on which the computer program is stored. [Effects of the Invention]

[0010] According to the present invention, it is possible to provide a technology that enables the suggestion of food and beverage recipes that meet user needs. [Brief explanation of the drawing]

[0011] [Figure 1] This diagram conceptually shows an example of the hardware configuration of the recipe suggestion device (the proposed device) according to this embodiment. [Figure 2] This diagram conceptually shows an example of the software configuration of the recipe suggestion device (the proposed device) according to this embodiment. [Figure 3] This diagram conceptually illustrates an example of updating time-series data on sensory changes in a recipe database (DB). [Figure 4] This diagram shows the correspondence between feedback information, recipe information, and additional removal-related cooking steps. [Figure 5] This is a flowchart showing the recipe proposal method according to this embodiment (the proposed method). [Modes for carrying out the invention]

[0012] Hereinafter, embodiments of the present invention (hereinafter referred to as "the present embodiments") will be described. Note that the embodiments described below are examples, and the present invention is not limited to the configurations of the following embodiments.

[0013] First, an overview of the recipe proposal device and the recipe proposal method according to the present embodiments will be described. The recipe proposal device according to the present embodiments includes at least a desire acquisition means for acquiring user desire information regarding food and drink, a generation means for generating recipe proposal information including at least recipe information and evaluation information regarding a proposed recipe according to the user desire information using a learned model including a large language model, and a display processing means for displaying the recipe proposal information on a display unit.

[0014] The "user desire information regarding food and drink" acquired by the desire acquisition means is information indicating the desires of the user regarding food and drink that the user wants to eat, make, or provide. "Food and drink" is used in a broad sense including not only food but also drinks such as soup, stew, broth, cocktail, juice, etc. The "user desire information" may include information indicating the user's desires regarding the recipe of food and drink such as ingredients, cooking methods, dish names, etc., or may include information indicating the sensory desires of food and drink such as taste, aroma, texture, etc. regarding the deliciousness of food and drink, or may include information indicating the emotional or physiological desires regarding food and drink such as relaxing, filling, and having a good satiety, or may include information indicating other desires such as purpose, scene, image, health information, constitution, diet history, etc., or may be a combination thereof. For example, character string data such as "Please teach me a dish that is suitable for breakfast, refreshing, has a gentle taste, and has a good satiety" is acquired as the user desire information.

[0015] The method for obtaining user demand information by the demand acquisition means is not limited in any way. For example, the demand acquisition means may generate the user demand information based on the information input by the user via an input screen, or the user demand information itself may be input by the user. Also, in this embodiment, the data format of the user demand information to be obtained is a character string data in a description format consisting of character data such as a sentence format, a bullet point format, or a set of words, but the data format is arbitrary.

[0016] The generation means uses a trained model including a large language model (LLM (Large Language Models)) to generate recipe proposal information corresponding to the user demand information obtained by the demand acquisition means. The "trained model" here may be any machine learning model including an LLM, and may be composed only of an LLM, or may include other machine learning models in addition to the LLM. The data structure, learning algorithm, etc. of the "trained model" are not limited. An LLM is a trained natural language model constructed using a large-scale text dataset and deep learning technology, enabling the recognition and generation of natural language. A large language model is a type of generative AI, and specific examples include "BERT" and "GPT-4". Other machine learning models may include other language models called small language models (SLMs), may include regression equations obtained by regression analysis, or may include neural network models obtained by principal component analysis, deep learning, etc. For example, these machine learning models are realized by a combination of a computer program and parameters, a combination of a plurality of functions and parameters, etc. The trained model may be held by the recipe proposal device itself, or may be held by another computer that the recipe proposal device can access via communication.

[0017] The generated recipe suggestion information will include at least recipe information and evaluation information related to the suggested recipes that meet the user's request information. "Recipe information" here includes at least the cooking method and the information on the ingredients required for the proposed recipe. The preparations listed in the recipe information may include all items necessary for the suggested recipe, such as ingredients, seasonings, cooking utensils, and cooking equipment. "Evaluation information" refers to information that indicates the evaluation of food and beverages created using the proposed recipe. The evaluation information only needs to show the evaluation of one or more sensory evaluation items such as taste (sweetness, sourness, saltiness, umami, bitterness, astringency, spiciness, richness, etc.), aroma (flavor), texture (touch), temperature, sound (chewing sounds, swallowing sounds, etc.) (auditory), and appearance (sight). In addition, the evaluation information may include information showing the evaluation over time in relation to changes in sensory perception over time during consumption, as an example of the evaluation of sensory evaluation items. Furthermore, in addition to sensory evaluations, the evaluation information may also show evaluations of sensory or physiological evaluation items related to the food and beverage prepared using the proposed recipe, such as comforting, hearty, and filling. In other words, the evaluation information may show evaluations of limited evaluation items from among multiple types of sensory evaluation items, sensory evaluation items, and physiological evaluation items, or it may show evaluations of multifaceted evaluation items including one or more sensory evaluation items, sensory evaluation items, and physiological evaluation items.

[0018] The generated recipe suggestion information only needs to include recipe information and evaluation information related to the suggested recipe, as described above, and there are no restrictions on the format of the information. The recipe information and evaluation information included in the recipe suggestion information may be formed solely from string data, or it may be formed from string data plus one or more types of data such as graphs, charts, images, videos, audio, and animations. In addition, the recipe suggestion information may include any other information related to the suggested recipe, such as health information and cultural information.

[0019] The generation means may input user request information or derived information obtained from such information into the trained model, and the information output from the trained model may be used as recipe suggestion information as is, or the recipe suggestion information may be generated by applying some kind of processing to the information output from the trained model. In this embodiment, the generation means extracts information corresponding to the user request information from a recipe database that stores the recipe name, recipe information, and evaluation information for each of a plurality of recipes, in order to generate recipe suggestion information. This extracted information may be input into the trained model as at least part of the derived information described above, or the derived information described above may be generated from this extracted information, or at least part of this extracted information may be included as is in the recipe suggestion information.

[0020] Ideally, the recipe database should be built to include recipe data not used in LLM learning, as well as the latest recipe data, so as to supplement the knowledge that LLMs lack. However, its structure and implementation are not restricted. For example, if the recipe database is built as a vector database, the generation method can be enhanced by using Retrieval Augmented Generation (RAG) technology to retrieve necessary information from the recipe database, thereby expanding and strengthening the input prompts for the LLM.

[0021] The definitions of the meanings of the recipe information and evaluation information stored in the recipe database are as described above. However, in this embodiment, one or more evaluation information stored in the recipe database includes sensory change time series data that quantitatively shows the changes in sensory perception over time when consuming food or beverage prepared using the target recipe, and evaluation expressions related to the food or beverage prepared using that target recipe.

[0022] The "sensory change time series data" stored in the recipe database is preferably data quantified by sensory evaluation or biometric measurements (electromyography, chewing sounds, swallowing sounds, etc.) using measuring instruments at the time of actual consumption of each food and beverage. However, this sensory change time series data can also be estimated using a trained model, as described later. Furthermore, the time-series data of sensory changes only needs to be quantified with respect to one or more of the sensory evaluation items mentioned above. For example, it may be a set of time-series data quantified with respect to each of the four evaluation items: taste, aroma, texture, and chewing sound, or it may be a set of time-series data quantified with respect to each of the five evaluation items of the five basic tastes.

[0023] The "evaluation terms" stored in the recipe database are words that describe what was felt when consuming food or drink made using the recipe in question. These can be sensory expressions such as "it has a light / intense taste," "it has a solid flavor," "it has a light taste," "it leaves a lingering aftertaste," or "it has a clean finish," or they can be emotional or physiological expressions such as "it feels refreshing," "it feels gentle," "it feels nostalgic," "it leaves you feeling full," "it's refreshing," or "it energizes you." Thus, it is preferable that the evaluation expression is an evaluation expression for multifaceted evaluation items (multifaceted evaluation expression) that includes at least sensory expressions, emotional expressions, and physiological expressions, but it may also be an evaluation expression for limited evaluation items among sensory expressions, emotional expressions, and physiological expressions. Furthermore, since the evaluation expression is also associated with the above-mentioned time-series data of sensory changes, it is preferable that it includes words that express changes in sensory perception over time, such as "the aroma is amazing the moment it enters the mouth," "it goes down smoothly," and "it has a lingering umami flavor."

[0024] The display processing means causes the recipe suggestion information generated by the generation means in this manner to be displayed on the display unit. The display unit may be the recipe suggestion device itself according to this embodiment, or it may be a computer that can communicate with the recipe suggestion device. Furthermore, the display processing means only needs to enable the display unit to display recipe suggestion information by some means, and there are no limitations on the display format or method of the recipe suggestion information. For example, sending recipe suggestion information to another computer using a communication medium such as email or electronic message is one example of a method by which the display processing means causes the recipe suggestion information to be displayed on the display unit. By sending the recipe suggestion information using the communication medium, the recipe suggestion information becomes displayable through the functions on that other computer (email function, message viewing function, etc.) and user operations on that computer.

[0025] The recipe suggestion method according to this embodiment is performed by one or more computers, such as the recipe suggestion device described above. In this recipe suggestion method, the computer performs the following steps: acquiring user request information regarding food and beverages; generating recipe suggestion information, which includes at least recipe information and evaluation information for suggested recipes corresponding to the user request information, using a trained model including a large-scale language model; and display processing, which displays the recipe suggestion information on a display unit. The display unit that displays recipe suggestion information in the display processing step may be provided by one or more computers that execute the recipe suggestion method according to this embodiment, or it may be provided by other computers that can communicate with said computers. In the generation process, for the purpose of generating the recipe suggestion information, information corresponding to the user request is extracted from a recipe database that stores the recipe name, recipe information, and evaluation information for each of the multiple recipes. Furthermore, as described above, one or more evaluation pieces of information stored in the recipe database include time-series data of sensory changes that quantitatively show the changes in sensory perception over time when consuming the food or beverage created by the target recipe, and evaluation expressions related to the food or beverage created by the target recipe.

[0026] In this embodiment, when generating and presenting recipe suggestion information for a specific recipe that corresponds to user requests using a trained model including an LLM, a recipe database containing sensory change time series data and evaluation expression words is referenced as evaluation information for each recipe, and information corresponding to the user requests is extracted from that recipe database. This allows for the extraction of time-series data of sensory changes as evaluation information for recipes corresponding to user requests, based on evaluation terms corresponding to user requests. Recipe suggestion information can then be generated using this time-series data of sensory changes, enabling the presentation of recipe suggestion information that includes detailed evaluation information for recipes that accurately match the user's requests.

[0027] Hereafter, the recipe suggestion device (hereinafter sometimes abbreviated as "the proposed device") and the recipe suggestion method (hereinafter sometimes abbreviated as "the proposed method") according to this embodiment will be described in detail.

[0028] <Recipe suggestion device> Figure 1 is a conceptual diagram showing an example of the hardware configuration of the recipe suggestion device (proposed device) 10 according to this embodiment. The proposed device 10 is a so-called computer and includes a CPU (Central Processing Unit) 1, memory 2, input / output interface (I / F) 3, communication unit 4, etc. CPU1 refers to a so-called processor, and in addition to general CPUs, it may also include application-specific integrated circuits (ASICs), DSPs (Digital Signal Processors), GPUs (Graphics Processing Units), etc. Memory 2 includes RAM (Random Access Memory), ROM (Read Only Memory), and auxiliary storage devices (such as hard disks).

[0029] The input / output interface 3 can be connected to user interface devices such as the display device 5 and the input device 6. The display device 5 is a device that displays a screen corresponding to drawing data processed by the CPU 1, such as an LCD (Liquid Crystal Display) or CRT (Cathode Ray Tube) display. The input device 6 is a device that accepts user input such as a keyboard or mouse. The display device 5 and the input device 6 may be integrated and implemented as a touch panel. The communication unit 4 communicates with other computers via the communication network 11 and exchanges signals with other devices such as printers. Portable recording media may also be connected to the communication unit 4. In this embodiment, the communication unit 4 is connected to a user terminal 15 or the like via the communication network 11 so as to be able to communicate.

[0030] The hardware configuration of the proposed device 10 is not limited to the example in Figure 1. The proposed device 10 may include other hardware elements not shown. The number of each hardware element is also not limited to the example in Figure 1. For example, the proposed device 10 may have multiple CPUs 1. Furthermore, the proposed device 10 may be implemented by multiple computers consisting of multiple enclosures.

[0031] The communication network 11 consists of one or more of the following: a public network such as the Internet, a WAN (Wide Area Network), a LAN (Local Area Network), a wireless communication network, etc. However, in this embodiment, the communication configuration between the proposed device 10 and the user terminal 15 is not limited.

[0032] The user terminal 15 is a so-called computer, and may be a stationary computer or a portable computer. The user terminal 15 can access the proposed device 10 and output the information provided by the proposed device 10. As long as such information output is possible, the hardware and software configuration of the user terminal 15 is not limited. The user terminal 15 displays the screen that serves as the user interface for the proposed device 10. The output format of the user terminal 15 will be described later.

[0033] The proposed device 10 generates and outputs recipe suggestion information using the software configuration shown in Figure 2. Figure 2 is a conceptual diagram showing an example of the software configuration of the recipe suggestion device (proposed device) 10 according to this embodiment. As shown in Figure 2, the proposed device 10 includes a request acquisition unit 21, a generation unit 22, an AI model 23, a recipe database (DB) 24, a display processing unit 25, a feedback processing unit 26, an update processing unit 27, and the like. Each of these processing modules is realized, for example, by the execution of a computer program stored in memory 2 by the CPU 1. This computer program may be installed via input / output I / F 3 or a communication unit 4 from a portable recording medium such as a CD (Compact Disc) or memory card, or from another computer on a network, and stored in memory 2.

[0034] The request acquisition unit 21 corresponds to the request acquisition means described above and acquires user request information regarding food and beverages. The user request information to be acquired and the method of acquiring it are as described above with respect to the request acquisition means. For example, the request acquisition unit 21 may generate user request information based on information entered by the user via an input screen displayed on the user terminal 15 via communication, or the user request information itself may be entered by the user through operation of the user terminal 15. Thus, the method of acquiring information by the request acquisition unit 21 is not limited in any way.

[0035] To improve the degree to which the suggested recipes match user requests, it is desirable to obtain as much detailed user request information as possible. Therefore, in this embodiment, the request acquisition unit 21 includes a chat means for exchanging messages with the user terminal 15. This chat mechanism, like the request acquisition unit 21, is implemented as a software component (chat processing unit) and conducts interactive chat with the user terminal 15 via communication. The interactive chat proceeds in a conversational format, with the user terminal 15 exchanging text or voice messages. Since interactive chat can be implemented using various existing methods, such as those employing specific LLMs, a detailed explanation is omitted here.

[0036] For example, the chat processing unit (chat method) conducts interactive chat to acquire as many expressions as possible that indicate the user's preferences regarding food and beverages. Examples of expressions that indicate the user's preferences include expressions that indicate recipe information such as ingredients and cooking methods, expressions that indicate mood, feeling, or purpose ("light," "for breakfast," "for lunch," "quick," "for celebrations," etc.), and expressions that indicate sensory, emotional, or physiological evaluations ("spicy," "hearty," "filling," etc.). As described above, when user request information containing evaluation expressions stored in the recipe database in association with time-series data of sensory changes is acquired, it is preferable that interactive chat be conducted so that such evaluation expressions are derived, as the time-series data of sensory changes corresponding to the user request information can be used. In this way, the request acquisition unit 21 can acquire user request information containing evaluation expressions based on the messages exchanged by the chat processing unit (chat means). By obtaining user request information, including necessary evaluation terms, through chat, it becomes possible to suggest recipes that match the user's requests.

[0037] The generation unit 22 corresponds to the generation means described above and generates recipe suggestion information corresponding to the user request information acquired by the request acquisition unit 21. In this embodiment, the generation unit 22 uses the AI ​​model 23 and the recipe DB 24 to generate the recipe suggestion information.

[0038] AI model 23 corresponds to the pre-trained model including the LLM described above. In this embodiment, AI model 23 includes a small language model (SLM) in addition to the LLM. In this embodiment, the LLM is preferably constructed by fine-tuning a pre-trained natural language model using a large number of recipe information (e.g., hundreds of thousands to millions) stored on multiple recipe information websites, as well as descriptions, comments, and evaluation information for each recipe, as training data. Furthermore, the SLM consists of fewer parameters than the LLM and is trained specifically for food and beverages.

[0039] As described above, Recipe DB24 stores at least the recipe name, recipe information, and evaluation information for each of the multiple recipes. For example, the recipe information includes ingredient information and cooking information, the evaluation information includes time-series data of sensory changes, sensory evaluation values, sensory evaluation expressions, affective / physiological evaluation expressions, and comments, and also includes food background information and physiological function information for each recipe.

[0040] Examples of material information include the name of the material, quantity, shape, preparation method, nutritional information, cultural information, and allergy information. Examples of cooking information include cooking methods such as baking, boiling, steaming, mixing, and coating, as well as time, temperature, sequence, ingredients used, utensils used, precautions, and the state of each cooking stage. As described above, the time-series data for sensory changes includes time-series data for each of the multiple evaluation items based on the physiological sensations that food and drink evoke during eating and drinking. The time-series data for sensory changes only needs to include time-series evaluation data (time-series data) for each of one or more sensory evaluation items. This time-series data is formed, for example, by a sequence of time-series evaluation values ​​over a predetermined period of time during eating and drinking (for example, from just before putting it in the mouth to a predetermined time after swallowing).

[0041] A sensory evaluation score is an evaluation value for each of one or more sensory evaluation items. Sensory and emotional / physiological evaluation terms refer to words that describe, sensorily, emotionally, or physiologically, what one feels when consuming food or beverages prepared using the recipe in question. Examples of sensory evaluation terms include words like "it feels light / fluffy / hits hard in the mouth," "it has a solid taste," "it has a light taste," "it has a good chewiness," "it leaves a lingering feeling," and "it has a clean finish." Examples of emotional / physiological evaluation terms include words like "it leaves you feeling refreshed," "it feels gentle," "it leaves you feeling nostalgic," "it leaves you feeling full," "it refreshes you," and "it energizes you." Examples of food background information include eating habits, food culture (Japanese food, Western food, etc.), and intended use (breakfast, party, bento box, etc.). Examples of physiological function information include nutrition, functional components, high-protein diets, low-carbohydrate diets, and allergens. However, the information stored in Recipe DB24 is not limited to these examples.

[0042] Recipe DB24 is constructed from a set of datasets containing recipe information and evaluation information generated when evaluators actually consume the food and beverages prepared using each of multiple recipes and perform sensory evaluations or biometric measurements using measuring instruments. In this embodiment, the recipe DB24 is constructed to include information output from a trained recipe evaluation model (hereinafter sometimes abbreviated as the recipe evaluation model) that is built using such datasets as training data (recipe training data).

[0043] The recipe evaluation model is a machine learning model as described above, and its specific data structure and learning method are not limited. For example, the recipe evaluation model may be constructed to take recipe information as input and output time-series data of sensory changes, sensory evaluation values, evaluation expressions (sensory evaluation expressions, affective / physiological evaluation expressions, etc.), and a review. Alternatively, the recipe evaluation model may be constructed to take recipe information and evaluation expressions as input and output time-series data of sensory changes, sensory evaluation values, and a review. In other words, the recipe training data includes, for each of the multiple recipes, the recipe name, recipe information, and evaluation information including time-series data of sensory changes and evaluation expressions, and the recipe evaluation model can be said to output at least time-series data of sensory changes.

[0044] Thus, in constructing the Recipe DB24, by using not only datasets generated from actually consuming food and beverages created using recipes, but also estimated information output from a recipe evaluation model, it is possible to store a large amount of recipe information in the Recipe DB24. While preparing these datasets requires actual consumption and evaluation, by using a recipe evaluation model, it is possible to obtain evaluation information by preparing only the minimum amount of new recipe information, thus enabling the storage of a large amount of recipe information in the Recipe DB24.

[0045] The generation unit 22 only needs to be able to generate recipe suggestion information using such an AI model 23 and recipe DB 24, and there are various ways in which the AI ​​model 23 and recipe DB 24 can be used. For example, Retrieval Augmented Generation (RAG) technology can be used as a method to extend and enhance the input prompts to the LLM using the recipe DB 24.

[0046] The generation unit 22 extracts time-series data of sensory changes corresponding to at least the user request information obtained by the request acquisition unit 21 from the recipe DB 24, and obtains recipe suggestion information by providing the prompt (input prompt) generated based on this extracted time-series data of sensory changes to the LLM of the AI ​​model 23. At this time, the generation unit 22 may also search the recipe DB 24 based on evaluation expression words included in the user request information and extract time-series data of sensory changes associated with those evaluation expression words from the recipe DB 24. In this way, it is possible to generate recipe suggestion information that incorporates information on changes in perception over time as evaluation information for the proposed recipe.

[0047] Furthermore, the generation unit 22 may search the recipe DB 24 based on the expressions contained in the user request information, and extract recipe names, recipe information, sensory change time series data, sensory evaluation values, sensory evaluation expressions, affective / physiological evaluation expressions, reviews, food background information, physiological function information, etc., that are associated with those expressions in the recipe DB 24. The generation unit 22 may generate an input prompt based on this information extracted from the recipe DB 24 and the user request information, and provide this to the SLM of the AI ​​model 23 to generate basic sentences for recipe information and evaluation information. The generation unit 22 may then generate an input prompt based on the data of these generated basic sentences and the user request information, and provide this input prompt to the LLM of the AI ​​model 23 to obtain recipe suggestion information.

[0048] Preferably, the input prompts provided to the LLM are generated in such a way that the LLM outputs recipe suggestion information that includes one or more types of additional information, such as graphs, charts, images, videos, audio, and animations, in addition to basic recipe and evaluation information. For example, by graphing, animating, or creating videos of the sensory change time series data in the recipe suggestion information, the user can easily see how the evaluation values ​​change over time for each sensory evaluation item indicated by the sensory change time series data. In addition, one or more features (strength, rise (gradient), persistence, sharpness (gradient), etc.) of the sensory change time series data may be calculated using methods such as those described in Patent Documents 2 and 3, and these features may be included as evaluation information in the recipe suggestion information. Furthermore, by including the sensory change time series data in the LLM input prompt, evaluation expressions that verbalize the sensory change time series data may be included as evaluation information in the recipe suggestion information. Furthermore, by including images such as diagrams illustrating the cooking process, illustrative diagrams showing the state of food and beverages during the cooking process, and serving suggestions as additional information in the recipe suggestion information, the suggested recipes can be presented to users in an easy-to-understand manner. As shown in the example above, the user request information is significantly expanded by extracting information from the recipe DB24 and outputting from an SLM specifically built for food and beverages. This information is further supplemented by the LLM, and the recipe suggestion information is then formatted to be easily understood by the user. As a result, recipe suggestion information that matches the user's requests and is easy for the user to grasp can be generated.

[0049] The display processing unit 25 corresponds to the display processing means described above and causes the recipe suggestion information generated by the generation unit 22 to be displayed on the display unit. In this embodiment, the display processing unit 25 may display the recipe suggestion information on the display unit of the user terminal 15 or on the display device 5. The display processing unit 25 may also cause the recipe suggestion information to be printed on a printing device (not shown). Thus, the output format of the recipe suggestion information is not limited.

[0050] The feedback processing unit 26 acquires user feedback information regarding the suggested recipe indicated in the recipe suggestion information. For example, the feedback processing unit 26 displays an input screen on the user terminal 15 that allows the user to input their evaluation of the suggested recipe, and acquires the feedback information based on the information entered by the user through that input screen. At this time, the feedback processing unit 26 may also initiate a chat through the display screen of the user terminal 15 to guide the user to input their evaluation of the suggested recipe in as much detail as possible.

[0051] The feedback information obtained may include one or more of the following: text data of evaluation expressions such as "The colors were nice, but the taste was a little too strong and made me thirsty" or "Too sweet"; numerical data indicating satisfaction levels or deviations from preferences; or symbolic data such as ASCII art emoticons. It is preferable that the numerical data be obtained by specifying sensory evaluation items, such as saltiness being -2.0 (weak) and sweetness being +1.0 (strong).

[0052] The update processing unit 27 tunes the recipe DB 24 based on the feedback information obtained by the feedback processing unit 26. For example, if the recipe DB24 is constructed as a graph database using a property graph, and recipe data related to food and beverages is stored in a configuration that includes each element and the properties attached between those elements, the update processing unit 27 updates the properties between specific elements for the proposed recipe based on the feedback information. If the feedback information indicates that it was "a little salty," 0.05 is added to the property value between the element "sensory evaluation" and the element "saltiness" in the recipe DB24. Also, if the feedback information indicates that it would have been better if it had been chewier, 0.1 is subtracted from the property value between the element "sensory evaluation" and the element "texture." However, the data structure of Recipe DB24 is not limited to a graph database; it may also be a vector database.

[0053] The update processing unit 27 updates the time-series data of sensory changes included in the evaluation information of the proposed recipe stored in the recipe DB 24, if necessary, based on the feedback information obtained by the feedback processing unit 26. The update processing unit 27 may include an analysis processing unit (not shown). In this case, the analysis processing unit is configured to analyze the feedback information and output the data items to be updated in the recipe DB 24 (evaluation items in the evaluation information) and numerical values ​​for the updates. As the feedback information can include various data formats as described above, the analysis processing unit may use AI models such as language models like LLM or other machine learning models. The analysis processing unit can also analyze feedback information to identify evaluation items to be updated and the update data from among multiple evaluation items of the sensory change time series data. In this case, the update processing unit 27 updates the time series data of the evaluation items identified by the analysis processing unit in the sensory change time series data included in the evaluation information of the proposed recipe stored in the recipe DB 24, based on the update data identified in the same way.

[0054] Furthermore, the analysis processing unit may identify the update data in such a way that it can distinguish between time periods to be updated and time periods not to be updated within the time-series data of the evaluation items to be updated. For example, the information and update data for the evaluation value immediately after putting the food in the mouth within the time-series data of sensory changes may be identified, while other values ​​(e.g., evaluation values ​​during chewing or after swallowing) may be identified as not to be updated. Then, the update processing unit 27 updates all or part of the time-series data of the identified evaluation items in the recipe DB 24 based on the update data.

[0055] Figure 3 is a conceptual diagram illustrating an example of updating time-series data on sensory changes in the recipe database (DB) 24. Example 1 in Figure 3 illustrates a case where the feedback information includes the text data, "It looked good, but it was saltier than I liked." In this case, the analysis processing unit receives this feedback information, identifies the time-series data of the change in perceived saltiness as an evaluation item to be updated, and identifies the update data to be added to the entire time-series data. As a result, the update processing unit 27 adds up the time-series data of the change in perceived saltiness included in the evaluation information of the proposed recipe in the recipe DB 24. Example 2 in Figure 3 illustrates a case where the feedback information includes the text data, "It had a mild flavor, but overall it was unsatisfying." In this case, the analysis processing unit receives this feedback information, identifies the time-series data of sensory changes related to the flavor of the meat and vegetables as evaluation items to be updated, and identifies the update data to be subtracted from the time-series data overall. As a result, the update processing unit 27 subtracts the time-series data of sensory changes related to the flavor of the meat and vegetables included in the evaluation information of the proposed recipe in the recipe DB 24 overall. Example 3 in Figure 3 illustrates a case where the feedback information includes text data such as "The taste was good, but it was a little heavy." In this case, the analysis processing unit receives this feedback information, identifies the time-series data of sensory changes related to the flavor of the oil as an evaluation item to be updated, and adds up the time-series data overall and identifies update data that extends the aftertaste. As a result, the update processing unit 27 updates the time-series data of sensory changes related to the flavor of the oil included in the evaluation information of the proposed recipe in the recipe DB 24.

[0056] In this way, the time-series data of sensory changes in suggested recipes stored in the recipe DB24 can be updated to match each user's perception. In other words, according to this embodiment, the recipe DB24 can be tuned to match each user's perception of food and drink, making it possible to suggest recipes that match each user's preferences.

[0057] Furthermore, the update processing unit 27 can update the recipe information of the proposed recipe stored in the recipe DB 24 based not only on the time-series data of sensory changes and the evaluation information containing it, but also on the feedback information. Specifically, the update processing unit 27 can add one or more removal-type cooking steps, which are identified based on the relationship between the expressions contained in the feedback information and the ingredients or cooking steps indicated in the recipe information of the proposed recipe, to the cooking steps of the recipe information of the proposed recipe stored in the recipe DB 24.

[0058] Figure 4 shows the correspondence between feedback information, recipe information, and additional removal-type cooking steps. To make each removal-type cooking step easier to understand, Figure 4 also indicates the target of removal for each step. Thus, removal-type cooking steps are processes that remove certain components to improve the taste of the food or beverage being produced, such as off-flavors, bitterness, cloudiness, astringency, oiliness, odor, scum, sliminess, excessive salt or moisture, and fiber. Figure 4 illustrates the following removal-type cooking processes: removing scum, removing bitterness, removing oil, blanching, removing salt, blanching, soaking in water, removing tendons, removing membranes, peeling, removing cores / seeds, removing small bones, draining, removing slime, soaking in milk (washing with milk), washing with vinegar, and rubbing with salt. However, Figure 4 is illustrative, and further information regarding removal-type cooking processes not shown in Figure 4 may be added.

[0059] The update processing unit 27 may maintain a table showing the correspondence relationships as shown in Figure 4, and identify the removal-type cooking steps to be added by referring to this table based on the feedback information and the recipe information of the proposed recipe. In the example in Figure 4, for example, if the feedback information includes an expression such as "strong bitterness" and the recipe information of the proposed recipe includes the cooking step "boiling", then the "scum removal" step is added as a removal-type cooking step to the recipe information of the proposed recipe stored in the recipe DB 24. Also, if the feedback information includes an expression such as "smelly" and the recipe information of the proposed recipe includes seafood ingredients, then the "blanching" step is added as a removal-type cooking step to the recipe information of the proposed recipe stored in the recipe DB 24. However, identifying the additional removal-type cooking steps that need to be added can be achieved not only using the table method described above, but also by using AI models such as language models or pre-trained models. In this case, the AI ​​model should be constructed to take feedback information and recipe information as input and output information on one or more additional removal-type cooking steps.

[0060] By adding necessary removal-type cooking steps to each recipe based on user feedback, the recipe information stored in the recipe DB24 can be tuned to match user preferences, ultimately enabling the provision of recipe suggestions that better suit user tastes.

[0061] <Recipe suggestion method> The details of the recipe proposal method according to this embodiment (the proposed method) will be explained below with reference to Figure 5. Figure 5 is a flowchart showing the recipe proposal method according to this embodiment (the proposed method). The proposed method is executed by one or more computers, such as the proposed device 10 described above. The proposed device 10 can execute the proposed method by having the CPU 1 execute a computer program stored in memory 2. Alternatively, it can be said that the CPU 1 can execute the proposed method through the execution of a computer program stored in memory 2. This computer program is installed via input / output I / F 3 or communication unit 4 from a portable recording medium such as a CD (Compact Disc) or memory card, or from another computer on a network, and stored in memory 2.

[0062] The proposed method includes steps (S51) through (S59), as shown in Figure 5. In the following, the proposed device 10 will be described as the main entity executing each process. However, the main entity executing each process can also be described as a computer or CPU, or as one of the processing modules of the proposed device 10. Since the content of each process is the same as the processing content of each processing module of the proposed device 10, the details of each process will be omitted as appropriate.

[0063] In step (S51), the proposed device 10 (request acquisition unit 21) acquires user request information regarding food and beverages. The user request information to be acquired and the method of acquiring it are as described above for the request acquisition unit 21. The proposed device 10 may also guide the user in a conversational format by performing an interactive chat via communication with the user terminal 15. In this way, user request information that includes as many expressions as possible indicating user requests is acquired.

[0064] In step (S52), the proposed device 10 (generation unit 22) generates input prompts for LLM based on the user request information acquired in step (S51). In this embodiment, the proposed device 10 uses the SLM of the AI ​​model 23 and the recipe DB 24 to generate input prompts. As described above, Recipe DB24 stores at least the recipe name, recipe information, and evaluation information for each of the multiple recipes, and the evaluation information may include time-series data of sensory changes and evaluation expressions. Recipe DB24 is constructed from a set of datasets generated by evaluators actually consuming the food and beverages created in each recipe and performing sensory evaluations or biometric measurements, as well as output information from a trained recipe evaluation model. As mentioned above, SLM also consists of fewer parameters than LLM and is a language model specifically trained for food and beverages.

[0065] For example, the proposed device 10 (generation unit 22) searches the recipe DB 24 based on the expression words contained in the user request information, and extracts the recipe name, recipe information, evaluation information (including time-series data of sensory changes), etc., that are associated with those expression words in the recipe DB 24. The proposed device 10 generates an input prompt based on this information extracted from the recipe DB 24 and the user request information, and provides this to the SLM of the AI ​​model 23 to generate basic sentences for recipe information and evaluation information. Based on the data of these generated basic sentences and the user request information, it generates an input prompt for LLM. In this embodiment, time-series data of sensory changes is identified from the recipe DB24 as evaluation information for recipes corresponding to user request information, and input prompts for LLM are generated based on the user request information, this time-series data of sensory changes, and other information. By using the recipe DB24 and SLM, the input prompts for LLM are enhanced, suppressing hallucination and noise data in the output from LLM described later, and optimizing the output of LLM.

[0066] In step (S53), the proposed device 10 (generation unit 22) generates recipe suggestion information by providing the input prompt generated in step (S52) to the LLM of the AI ​​model 23. The generated recipe suggestion information is as described above, and in addition to the basic information of recipe information and evaluation information related to the suggested recipe, it is desirable to include one or more types of additional information such as graphs, charts, images, videos, audio, and animations. For example, by graphing, animating, and creating videos of sensory change time series data in the recipe suggestion information, the user can be shown in an easy-to-understand manner how the evaluation values ​​change over time for each sensory evaluation item indicated by the sensory change time series data. In addition, images such as diagrams showing the cooking process, image diagrams showing the state of food and beverages during the cooking process, and serving examples may be included in the recipe suggestion information as additional information.

[0067] In step (S54), the proposed device 10 (display processing unit 25) displays the recipe suggestion information generated in step (S53) on the user terminal 15. The display screen at this time may be configured to present the recipe information and evaluation information related to the suggested recipe side by side, or it may be configured to display the recipe information and evaluation information in a switchable manner. Furthermore, if the recipe suggestion information includes one or more types of additional information such as graphs, charts, images, videos, audio, or animations, these additional types of information will also be displayed on the user terminal 15. The configuration of such a display screen is not limited in any way.

[0068] Furthermore, the proposed device 10 (display processing unit 25) only needs to enable the display of recipe suggestion information on the user terminal 15 by some means, and the display format and method of the recipe suggestion information are not limited in any way. For example, the proposed device 10 may enable the display of recipe suggestion information on the user terminal 15 by transmitting the recipe suggestion information to the user terminal 15 using a communication medium such as email or electronic message.

[0069] Thus, through steps (S51) to (S54), the user can not only obtain recipe information related to the suggested recipe by providing information about their desired food and beverages (user request information) via the user terminal 15, but also check evaluation information about the food and beverages created using the suggested recipe. Furthermore, since the evaluation information presented in this embodiment includes information based on time-series data of sensory changes (graphs, animation charts, videos, etc.), the deliciousness of the food and beverages created using the suggested recipe can be easily grasped before actually cooking. In other words, the displayed recipe suggestion information is very useful for the user in selecting which suggested recipes to use.

[0070] The proposed device 10 (feedback processing unit 26) acquires user feedback information regarding the suggested recipe shown in the recipe suggestion information displayed in step (S54) (S55; YES) (S56). If there is no feedback information (S55; NO), the process ends.

[0071] As described above, the proposed device 10 displays an input screen on the user terminal 15 and acquires feedback information based on the information entered by the user via that input screen. At this time, the proposed device 10 may also perform a chat via the display screen of the user terminal 15 to guide the user to enter their evaluation of the proposed recipe in as much detail as possible. The feedback information obtained is not limited in its specific content, but it is preferable that it includes one or more of the following: text data of evaluation expressions such as "The colors were nice, but the taste was a little too strong and made me thirsty"; numerical data indicating satisfaction levels or deviations from preferences; and symbolic data such as ASCII art emoticons.

[0072] Next, in step (S57), the proposed device 10 (the analysis processing unit of the update processing unit 27) analyzes the feedback information acquired in step (S56) and outputs the data items to be updated in the recipe DB 24 (evaluation items in the evaluation information) and the numerical values ​​for the updates. The method for analyzing the feedback information is not limited in any way, as long as it is possible to identify the data items to be updated and the numerical values ​​for the updates from the feedback information. For example, the analysis of the feedback information can be performed using AI models such as language models like LLM or other machine learning models.

[0073] The proposed device 10 (update processing unit 27) determines whether or not the recipe DB 24 needs to be updated by analyzing the process (S57) (S58). For example, if the feedback information contains only text data of descriptive words such as "It looked and tasted just right and was very delicious," the proposed device 10 may determine that the recipe DB 24 does not need to be updated without specifying the data items to be updated and the numerical values ​​for the update (S58; NO). In this case, the process is terminated.

[0074] If the proposed device 10 (update processing unit 27) determines that the recipe DB 24 needs to be updated (S58; YES), it updates the recipe DB 24 based on the analysis results of process (S57) (S59). For example, if the analysis of process (S57) identifies an evaluation item to be updated and the update data from among multiple evaluation items of the sensory change time series data, the proposed device 10 updates the time series data of the identified evaluation item in the sensory change time series data included in the evaluation information of the proposed recipe stored in the recipe DB 24 based on the similarly identified update data (S59). Furthermore, if the analysis of process (S57) identifies update data that allows for the distinction between time periods to be updated and time periods not to be updated within the time-series data of sensory changes for an evaluation item to be updated, the proposed device 10 can also update all or part of the time-series data of the identified evaluation item in the recipe DB 24 based on that update data (S59). An example of updating part of the time-series data of sensory changes for an identified evaluation item is when feedback information indicating "the flavor was great the moment it entered the mouth" or "it was too heavy" is obtained. In the former case, only the data values ​​in the first half of the time-series data of sensory changes related to flavor are updated, and in the latter case, only the data values ​​in the second half of the time-series data of sensory changes related to the flavor of oils and fats are updated. Updating the recipe DB24 based on feedback information may include not only time-series data on sensory changes but also other evaluation information.

[0075] Furthermore, the proposed device 10 (update processing unit 27) can also update the recipe information of the proposed recipe stored in the recipe DB 24 based on feedback information (S59). For example, the proposed device 10 adds one or more removal-type cooking steps, which are identified based on the relationship between the expressions included in the feedback information and the preparations or cooking procedures indicated in the recipe information of the proposed recipe, to the cooking procedures of the recipe information of the proposed recipe stored in the recipe DB 24. The proposed device 10 may identify the removal-type cooking steps to be added by referring to a table showing the correspondence relationships as shown in Figure 4, or it may input feedback information and recipe information into an AI model such as a language model or a trained model to obtain information on one or more removal-type cooking steps to be added.

[0076] In this way, through steps (S55) to (S59), the user can provide feedback information regarding the suggested recipes presented in the recipe suggestion information via the user terminal 15, thereby tailoring the recipe information (including cooking steps) and evaluation information (including time-series data of sensory changes) stored in the recipe DB 24 to their own preferences. Since the personalized information from the recipe DB 24 is used when generating the recipe suggestion information, the user will be able to obtain recipe suggestion information that suits their preferences in the future as well.

[0077] [Differentiation] The above-described embodiment is merely an example and may be partially modified as appropriate. For example, part or all of the AI ​​model 23 (LLM, SLM, etc.) may be stored in an operable format on another computer accessible by the proposed device 10 via communication. Similarly, the recipe DB 24 may also be stored on the same other computer. In this case, the proposed device 10 can transmit the input information for the AI ​​model 23 to the other computer and receive the output information from the AI ​​model 23 from the other computer. Similarly, the proposed device 10 can retrieve the extracted information from the recipe DB 24 on the other computer via communication.

[0078] Furthermore, in the example described above, a configuration in which the proposed device 10 displays a screen that serves as the user interface on the user terminal 15, a so-called server-client configuration, was illustrated. However, the proposed device 10 may also be configured to utilize the display device 5 and input device 6 as the user interface. The proposed device 10 may be a stationary computer or a portable computer.

[0079] Furthermore, although Figure 5 shows multiple processes in sequence, the execution order of each process is not limited to the example shown and can be changed as appropriate without deviating from the intended purpose. For example, processes (S51) to (S54) and processes (S55) to (S59) may be executed separately. Furthermore, SLM is not necessarily required in the process of generating input prompts for LLM (S52). In this case, input prompts for LLM can be generated from user request information and information extracted from recipe DB24.

[0080] Furthermore, while we have described the recipe evaluation model used to construct the Recipe DB24, this recipe evaluation model may also be used for food and beverage marketing and quality comparisons during product development. For example, the evaluation information output from the model, with the recipe information of a new product as input, can be used as an indicator for evaluating the quality of that new product. Also, if the recipe evaluation model is constructed to take recipe information as input and output evaluation expressions and videos / images of that recipe, the evaluation expressions and videos / images output from the model, with the information of a newly devised recipe as input, can be used as promotional text and promotional images / videos for that new recipe.

[0081] Furthermore, the proposed device 10 and method can provide new services through collaboration with external devices or other application functions. For example, the proposed device 10 and method may transmit some or all of the recipe information contained in the generated recipe suggestion information to an automatic cooker. This allows the automatic cooker to automatically cook using the received recipe information, thus providing a convenient, personalized, and delicious meal.

[0082] Furthermore, the proposed device 10 and method can also use the recipe information contained in the generated recipe suggestion information to manage the dietary habits of each user, such as their meal content and nutritional intake, as well as their food preferences. For example, by having the user input whether or not they have adopted the suggested recipe shown in the displayed recipe suggestion information, the proposed device 10 and method can obtain information on the user's meal content and nutritional intake based on the recipe information of that recipe suggestion information. In addition, by maintaining a history of recipe suggestion information generation for each user, it is possible to manage changes in each user's food preferences. Furthermore, the proposed device 10 and method may acquire biometric information such as body temperature, blood pressure, and pulse rate, as well as behavioral information from the user, manage this information, and use it to generate recipe suggestion information. This makes it possible to manage the user's physical condition and provide recipe suggestions that take the user's physical condition into account.

[0083] Some or all of the above embodiments and modifications may also be specified as follows; however, the above embodiments and modifications are not limited to those described below.

[0084] <1> A means for obtaining user requests regarding food and beverages, A generation means that generates recipe suggestion information, which includes at least recipe information and evaluation information, regarding suggested recipes that correspond to the user request information, using a trained model that includes a large-scale language model, A display processing means for displaying the aforementioned recipe suggestion information on a display unit, Equipped with, The generation means extracts information corresponding to the user request information from a recipe database that stores the recipe name, recipe information, and evaluation information for each of the multiple recipes, in order to generate the recipe suggestion information. The one or more evaluation pieces of information stored in the recipe database include sensory change time series data that quantitatively shows the changes in sensory perception over time when consuming food or beverages prepared using the target recipe, and evaluation expressions related to the food or beverages prepared using the target recipe. Recipe suggestion device. <2> The generation means extracts the time-series data of sensory changes corresponding to the user request information from the recipe database, and obtains the recipe suggestion information by providing the large-scale language model with prompts generated based on the extracted time-series data of sensory changes. <1> The recipe suggestion device described above. <3> A means for obtaining user feedback information regarding the proposed recipe, An update means that updates the time-series data of sensory changes included in the evaluation information of the proposed recipe stored in the recipe database based on the acquired feedback information, It also has <1> or <2> The recipe suggestion device described above. <4> The aforementioned time-series data of sensory changes includes time-series data for each of several evaluation items based on the physiological sensations induced by food and drink during consumption. The update means includes an analysis means that analyzes the feedback information to identify an evaluation item to be updated and update data from among the multiple evaluation items of the sensory change time series data, and updates the time series data of the identified evaluation item in the sensory change time series data included in the evaluation information of the proposed recipe stored in the recipe database based on the identified update data. <3> The recipe suggestion device described above. <5> The analysis means identifies the update data in such a way that it is possible to distinguish between the time periods to be updated and the time periods not to be updated within the time series data of the evaluation item to be updated. The update means updates all or part of the time-series data of the identified evaluation items in the recipe database based on the update data. <4> The recipe suggestion device described above. <6> The update means adds one or more removal-type cooking steps, identified based on the relationship between the expression words included in the acquired feedback information and the ingredients or cooking steps indicated in the recipe information of the proposed recipe, to the cooking steps of the recipe information of the proposed recipe stored in the recipe database. <3> from <5> A recipe suggestion device described in any one of the following. <7> The aforementioned recipe database includes information output from a trained recipe evaluation model constructed using recipe training data. The aforementioned recipe training data includes, for each of the multiple recipes, the recipe name, the recipe information, and the evaluation information including the time-series data of sensory changes and the evaluation expression words. The trained recipe evaluation model outputs at least the time-series data of sensory changes. <1> from <6> A recipe suggestion device described in any one of the following. <8> The aforementioned means for obtaining requests is, This includes a chat method for exchanging messages with the user's terminal, Based on the messages exchanged via the chat means, the user request information, including the evaluation expression, is obtained. <1> from <7> A recipe suggestion device described in any one of the following. <9> A recipe suggestion method performed by a computer, The aforementioned computer, The process involves obtaining user request information regarding food and beverages, A generation step that generates recipe suggestion information, which includes at least recipe information and evaluation information, regarding suggested recipes that correspond to the user request information, using a trained model that includes a large-scale language model. A display processing step for displaying the aforementioned recipe suggestion information on the display unit, Execute, In the generation process, for the purpose of generating the recipe suggestion information, information corresponding to the user request information is extracted from a recipe database that stores the recipe name, recipe information, and evaluation information for each of the multiple recipes. The one or more evaluation pieces of information stored in the recipe database include sensory change time series data that quantitatively shows the changes in sensory perception over time when consuming food or beverages prepared using the target recipe, and evaluation expressions related to the food or beverages prepared using the target recipe. Recipe suggestion methods. <10> In the generation process, the computer extracts the time-series data of sensory changes corresponding to the user request information from the recipe database, and obtains the recipe suggestion information by providing the large-scale language model with prompts generated based on the extracted time-series data of sensory changes. <9> Recipe suggestion method as described above. <11> The aforementioned computer, The process of obtaining user feedback information regarding the proposed recipe, An update step of updating the time-series data of sensory changes included in the evaluation information of the proposed recipe stored in the recipe database based on the acquired feedback information, Perform further <9> or <10> Recipe suggestion method as described above. <12> The aforementioned time-series data of sensory changes includes time-series data for each of several evaluation items based on the physiological sensations induced by food and drink during consumption. The update step includes an analysis step that analyzes the feedback information to identify the evaluation items to be updated and the update data from among the multiple evaluation items of the sensory change time series data, In the update step, the computer updates the time-series data of the identified evaluation item in the time-series data of sensory changes included in the evaluation information of the proposed recipe stored in the recipe database, based on the identified update data. <11> Recipe suggestion method as described above. <13> In the analysis step, the computer identifies the update data in such a way that it can distinguish between the time periods to be updated and the time periods not to be updated within the time series data of the evaluation item to be updated. In the update step, the computer updates all or part of the time-series data of the identified evaluation items in the recipe database based on the updated data. <12> Recipe suggestion method as described above. <14> In the update step, the computer adds one or more removal-type cooking steps, which are identified based on the relationship between the expression words included in the acquired feedback information and the ingredients or cooking steps indicated by the recipe information of the proposed recipe, to the cooking steps of the recipe information of the proposed recipe stored in the recipe database. <11> from <13> Recipe suggestion method described in one of the following ways. <15> The aforementioned recipe database includes information output from a trained recipe evaluation model constructed using recipe training data. The aforementioned recipe training data includes, for each of the multiple recipes, the recipe name, the recipe information, and the evaluation information including the time-series data of sensory changes and the evaluation expression words. The trained recipe evaluation model outputs at least the time-series data of sensory changes. <9> from <14> Recipe suggestion method described in one of the following ways. <16> The aforementioned request acquisition process includes a chat process in which messages are exchanged with the user terminal. In the request acquisition step, the computer acquires the user request information, including the evaluation expression, based on the messages exchanged in the chat step. <9> from <15> Recipe suggestion method described in one of the following ways. <17> A computer program stored in one or more memory locations and executed by one or more processors, <9> from <15> A computer program that causes one or more computers equipped with one or more memory and one or more processors to execute one of the recipe suggestion methods described in any one of the above. [Explanation of symbols]

[0085] 1 CPU 2 memory 3 Input / Output Interfaces 4. Communication Unit 5 Display device 6 Input devices 10. Recipe suggestion device (this proposed device) 11. Communication Network 15 User terminals 21 Request Acquisition Department 22 Generation part 23 AI Models 24 Recipe Database (DB) 25 Display Processing Unit 26 Feedback Processing Unit 27 Update Processing Unit

Claims

1. A means for obtaining user requests regarding food and beverages, A generation means that generates recipe suggestion information, which includes at least recipe information and evaluation information for a suggested recipe corresponding to the user request information, using a recipe database that stores recipe names, recipe information, and evaluation information for each of multiple recipes, including a trained model that includes a large-scale language model, A display processing means for displaying the aforementioned recipe suggestion information on a display unit, Equipped with, The one or more evaluation pieces of information stored in the recipe database include time-series data of sensory changes that quantitatively show the changes in sensory perception over time when consuming food or beverages prepared using the target recipe. The generation means extracts the time-series data of sensory changes corresponding to the user request information from the recipe database, and obtains the recipe suggestion information by providing the large-scale language model with prompts generated using at least the extracted time-series data of sensory changes. Recipe suggestion device.

2. The one or more evaluation pieces of information stored in the recipe database include the time-series data of sensory changes relating to the target recipe and evaluation expressions relating to the food and beverage created by the target recipe, The request acquisition means acquires the user request information, which includes evaluation terms related to food and beverages. The generation means extracts the time-series data of sensory changes corresponding to the evaluation expression words included in the acquired user request information from the recipe database. The recipe suggestion device according to claim 1.

3. A means for obtaining user feedback information regarding the proposed recipe, An update means that updates the time-series data of sensory changes included in the evaluation information of the proposed recipe stored in the recipe database based on the acquired feedback information, The recipe suggestion device according to claim 1, further comprising the following:

4. The aforementioned time-series data of sensory changes includes time-series data for each of several evaluation items based on the physiological sensations induced by food and drink during consumption. The update means includes an analysis means that analyzes the feedback information to identify an evaluation item to be updated and update data from among the multiple evaluation items of the sensory change time series data, and updates the time series data of the identified evaluation item in the sensory change time series data included in the evaluation information of the proposed recipe stored in the recipe database based on the identified update data. The recipe suggestion device according to claim 3.

5. The analysis means identifies the update data in such a way that it is possible to distinguish between the time periods to be updated and the time periods not to be updated within the time series data of the evaluation item to be updated. The update means updates all or part of the time-series data of the identified evaluation items in the recipe database based on the update data. The recipe suggestion device according to claim 4.

6. The update means adds one or more removal-type cooking steps, identified based on the relationship between the expression words included in the acquired feedback information and the ingredients or cooking steps indicated in the recipe information of the proposed recipe, to the cooking steps of the recipe information of the proposed recipe stored in the recipe database. A recipe suggestion device according to any one of claims 3 to 5.

7. The aforementioned recipe database includes information output from a trained recipe evaluation model constructed using recipe training data. The aforementioned recipe training data includes, for each of the multiple recipes, the recipe name, the recipe information, and the evaluation information including the time-series data of sensory changes. The trained recipe evaluation model outputs at least the time-series data of sensory changes. The recipe suggestion device according to claim 1.

8. The aforementioned means for obtaining requests is, This includes a chat method for exchanging messages with the user's terminal, Based on the messages exchanged via the chat means, the user request information, including the evaluation expression, is obtained. The recipe suggestion device according to claim 2.

9. A recipe suggestion method performed by a computer, The aforementioned computer, The process of obtaining user request information regarding food and beverages, A generation process that generates recipe suggestion information, which includes at least recipe information and evaluation information for a suggested recipe corresponding to the user request information, using a recipe database that stores the recipe name, recipe information, and evaluation information for each of the trained models, including a large-scale language model, and multiple recipes, A display processing step for displaying the aforementioned recipe suggestion information on the display unit, Execute, The one or more evaluation pieces of information stored in the recipe database include time-series data of sensory changes that quantitatively show the changes in sensory perception over time when consuming food or beverages prepared using the target recipe. In the generation step, the time-series data of sensory changes corresponding to the user request information is extracted from the recipe database, and the recipe suggestion information is obtained by providing the large-scale language model with prompts generated using at least the extracted time-series data of sensory changes. Recipe suggestion methods.

10. A computer program stored in one or more memories and executed by one or more processors, which causes one or more computers equipped with the one or more memories and one or more processors to execute the recipe suggestion method described in claim 9.

Images