Recipe evaluation device and method based on review information of food and beverage

Abstract

To provide a technique capable of highly accurate recipe evaluation using review information.SOLUTION: A recipe evaluation device (10) comprises acquisition means (23) that acquires review information on food or beverage created with a target food recipe and specific information related to the food or beverage created with the target food recipe, and estimation means (23) that estimates evaluation values of multifaceted evaluation items of the food or beverage created with the target food recipe on the basis of the acquired review information and specific information. The recipe information may be acquired from a generative AI so as to match food preferences and requests, and the review information and the specific information may also be acquired from a generative AI. The recipe evaluation device can extract factor information (such as preparation steps and ingredients) indicating a causal relationship or a correlation relationship with the evaluation items using causal inference or feature selection.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a technique for evaluating food and drink recipes. [Background technology]

[0002] In recent years, Japan has been hit by the problems of an aging population and solitary eating, while globally, population growth has led to the maintenance of food supply and food waste. Meanwhile, information processing technologies such as artificial intelligence (AI) have rapidly advanced and become widespread, providing applications that suggest recipes based on ingredient information. For example, Patent Document 1 below discloses a recipe suggestion device that acquires ingredient information for ingredients selected by a user terminal as selected ingredient information, and extracts recipe information that includes at least one ingredient from the selected ingredient information based on recipe information that associates a recipe with the ingredients used in the recipe.

[0003] In addition to such recipe suggestions, there are also methods that use information processing technology to measure the overall deliciousness of a sample such as food. Patent Document 2 below discloses a deliciousness measurement device that calculates a score indicating the deliciousness of a sample based on the pressure that the sample exerts on a pressure detection surface divided into multiple areas. Furthermore, Patent Document 3 below discloses an information processing device that estimates the taste tendency of a dish prepared according to recipe information based on at least the information on ingredients and quantities used in the recipe information, and presents information on the estimated taste tendency to the user, thereby improving the convenience of the user in using the recipe. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2022-157379 [Patent Document 2] Patent Publication No. 2021-15062 [Patent Document 3] International Publication No. 2018 / 42589 Summary of the Invention [Problem to be solved by the invention]

[0005] As disclosed in Patent Document 1, it is possible to automatically generate and output recipe information using information processing technology, but the recipe has not been evaluated in terms of the taste and appearance of the food prepared based on the recipe, or its effects on the body. Although Patent Document 3 estimates the taste tendency of a dish prepared according to recipe information, it merely estimates the intensity of sweetness, sourness, saltiness, bitterness, and spiciness based on the amounts of various components contained in the ingredients and seasonings used, which is not sufficient for evaluating a recipe.

[0006] On the other hand, there are currently several recipe information websites on the Internet, which not only provide a large number of recipes but also post user reviews and comments for each recipe. Such recipe review information is important information that indicates the evaluation of each recipe.

[0007] The present invention has been made in view of the above circumstances, and provides a technique that enables highly accurate recipe evaluation using review information. [Means for solving the problem]

[0008] According to the present invention, a recipe evaluation device is provided which includes an acquisition means for acquiring review information of food and beverages prepared using a target recipe and specific information regarding the food and beverages prepared using the target recipe, and an estimation means for estimating evaluation values ​​of multifaceted evaluation items of the food and beverages prepared using the target recipe based on the acquired review information and specific information.

[0009] Furthermore, according to the present invention, there is provided a recipe evaluation method executed by one or more computers, in which the one or more computers execute the steps of acquiring review information for food and beverages prepared using a target recipe and specific information regarding the food and beverages prepared using the target recipe, and estimating evaluation values ​​for multiple evaluation items of the food and beverages prepared using the target recipe based on the acquired review information and specific information.

[0010] Furthermore, according to the present invention, there can also be provided a computer program stored in one or more memories and executed by one or more processors, which causes the recipe evaluation method to be executed by the one or more computers having the one or more memories and the one or more processors, and a recording medium on which the computer program is recorded. [Effects of the Invention]

[0011] According to the present invention, it is possible to provide a technique that enables highly accurate recipe evaluation using review information. [Brief explanation of the drawings]

[0012] [Figure 1] FIG. 1 is a diagram conceptually illustrating an example of the hardware configuration of a recipe evaluation device (present evaluation device) according to the present embodiment. [Figure 2] FIG. 2 is a diagram conceptually illustrating an example of the software configuration of the recipe evaluation device (the evaluation device) according to the present embodiment. [Figure 3] FIG. 1 is a diagram conceptually illustrating an example of tuning a recipe database (DB). [Figure 4] FIG. 10 is a diagram conceptually illustrating an example of structured recipe data. [Figure 5] FIG. 10 is a diagram showing an example of displaying the causal relationship between kokumi as an evaluation item and each factor. [Figure 6] 1 is a flowchart showing a recipe evaluation method (present evaluation method) according to the present embodiment. [Figure 7]10 is a flowchart showing a specific example of a recipe information generation process in the recipe evaluation method (present evaluation method) according to the present embodiment. [Figure 8] FIG. 10 is a diagram conceptually illustrating an example of the software configuration of a recipe evaluation device according to a modified example. DETAILED DESCRIPTION OF THE INVENTION

[0013] Hereinafter, an embodiment of the present invention (hereinafter referred to as the present embodiment) will be described. Note that the embodiment described below is an example, and the present invention is not limited to the configuration of the following embodiment.

[0014] First, an overview of the recipe evaluation device and recipe evaluation method according to this embodiment will be described. In this embodiment, review information for the food and beverages created using the target recipe and specific information regarding the food and beverages created using the target recipe are obtained, and based on the obtained review information and specific information, evaluation values ​​for the multifaceted evaluation items of the food and beverages created using the target recipe are estimated.

[0015] The food and drink created by the target recipe in this embodiment includes not only food but also drinks such as soup, stew, liquid, cocktail, juice, etc. "Review information" is information such as critiques, comments, evaluations, overviews, reports, opinions, and impressions about the food and drink created using the target recipe. "Specific information about food and drink" means certain specific information about food and drink, and includes some or all of the information indicated in the recipe information of the target recipe. For example, the specific information may be the name of the food and drink made by the target recipe (such as the name of the dish), or some of the preparations or preparation methods indicated in the recipe information. Hereinafter, the term "specific information" refers to this "specific information about food and drink." The "recipe information" of a target recipe includes at least information on how to prepare the food or drink to be created using the target recipe and the preparations required to prepare the food or drink. The preparations indicated in the recipe information may include any items required to prepare the target food or drink, such as food ingredients, seasonings, nutrients, functional ingredients, cooking utensils, cooking equipment, etc. The specific information may be formed from only one piece of information indicated in the recipe information, such as the name of the dish, but it is preferable to include more information in the recipe information, such as the preparations and the method of preparation, in order to improve the accuracy of the recipe evaluation.

[0016] The acquisition of such review information and specific information may be realized by obtaining output data from a trained model such as a large-scale language model, as in the embodiment described below, or may be acquired as input data from a user, or may be acquired from information posted on a recipe information providing site, and the acquisition form is not limited in any way.

[0017] Multifaceted evaluation items for food and beverages are multiple evaluation items from various aspects of food and beverages, and include, for example, any one, any plurality, or all of the following evaluation items: taste evaluation items, texture evaluation items, appearance evaluation items, temperature evaluation items, sound evaluation items during eating and drinking, functional evaluation items, sensory evaluation items after eating and drinking, palatability evaluation items, or combined evaluation items.

[0018] "Taste evaluation items" are evaluation items related to the taste of food and beverages, and include evaluation items related to one or more of the five basic tastes: sweetness, sourness, saltiness, umami, and bitterness, as well as astringency, spiciness, richness, and flavor (aroma). In other words, taste evaluation items can be said to include one or more of evaluation items related to tastes sensed through taste bud cells based on the sense of taste, such as the five basic tastes; evaluation items related to tastes sensed without the use of taste bud cells based on the sense of touch, such as astringency and spiciness; and evaluation items related to tastes sensed without the use of taste bud cells based on the sense of smell, such as flavor. Taste evaluation items can also be said to be evaluation items related to the secondary functions of food. While the five basic tastes can be identified as having specific chemical substances as their main cause, "kokumi" is a taste based on a complex set of factors, including multiple chemical substances and physical properties.

[0019] "Texture evaluation items" are evaluation items related to the texture of food and beverages, and may be items that comprehensively evaluate the degree of texture, or may include one or more evaluation items that are subdivided into chewiness, texture, tongue feel, throat feel, etc. "Appearance evaluation items" are evaluation items related to the appearance of food and beverages, and may be items that comprehensively evaluate the degree of beauty of the appearance, or may include one or more evaluation items that are subdivided into color, gloss, luster, texture, etc. "Temperature evaluation item" refers to an evaluation item related to the temperature of food or drink, and may be an item that comprehensively evaluates whether the temperature is appropriate for food or drink, or may include one or more evaluation items that are subdivided into the length of time the warmth lasts, the state of coldness (such as ice cold), etc. "Evaluation items for sounds made during eating and drinking" are evaluation items related to sounds made before, during, and after eating and drinking food and may be items that comprehensively evaluate the quality of the sound, or may include one or more evaluation items subdivided by type of sound, such as the sound of chewing, the sound of fizzing, or the sizzling sound of grilling.

[0020] "Functional evaluation items" are evaluation items related to the functionality of food and drink, and include one or more evaluation items related to nutritional components, functional components, functions, etc. Nutritional components refer to nutrients necessary for human life, while functional components refer to food components contained in food that are involved in bioregulatory functions and are expected to have functional effects such as maintaining health and mitigating harmful effects. In Japan, foods that meet certain standards are permitted to be labeled as functional health foods (functional food, nutrient-specific functional food, specified health food), and there are also systems independently certified by local governments, such as the Hokkaido Food Functionality Labeling System. In this document, the terms "nutritional components" and "functional components" are used to refer to general nutrition and functionality, including but not limited to the nutrition and functionality defined by the Japanese system. Evaluation items for nutritional components may include items that comprehensively evaluate nutritional value, or may include evaluation items that are subdivided into nutritional components such as carbohydrates, lipids, proteins, and vitamins. The evaluation items for functional ingredients relate to ingredients that are expected to have a functional effect, and may include items that comprehensively evaluate the level of the functional ingredient contained, or may include evaluation items that are subdivided by ingredient, such as lycopene, tea catechins, GABA, etc. Functional evaluation items relate to the expected physical effects of ingesting food or drink, and include evaluation items that are subdivided by type of function, such as regulating the stomach, thinning the blood, and relieving fatigue. In this way, functional evaluation items can also be said to be evaluation items related to the primary and tertiary functions of food. In addition, functional evaluation items may include evaluation items related to functions other than those described above, such as evaluation items related to functions to adapt to specific physical conditions such as need for nursing care, eating / swallowing disorders, chewing disorders, and allergic disorders. Evaluation items for functions related to need for nursing care, eating / swallowing disorders, and chewing difficulties include things like slowing down the speed at which food and drink enter the pharynx. Evaluation items for functions related to allergic disorders may include things like symptom relief.

[0021] "Sensory evaluation items after eating and drinking" are evaluation items related to the sensations obtained after consuming food and drink, and may include items that comprehensively evaluate the degree of satisfaction, or may include evaluation items that are subdivided into sensations such as the degree of satisfaction and the degree of fullness. "Preference evaluation items" are preference evaluation items related to food and drink, and may include items that comprehensively evaluate the degree of likes and dislikes, or may include one or more more detailed evaluation items such as physical condition, beauty condition, mental and physical condition, food culture, eating experience, eating habits, eating life, and eating environment. The evaluation items for physical condition may include an item for comprehensively evaluating the physical condition, or may include more detailed evaluation items for symptoms of poor physical condition, such as drowsiness, stomachache / bloating / constipation / diarrhea, slight fever, coughing, fatigue, and tiredness. The evaluation items for beauty condition may include skin dryness (e.g., dryness), moisture / oil content, firmness, and gloss, as indicators of personal beauty, and hair gloss, dryness, damage, and thickness, and may include more detailed evaluation items for each of these symptoms. The evaluation items for mental and physical condition include the joy, excitement, pleasure, calmness, irritability, concentration, vitality, anxiety, fear, and disappointment derived from the food or drink. The evaluation items for food culture may include an item for comprehensively evaluating the degree to which the food or drink's food culture is perceived, or may include one or more detailed evaluation items, such as the degree of Japaneseness, the degree of Koreanness, and the degree of locality. The evaluation items for food experience are those related to past memories evoked by the food or drink, such as the degree of nostalgia, the degree to which it evokes memories of places or people, and the degree of similarity to the home cooking of close friends such as mothers or grandmothers. The evaluation items for dietary habits are those related to the incorporation of the food or drink into one's eating habits, and include evaluation items subdivided into the degree to which it is suitable for breakfast, lunch, dinner, snacks, or late-night snacks, and the degree to which it is suitable for eating at home or eating out. The evaluation items for the eating environment include evaluation items such as the degree to which it is suitable for hot or cold climates, high or low humidity, the degree to which it is suitable for regional characteristics such as being in the mountains, valleys, along a river, or by the sea, and the degree to which it is suitable for eating alone, eating with multiple people, or eating with family.

[0022] "Combination evaluation items" are evaluation items related to food and drink combinations, and may include items that comprehensively evaluate the quality of the combination of ingredients and components contained in the food and drink, or the quality of the combination of the food and drink with other foods and drinks. For example, a good combination could be the effect of eating fried chicken with lemon, while a bad combination could be a negative combination that puts a strain on the stomach and intestines, such as a combination of fatty ingredients with protein. Furthermore, evaluation items for the degree to which a food and drink goes well with alcoholic beverages such as beer or sake may also be included as combination evaluation items.

[0023] The "evaluation value" estimated for such multifaceted evaluation items may be a value indicating the evaluation of each multifaceted evaluation item, and the specific form thereof is not limited. For example, the evaluation value may be indicated by a binary value indicating whether or not each evaluation item is applicable, or may indicate a level of applicability (e.g., 1 to 5) for each evaluation item, or may indicate the possibility of applicability for each evaluation item as a numerical value such as a probability value.

[0024] In this manner, in this embodiment, the evaluation values ​​of the multifaceted evaluation items of the food and drink prepared using the target recipe are estimated based on the review information and specific information acquired regarding the target recipe. Since the review information is based on the impressions of experienced people who have actually created the recipe in question, according to this embodiment, highly accurate recipe evaluation is possible by estimating the evaluation values ​​of multiple evaluation items for food and drink based on such review information.

[0025] Furthermore, the evaluation values ​​of the above-mentioned multifaceted evaluation items may further include a time-dependent evaluation value that indicates an evaluation over time when eating food and drink prepared according to a recipe in the recipe information. The "evaluation value over time" is composed of multiple evaluation values ​​at multiple timings over time when eating or drinking food and drink, for at least some of the above-mentioned multifaceted evaluation items. For example, it is composed of evaluation values ​​at two or more timings including before putting the food in the mouth, the moment it is put in the mouth, when chewing, when it is mixed with saliva, when it is swallowed, and after swallowing. Furthermore, the evaluation values ​​at each timing that make up the evaluation value over time may be evaluation values ​​at a certain moment, or may include evaluation values ​​of continuity. For example, as an evaluation value for the taste evaluation item, a time-dependent evaluation value may be estimated such that the sourness is strong the moment it is put in the mouth, the umami gradually increases during chewing, and a slight spiciness appears when it is swallowed. Such evaluation values ​​over time enable more accurate recipe evaluation.

[0026] Hereinafter, a recipe evaluation device (hereinafter sometimes abbreviated as the present evaluation device) and a recipe evaluation method (hereinafter sometimes abbreviated as the present evaluation method) according to this embodiment will be described in detail. As described above, this embodiment does not limit the food or drink prepared using the recipe to be evaluated. However, in the following specific examples, for ease of understanding, dishes, mainly food, prepared using the recipe to be evaluated will be used as examples.

[0027] <Recipe evaluation device> FIG. 1 is a diagram conceptually showing an example of the hardware configuration of a recipe evaluation device (present evaluation device) 10 according to this embodiment. The evaluation device 10 is a so-called computer, and includes a CPU (Central Processing Unit) 1, a memory 2, an input / output interface (I / F) 3, a communication unit 4, and the like. The CPU 1 is a so-called processor, and may include an application specific integrated circuit (ASIC), a digital signal processor (DSP), a graphics processing unit (GPU), and the like in addition to a general CPU. The memory 2 is a RAM (Random Access Memory), a ROM (Read Only Memory), or an auxiliary storage device (such as a hard disk).

[0028] The input / output I / F 3 can be connected to user interface devices such as a display device 5 and an input device 6. The display device 5 is a device that displays a screen corresponding to drawing data processed by the CPU 1 or the like, such as an LCD (Liquid Crystal Display) or CRT (Cathode Ray Tube) display. The input device 6 is a device that accepts input of user operations, such as a keyboard or mouse. The display device 5 and the input device 6 may be integrated and realized 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. A portable recording medium or the like can 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 with them.

[0029] The hardware configuration of the present evaluation device 10 is not limited to the example shown in FIG. 1. The present evaluation device 10 may include other hardware elements not shown. Furthermore, the number of each hardware element is not limited to the example shown in FIG. 1. For example, the present evaluation device 10 may have multiple CPUs 1. Furthermore, the present evaluation device 10 may be realized by multiple computers each having multiple housings.

[0030] The communication network 11 is composed of one or more of a public network such as the Internet, a wide area network (WAN), a local area network (LAN), a wireless communication network, etc. However, in this embodiment, the form of communication between the evaluation device 10 and the user terminal 15 is not limited.

[0031] 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 evaluation device 10 and output information provided by the evaluation device 10. As long as such information output is possible, the hardware and software configurations of the user terminal 15 are not limited. The user terminal 15 displays a screen that serves as a user interface for the evaluation device 10. The output format of the user terminal 15 will be described later.

[0032] The evaluation device 10 performs recipe evaluation using the software configuration shown in FIG. FIG. 2 is a diagram conceptually showing an example of the software configuration of the recipe evaluation device (present evaluation device) 10 according to this embodiment. 2, the evaluation device 10 includes a requirement acquisition unit 21, a recipe generation unit 22, an estimation unit 23, a display processing unit 25, an operation processing unit 26, a change processing unit 27, etc. Each of these processing modules is realized, for example, by the CPU 1 executing a computer program stored in the memory 2. This computer program may be installed from a portable recording medium such as a CD (Compact Disc) or a memory card, or from another computer on a network, via the input / output I / F 3 or the communication unit 4, and stored in the memory 2.

[0033] The request acquisition unit 21 acquires user request information regarding food and drink. The user request information is information indicating the user's request regarding the food or drink that the user wants to eat, make, or serve, and includes condition information for the food or drink, such as the taste, the ingredients or seasonings that the user wants to use, the food components or nutrients that the user wants to eat, whether the food or drink is hot or cold, the name of the menu, etc. For example, character string data such as "Please tell me a warm dish that uses lots of vegetables that children will enjoy" is acquired as the user request information. The user request information may also include cooking condition information for cooking food and drink, such as the cooking time required, difficulty level, etc. For example, character string data such as "Tell me a time-saving recipe using sardines" or "Tell me an easy pasta dish that can be made using just one frying pan" is acquired as the user request information. The data format of the user request information acquired in this embodiment is character string data in a description format consisting of character data such as a sentence format, an itemized format, or a set of words, but the data format is arbitrary.

[0034] The request acquisition unit 21 acquires user request information from another computer via communication. For example, the request acquisition unit 21 may generate the user request information based on information input by the user via an input screen displayed on the user terminal 15 via communication, or the user request information itself may be input by the user by operating the user terminal 15. In this way, the method of acquiring information by the request acquisition unit 21 is not limited in any way.

[0035] The recipe generation unit 22 generates recipe information and review information based on the user request information acquired by the request acquisition unit 21 using a large-scale language model. Large-scale language models (LLMs) are trained natural language models built using large text datasets and deep learning techniques, enabling natural language recognition and generation. Large-scale language models are a type of generative AI, and examples include "BERT" and "GPT-4." The large-scale language model used in this embodiment is constructed by fine-tuning a pre-trained natural language model using as training data a large number (e.g., hundreds of thousands to millions) of recipe information sets stored on multiple recipe information providing sites and review information sets such as user reviews and comments on each recipe. Hereinafter, this large-scale language model may also be referred to as generative AI.

[0036] As outlined above, the recipe information output from this generation AI includes at least information on the creation method (cooking method) and ingredients for the target recipe. For example, the recipe information is output as text data including information on the cooking method, ingredients, and cooking notes for the target dish. The cooking method includes cooking steps, cutting methods, preparation methods, stock, cooking conditions, etc. The ingredients include ingredients, seasonings, cooking utensils, etc. The cooking notes include the name of the dish, the category of the dish, the concept of the dish, the origin and expiration date of the ingredients, calories, nutrients, food components, shelf life, costs, etc.

[0037] The review information output by the generation AI is generated based on the training data used in the generation AI's learning process, i.e., a large number of review information sets from recipe information providers. Therefore, even if a recipe exactly identical to the target recipe is not posted on the recipe information provider, the information is likely to be quite close to how a user would feel about the food cooked using the target recipe. This review information includes qualitative information corresponding to each evaluation item among the multifaceted evaluation items of the dish cooked using the target recipe, and quantitative information corresponding to the evaluation value of each evaluation item. For example, the review information is generated as text data including multiple pieces of qualitative information and quantitative information corresponding to each piece of qualitative information. As a specific example, the generation AI generates review information such as, "It has a spicy kick, but also a sufficient amount of sweetness, so children will enjoy it." In this example, "spicy" and "sweet" are qualitative information corresponding to the evaluation items of taste, "children will enjoy it" is qualitative information corresponding to the "palatability evaluation item," and "spicy kick" and "sufficient sweetness" are quantitative information for the evaluation items of taste.

[0038] The generation AI may be stored in an operable state on the evaluation device 10 (see Figure 2), or may be stored in an operable state on another computer that the evaluation device 10 can access via communication. In the former case, the recipe generation unit 22 inputs user request information to the generation AI of the evaluation device 10, causing the generation AI to generate recipe information and review information. In the latter case, the recipe generation unit 22 transmits user request information to the other computer and inputs it to the generation AI, thereby acquiring the recipe information and review information output from the generation AI via communication.

[0039] The recipe generation unit 22 may apply a filter to the recipe information output from the generation AI to eliminate noise data. Noise data may be recipe information that is clearly impossible to achieve or unrealistic or inappropriate recipe information (e.g., stir-frying a poisonous snake as an ingredient). For example, the recipe generation unit 22 may determine whether or not to filter the text data of the recipe information based on a score obtained by applying an existing method (e.g., a BERT score) for scoring the grammatical correctness of text to the text data of the recipe information output from the generation AI. If it is determined that filtering is necessary, the recipe generation unit 22 may present the recipe information to the user by causing the display processing unit 25 (described later) to perform a display process on the recipe information, allowing the recipe information to be modified by user operation. Furthermore, the recipe generation unit 22 may search for words that may be noise from the text data of the output recipe information and automatically eliminate them. The recipe generation unit 22 may also apply a filter to the review information output from the generation AI in the same way as the recipe information, to remove noise data.

[0040] In addition, the recipe generation unit 22 can suppress hallucination and noise data in the LLM output and optimize the LLM output by expanding and strengthening the input information (input prompt) to the large-scale language model (LLM) using Retrieval Augmented Generation (RAG) technology, Small Language Model (SLM), etc. In this case, a RAG database (hereinafter sometimes referred to as recipe DB) is constructed with a large number of appropriate recipe data related to food and drink, and the recipe generation unit 22 queries the recipe DB based on user request information to acquire recipe information (hereinafter referred to as related recipe information) that is highly relevant to the user request. The recipe generation unit 22 generates an input prompt by providing the acquired related recipe information to a small language model (SLM) that has been trained specifically for food and drink. The SLM is composed of fewer parameters than the LLM.

[0041] It is desirable that the recipe DB be constructed to include recipe data not used in LLM learning and the latest recipe data so that it can supplement knowledge that LLM does not have. It is also desirable that SLM learns from such recipe data. Using such a recipe DB and SLM, it is possible to generate expanded and strengthened input prompts in response to user request information, and by using the input prompts generated in this way, it is possible to optimize the output of the LLM so as not to output recipe information that includes qualitatively or quantitatively inappropriate ingredients, seasonings, cooking methods, etc. For example, it is possible to prevent the output of recipe information that includes ingredients or seasonings that are difficult to digest, cannot be eaten or are not suitable for eating, or whose safety has not been confirmed, such as those with too high a concentration of spicy ingredients.

[0042] However, the format of data input to the SLM, the method of generating input data to the SLM from related recipe information obtained from the recipe DB, and the method of generating an input prompt for the LLM using the SLM are not limited. Related recipe information obtained from the recipe DB may be input directly to the SLM, or output data from the SLM may be used directly as an input prompt for the LLM. Furthermore, user request information and related recipe information obtained from the recipe DB may be input to the SLM, or an input prompt for the LLM may be generated from one or both of the user request information and related recipe information obtained from the recipe DB and the output data from the SLM.

[0043] FIG. 3 is a diagram conceptually illustrating an example of tuning a recipe database (DB). 3, the recipe DB 221 is constructed as a graph database using a property graph, and recipe data related to food and drink is stored in a structure including each element (circle in FIG. 3), the relationship between the elements (solid line in FIG. 3), and the properties assigned between the elements. However, the method of constructing the recipe DB is not limited to a graph database, and other databases such as a vector database may also be used.

[0044] As shown in FIG. 3, the recipe DB 221 may be tuned based on the evaluation information of the output recipe information (hereinafter referred to as recipe evaluation information). In the example of FIG. 3, properties of the recipe DB 221 are updated based on the output from a preference-specific optimization model (hereinafter sometimes abbreviated as optimization model) 222 based on the recipe evaluation information. For example, if the recipe evaluation information includes information such as "it was a little salty," the property between the element "taste" and the element "saltiness" is increased by 0.05, updating it from 0.5 to 0.55. If the recipe evaluation information includes information such as "it would have been better if it had more bite," the property between the element "taste" and the element "texture" is decreased by 0.1, updating it from 0.3 to 0.2. In such a case, the property between the element "taste" and the element "sweetness," which is unrelated to the acquired recipe evaluation information, is not changed.

[0045] The recipe evaluation information is information indicating the evaluation of one or more evaluation items among the multifaceted evaluation items of the food and drink created by the recipe of the output recipe information. The recipe evaluation information is input by the user to the recipe information displayed on the user terminal 15 in conjunction with processing by the display processing unit 25 (described later), and is acquired by the operation processing unit 26 (described later).

[0046] The optimization model 222 is a trained model used to update the properties of the recipe DB 221 based on recipe evaluation information. The definition of this "trained model" will be described later, but the optimization model 222 is constructed using training data that indicates the relationship between each evaluation indicated by a large amount of recipe evaluation information and the increase / decrease value or property value of the property in the recipe DB 221. As a result, the optimization model 222 outputs the property value or increase / decrease value corresponding to the evaluation indicated by the recipe evaluation information for the property corresponding to the evaluation item indicated by the recipe evaluation information in the recipe DB 221.

[0047] The recipe generation unit 22 updates the target property in the recipe DB 221 with the property value or increase / decrease value output from this optimization model 222. When the recipe generation unit 22 updates the property in the recipe DB 221, the recipe generation unit 22 may regenerate an input prompt for the LLM using the updated recipe DB 221, causing the LLM to regenerate recipe information.

[0048] In this way, the properties of the recipe DB 221 are updated based on the recipe evaluation information, thereby optimizing the recipe data stored in the recipe DB 221 to suit the user's preferences. Then, the recipe DB 221 is used to expand and strengthen the input prompts of the LLM, so that the recipe information output from the generation AI can be adapted to suit the user's preferences. Furthermore, by providing and tuning the recipe DB 221 for each user, it becomes possible to output personalized recipe information. The same is true for the SLM. That is, the SLM may also be provided for each user and trained using training data that matches the preferences of each user.

[0049] The recipe DB 221 may be stored in the evaluation device 10, or may be stored on another computer accessible via communication by the evaluation device 10. Similarly, the optimization model 222 and the SLM may be stored operably on the evaluation device 10, or may be stored on the other computer.

[0050] Furthermore, the recipe generation unit 22 may regenerate the recipe information and review information by inputting data into the generation AI that adds additional conditions for the dish, such as food ingredients, seasonings, flavors, nutrients, functional ingredients, etc., to the recipe information output from the generation AI. In this case, the recipe generation unit 22 may display the original recipe information to the user by having the display processing unit 25, which will be described later, display the information, and then acquire the additional condition information for the dish input by the user.

[0051] Furthermore, the recipe generation unit 22 may generate, as part of the recipe information, images such as diagrams showing the cooking process, images showing the state of food and drink during the cooking process, and examples of presentation. The recipe generation unit 22 can also generate audio, explanatory text, etc. instead of or in addition to these diagrams, images, and images. Such supplemental information may be extracted from the recipe DB 221, or may be generated using a separate generation AI.

[0052] The estimation unit 23 acquires specific information from the recipe information acquired by the recipe generation unit 22, and estimates the evaluation values ​​of the multifaceted evaluation items of the food and beverage created using the target recipe based on this specific information and the review information acquired by the recipe generation unit 22. The estimation unit 23 may acquire the recipe information output from the generation AI as the specific information as is, or may acquire a predetermined portion of the recipe information (dish name, ingredients, seasonings, cooking steps, etc.) as the specific information.

[0053] Furthermore, the estimation unit 23 may use the review information output from the generation AI as is to estimate the evaluation value, or may distinguish between qualitative information and quantitative information extracted from the review information and use them. Furthermore, the estimation unit 23 may search for review information corresponding to the specific information from other sites (including blogs, etc.) that were not adopted as training data for the generation AI, add the acquired review information to the review information output from the generation AI, and use it to estimate the evaluation value. Therefore, the estimation unit 23 corresponds to an acquisition means for acquiring review information and specific information and an estimation means for estimating an evaluation value. The "evaluation values ​​of multifaceted evaluation items of food and beverages" estimated by the estimation unit 23 have been outlined above, but in this embodiment, for example, evaluation values ​​of evaluation items are estimated according to the content of the acquired review information and specific information.

[0054] In this embodiment, the estimation unit 23 estimates the evaluation values ​​of the multifaceted evaluation items related to the dish cooked using the target recipe, using a trained model that has been trained in advance using a group of training data. The "trained model" here may be one or more models that can input review information and specific information and output evaluation values ​​for multifaceted evaluation items, and may include a regression formula obtained by regression analysis, a large-scale language model, or a learning model obtained by classification in supervised learning or deep learning, etc., and the data structure and learning algorithm of the model are not limited. For example, the trained model is realized by a combination of a computer program and parameters, a combination of multiple functions and parameters, etc. When the trained model is constructed using a neural network and the input layer, intermediate layer, and output layer are considered as units of one neural network, the trained model may refer to one neural network or a combination of multiple neural networks. Furthermore, the trained model may be composed of a combination of multiple multiple regression equations, or may be composed of one multiple regression equation. The trained model may be stored in the memory 2 of the evaluation device 10, or may be stored in the memory of another computer that the evaluation device 10 can access via communication.

[0055] For example, the estimation unit 23 estimates the evaluation value by using the generation AI (large-scale language model) used in the recipe generation unit 22 and other trained models. Specifically, the estimation unit 23 acquires review information corresponding to the specific information by inputting the specific information into the generation AI, and acquires evaluation values ​​for the multifaceted evaluation items by inputting the review information acquired together with the recipe information as described above and the review information corresponding to the specific information into the other trained models. The other trained model is constructed to input review information in a written format or qualitative information and quantitative information extracted from the review information, and output evaluation values ​​for multifaceted evaluation items. For example, the other trained model can apply natural language processing to the review information to identify qualitative information and quantitative information according to the meaning of each word, and output evaluation items corresponding to the identified qualitative information and evaluation values ​​corresponding to the identified quantitative information.

[0056] As a specific example, specific information such as "Tell me the taste of curry with a lot of onions and garlic" is input into the generation AI, and the generation AI outputs review information corresponding to the specific information, such as "Curry with a lot of onions and garlic has a deeper, richer flavor. Onions add sweetness, while garlic enhances the flavor. Adding too much onion can make the curry too sweet, but adding the right amount will give the curry a deeper flavor. Adding the right amount of garlic will also enrich the curry's flavor." Review information corresponding to such specific information and review information acquired together with recipe information are input into the other trained model, and evaluation values ​​for the multifaceted evaluation items are output.

[0057] However, the trained model used by the estimation unit 23 is not limited to the above example. When all of the recipe information or various types of information within the recipe information are considered to be specific information, the trained model may be configured to input the specific information and output evaluation values ​​for the multifaceted evaluation items related to the target recipe corresponding to the specific information. In this case, the training data group used to train the trained model includes multiple recipe information and evaluation values ​​for the multifaceted evaluation items related to each recipe information. In this case, the evaluation values ​​of the evaluation items of taste, texture, appearance, temperature, and sound during eating and drinking, which are included in the training data, can be obtained by sensory evaluation or evaluation tests using measuring equipment, etc., and the evaluation values ​​of the functional evaluation items can be obtained by investigating the ingredients of ingredients or evaluation tests using measuring equipment, etc. Furthermore, the sensory evaluation items, preference evaluation items, and combination evaluation items after eating and drinking, which are included in the training data, can be obtained by evaluation using a questionnaire, etc. In this way, the training data group used for training the trained model includes evaluation values ​​for multifaceted evaluation items obtained by actually tasting each food or drink prepared using the recipe of each recipe information, such as sensory evaluation, evaluation using a questionnaire, etc. In particular, it is difficult to obtain evaluation values ​​for the sensory evaluation items, preference evaluation items, and combination evaluation items after eating without actually tasting the food or drink, such as evaluation using a questionnaire. By using a group of training data that includes evaluation values ​​that are difficult to obtain without actually eating the food, it becomes possible to estimate evaluation values ​​for evaluation items for which it is difficult to obtain evaluation results.

[0058] Furthermore, if the specific information is unstructured data such as text-format character string data, the estimation unit 23 may convert the specific information into structured data and input it to the AI ​​model. In this case, the AI ​​model is constructed so as to input the structured data of the specific information (hereinafter referred to as recipe structured data) and output evaluation values ​​of the multifaceted evaluation items.

[0059] FIG. 4 is a diagram conceptually illustrating an example of structured recipe data. 4, the recipe structured data is data in which various pieces of information included in the specific information are associated with each other by meaning and structured. For example, the estimation unit 23 can convert the specific information into recipe structured data structured according to the meaning of each word by applying natural language processing to the specific information. It is not necessary for all of the information items shown in FIG. 4 to be included in the specific information, and it is preferable that the specific information include at least information on the preparation method (cooking method) and ingredients to be prepared.

[0060] The display processing unit 25 displays a screen serving as a user interface of the evaluation device 10 on the user terminal 15 or the display device 5. The display processing by the display processing unit 25 may be executed in roughly the same way on both the user terminal 15 and the display device 5, and therefore, for ease of understanding, the display processing for the user terminal 15 will be described here as an example. Furthermore, the display processing unit 25 may cause a printing device (not shown) to print a report containing the same content as that displayed on the screen.

[0061] The display processing unit 25 displays character string data in a description format corresponding to the evaluation values ​​of the multifaceted evaluation items estimated by the estimation unit 23 on the user terminal 15. The display processing unit 25 may display character string data in a sentence format, a list format, or the like as a description format. For example, the display processing unit 25 displays character string data such as "slightly less sweet," "normally salty," "strongly sour," "shiny," "chewy," and "feels nostalgic." By displaying the evaluation results of the multiple evaluation items as character string data in this way, the evaluation results can be presented in a way that is easy for the user to understand.

[0062] The display processing unit 25 may hold a data table in which evaluation values ​​and character string data are associated with each other for each evaluation item, and may use this data table to convert the evaluation values ​​of the multifaceted evaluation items into character string data. The display processing unit 25 may also use a generation AI capable of language generation to acquire character string data in a description format corresponding to the evaluation values ​​of the multifaceted evaluation items. In this way, the method for acquiring character string data in a description format corresponding to the evaluation values ​​of the multifaceted evaluation items is not limited in any way.

[0063] Furthermore, there are no limitations on the display format of the evaluation results of the multifaceted evaluation items by the display processing unit 25. The display processing unit 25 may further display the evaluation values ​​of the multifaceted evaluation items in radar chart format on the user terminal 15. In this embodiment, since there are a large number of evaluation items, multiple radar charts may be displayed for each type of evaluation item, or the evaluation values ​​of some of the multifaceted evaluation items selected by the user from the multifaceted evaluation items may be displayed in radar chart format.

[0064] Furthermore, the display processing unit 25 can also display on the user terminal 15 the recipe information generated by the recipe generation unit 22 together with information (evaluation results) corresponding to the evaluation values ​​of the multifaceted evaluation items estimated by the estimation unit 23. When the recipe information includes supplemental information such as a diagram showing the cooking process, an image showing the state of food and drink during the cooking process, or an image of an example of presentation, the display processing unit 25 can also display or output the supplemental information as audio on the user terminal 15.

[0065] Furthermore, when evaluation values ​​over time have been output for at least some of the multifaceted evaluation items, the display processing unit 25 can also display on the user terminal 15 an animated chart or video (referred to as an evaluation change over time video) that dynamically shows the change in evaluation over time during eating and drinking that corresponds to the evaluation values ​​over time. Such an evaluation change over time video shows how the evaluation value changes over time for each evaluation item indicated by the evaluation values ​​over time, and thereby presents how the evaluation value for each evaluation item changes over time, for example, the sourness is strong the moment the food is put in the mouth, and as chewing progresses, the spiciness increases in addition to the sourness, and as chewing progresses further, the umami also increases.

[0066] The evaluation time-varying video may be generated by separately using a generation AI and inputting information based on the time-varying evaluation values ​​to the generation AI. The evaluation time-varying video is acquired by the estimation unit 23 or the display processing unit 25.

[0067] In addition, the display processing unit 25 may be configured to display on the user terminal 15 factor information that has a causal or correlational relationship with at least some of the evaluation items in the multifaceted evaluation items, and that includes at least some of the creation steps (cooking steps) and preparation items indicated in the recipe information acquired by the recipe generation unit 22.

[0068] The display processing unit 25 can acquire the factor information using causal inference, which is a method for statistically extracting causal relationships from data. For example, the display processing unit 25 performs causal inference using some evaluation items (e.g., full-bodied taste) among the multifaceted evaluation items as the result (objective) variable and at least some of the cooking steps and preparations (e.g., seasonings, ingredients, etc.) indicated in the recipe information as explanatory (processing) variables. The display processing unit 25 identifies, as the factor information, processing variables (e.g., seasonings and ingredients such as garlic and sauce) that are causally related to the evaluation item (full-bodied taste) as the result variable, as well as manipulated variables and confounding factors (e.g., seasonings and ingredients such as yogurt) that affect each processing variable. Causal inference can be performed using various existing tools such as DirectLiNGAM and NOTEARS, or, in the case of time-series causal inference, VAR-LiNGAM and CCM (Convergent Cross Mapping), etc.

[0069] The display processing unit 25 can also acquire the factor information using a feature selection method based on correlation. For example, the display processing unit 25 performs feature importance calculations using some evaluation items (e.g., body flavor) among the multifaceted evaluation items as the objective variable and at least some of the cooking steps and preparations (e.g., seasonings, ingredients, etc.) indicated in the recipe information as the explanatory variables, and identifies important factors (e.g., garlic, etc.) for the objective variable (body flavor) by calculating the feature importance of each explanatory variable. As such a feature selection method, various existing methods such as null importance using a model (learner) such as Random Forest or LightBGM can be used.

[0070] The display processing unit 25 may display the factor information for all evaluation items for which evaluation values ​​have been estimated on the user terminal 15, or may display the factor information for evaluation items selected by user operation or the like on the user terminal 15. For example, for the evaluation item of "rich taste," prepared ingredients included in the recipe information, such as "garlic," "yogurt," and "sauce," which are calculated to have a causal or correlation relationship with rich taste, as well as a cooking step included in the recipe information, such as "heating," are displayed as the factor information. The displayed factor information only needs to include at least some of the creation steps (cooking steps) and preparations indicated in the recipe information as factors, and may also include information not included in the recipe information (preparation items, etc.) as factors.

[0071] Furthermore, the display processing unit 25 can also display information indicating a causal relationship or a correlation between each factor included in the above-mentioned factor information and at least some of the evaluation items in the multifaceted evaluation items on the user terminal 15. The form of presentation of the causal relationship or the correlation is not limited. For example, the display processing unit 25 may display, in a table format, a list of factor information that is causally or correlatively related to each evaluation item. In this case, the strength of the causal relationship or the importance of the correlation (feature importance) can also be associated with each factor and displayed.

[0072] FIG. 5 is a diagram showing an example of a display of the causal relationship between kokumi as an evaluation item and each factor. As shown in Fig. 5, the display processing unit 25 may display a network diagram of a directed graph as a form of presenting the causal relationships. Although not shown in Fig. 5, the presence or absence of a preparation in stock (surplus information) may be displayed as an operational variable for each preparation as a factor. For example, "garlic in stock" may be displayed by connecting the "garlic" factor with an arrow.

[0073] In this way, by displaying factor information that has a causal or correlative relationship with at least some of the evaluation items in the multifaceted evaluation items and that includes at least some of the creation steps (cooking steps) and preparations indicated in the recipe information, or by indicating the causal or correlative relationship, the reliability of the evaluation values ​​of the multifaceted evaluation items can be clearly presented to the user. Furthermore, by allowing the user to understand the relationship between the cooking steps and preparations included in the recipe information and each evaluation item of the dish cooked therefrom, the user's interest in cooking can be increased.

[0074] The operation processing unit 26 performs processing related to user operations on a screen displayed on the user terminal 15 or the display device 5. The operation processing by the operation processing unit 26 may be executed in roughly the same way on both the screens displayed on the user terminal 15 and the display device 5, and therefore, for ease of understanding, the processing on a screen displayed on the user terminal 15 will be described here as an example. When user request information is generated based on information input by the user via an input screen displayed on the user terminal 15, the operation processing unit 26 acquires the information input by the user and sends it to the request acquisition unit 21. In addition, when the user request information itself is input by the user by operating the user terminal 15, the operation processing unit 26 acquires the input user request information and sends it to the request acquisition unit 21. In addition, when recipe information displayed on the user terminal 15 is modified by user operation or when additional condition information is added to the recipe information, the operation processing unit 26 acquires the modified information or added information in response to the user operation.

[0075] The operation processing unit 26 acquires information on evaluation items selected by the user from among the multifaceted evaluation items. For example, the operation processing unit 26 acquires information on evaluation items selected by a user operation on a screen showing all evaluation items displayed on the user terminal 15. Therefore, the operation processing unit 26 can correspond to a selection processing means. Based on the information thus acquired by the operation processing unit 26, the display processing unit 25 displays information on the creation steps (cooking steps) and preparations indicated in the recipe information that are causally or correlatively related to the evaluation item selected by the user on the user terminal 15. The display processing unit 25 can also selectively display the evaluation values ​​of the evaluation items selected by the user in a radar chart format.

[0076] The change processing unit 27 changes the recipe information in accordance with a change operation by the user. Specifically, the display processing unit 25 displays the preparations indicated in the recipe information on the user terminal 15, and when the user performs a change operation on the display of the preparations, the operation processing unit 26 acquires change information on the preparations caused by the change operation, and the change processing unit 27 changes the recipe information acquired by the recipe generation unit 22 based on the change information. In this case, the estimation unit 23 acquires the specific information (referred to as "change specific information") from the changed recipe information, and based on this change specific information, further estimates the evaluation values ​​of the multifaceted evaluation items of the food and drink created using the changed recipe indicated by the changed recipe information. In this case, the estimation unit 23 may input the change specific information into the generation AI to acquire review information (change review information) corresponding to the change specific information, and input this change review information into the trained model to estimate the evaluation values ​​of the multifaceted evaluation items of the food and drink created using the changed recipe. The display processing unit 25 causes the user terminal 15 to display information corresponding to the evaluation values ​​of the multifaceted evaluation items before and after the change, output based on the recipe information before and after the change, so that they can be compared. For example, the display processing unit 25 may display radar charts corresponding to the evaluation values ​​of the multifaceted evaluation items before and after the change so that they can be compared, or may display descriptive character string data corresponding to the evaluation values ​​of the multifaceted evaluation items before and after the change so that they can be compared. Alternatively, the display processing unit 25 may display only the evaluation values ​​of evaluation items that differ between before and after the change so that they can be compared. Such display formats are not limited.

[0077] In this way, recipe information generated based on user request information is presented to the user in a modifiable manner, and information corresponding to the evaluation values ​​of the multifaceted evaluation items for the recipe information before and after the change is displayed so that they can be compared with each other. This allows the user to obtain a better recipe by modifying the recipe information while referring to the information corresponding to the evaluation values ​​of the multifaceted evaluation items.

[0078] <Recipe evaluation method> Hereinafter, the recipe evaluation method according to this embodiment (the evaluation method) will be described in detail with reference to FIG. FIG. 6 is a flowchart showing the recipe evaluation method (the present evaluation method) according to this embodiment. This evaluation method is executed by one or more computers such as the above-described evaluation device 10. This evaluation device 10 can execute this evaluation method by having the CPU 1 execute a computer program stored in the memory 2. It can also be expressed as the CPU 1 executing this evaluation method through the execution of the computer program stored in the memory 2. This computer program is installed from a portable recording medium such as a CD (Compact Disc) or a memory card, or from another computer on a network, via the input / output I / F 3 or the communication unit 4, and stored in the memory 2.

[0079] As shown in FIG. 6, this evaluation method includes steps (S51) to (S59). In the following, the evaluation device 10 will be described as the entity that executes each step, but each step can also be described as being executed by one or more computers or CPUs, or by each of the above-mentioned processing modules of the evaluation device 10. Since each step is similar to the processing content of each processing module of the evaluation device 10, details of each step will be omitted as appropriate.

[0080] In step (S51), the evaluation device 10 (request acquisition unit 21) acquires user request information. For example, the evaluation device 10 displays an input screen on the user terminal 15 and acquires user request information based on information input by the user on the input screen. As described above, the user request information includes condition information for the food and drink desired by the user, and is composed of character string data such as, for example, "Please tell me a recipe for a rich curry with plenty of vegetables that even children can eat." The user request information may also include cooking condition information such as "time-saving cooking." However, the method of acquiring the user request information is not limited in any way.

[0081] In step (S52), the evaluation device 10 (recipe generation unit 22) uses a large-scale language model to generate recipe information and review information based on the user request information acquired in step (S51). The generated recipe information is as described above, and is output as text-format character string data including information on cooking methods, preparations, and cooking notes, for example. The large-scale language model in this embodiment is constructed by fine-tuning using a large number of recipe information groups stored on multiple recipe information providers and review information groups for each recipe as training data, so that review information can be generated in addition to recipe information. The generated review information indicates criticism of the recipe indicated by the generated recipe information and is output, for example, as character string data in a sentence format. Furthermore, as described above, the large-scale language model may be stored in an operable state on the evaluation device 10, or may be stored in an operable state on another computer that the evaluation device 10 can access via communication.

[0082] In step (S52), the evaluation device 10 may apply a filtering process to the generated recipe information or review information to remove noise data from the recipe information or review information, or may re-input data with additional conditions added to the generated recipe information into the large-scale language model to regenerate the recipe information and review information.

[0083] In step (S53), the evaluation device 10 (estimation unit 23) acquires the specific information from the recipe information generated in step (S52). The evaluation device 10 may acquire the recipe information generated in step (S52) as the specific information as is, or may acquire a predetermined part of the recipe information (such as the name of the dish, ingredients, seasonings, cooking steps, etc.) as the specific information.

[0084] In step (S54), the evaluation device 10 (estimation unit 23) estimates evaluation values ​​of multifaceted evaluation items for a food or drink prepared using a recipe corresponding to the specific information, based on the review information generated in step (S52) and the specific information acquired in step (S53). As described above, the estimation process in step (S54) utilizes a trained model that has been trained in advance using a group of training data. This trained model may also be stored in memory 2 within the evaluation device 10, or may be stored in the memory of another computer that the evaluation device 10 can access via communication.

[0085] The trained model used in step (S54) may be composed of two types of models: a first type model that estimates an evaluation value from specific information and a second type model that estimates an evaluation value from review information. In this case, the evaluation device 10 may estimate evaluation values ​​for the multifaceted evaluation items by combining an evaluation value output from a first type trained model obtained by inputting the specific information into the first type trained model and an evaluation value output from a second type trained model obtained by inputting review information into the second type trained model. Furthermore, the trained model used in step (S54) may be composed of two types of models: a first type model that outputs review information from specific information, and a second type model that estimates an evaluation value from the review information. In this case, the evaluation device 10 may input the specific information into the trained model of the first type, acquire review information corresponding to the specific information from the trained model, and input the review information corresponding to the specific information and the review information generated in step (S52) into the trained model of the second type, thereby estimating the evaluation value of the multi-faceted evaluation item. If the recipe information generated in step (S52) is unstructured data such as text-format character string data, the evaluation device 10 may extract specific information from the recipe information, convert this specific information into structured data (see Figure 4), and then input it into the trained model.

[0086] In step (S55), the evaluation device 10 (display processing unit 25) displays on the user terminal 15 the recipe information generated in step (S52) and information (evaluation results) corresponding to the evaluation values ​​of the multifaceted evaluation items estimated in step (S54). The display screen at this time may be configured to present the recipe information and a radar chart corresponding to the evaluation values ​​of the multifaceted evaluation items side by side, or may be configured to present the recipe information and the radar chart in a switchable manner. Furthermore, multiple radar charts may be displayed for each type of evaluation item, or the evaluation values ​​of some of the multifaceted evaluation items selected by the user may be displayed in radar chart format. Furthermore, as evaluation result information, descriptive character string data corresponding to the evaluation values ​​of the multifaceted evaluation items may be presented together with or instead of the radar chart. The configuration of such a display screen is not limited in any way.

[0087] Furthermore, if the recipe information includes supplementary information such as diagrams showing the cooking process, images showing the state of food and drink during the cooking process, and images of sample presentations, the present evaluation device 10 (display processing unit 25) can display or output audibly this supplementary information on the user terminal 15. Furthermore, if time-dependent evaluation values ​​are output for at least some of the multifaceted evaluation items, the present evaluation device 10 (display processing unit 25) can also display on the user terminal 15 a time-dependent evaluation change video that dynamically shows the change in evaluation over time during eating and drinking that corresponds to the time-dependent evaluation value.

[0088] In this way, through steps (S51) to (S55), the user can not only obtain recipe information but also check evaluation information regarding the food and drink made with the recipe by providing desired information (condition information) for the food and drink they desire via user terminal 15. Furthermore, the evaluation information presented in this embodiment is not only specific information obtained from the recipe information, but also evaluation values ​​for multifaceted evaluation items based on review information generated based on user reviews for each recipe stored on the recipe information providing site, and is therefore very useful information when the user selects and rejects recipe information.

[0089] 6, the evaluation device 10 is capable of performing a deeper analysis (represented as recipe information optimization analysis in FIG. 6) of the causal relationship or correlation between recipe information and the evaluation result of that recipe, and allows the user to select whether or not the recipe information optimization analysis is required. For example, a menu that allows the user to select whether or not the recipe information optimization analysis is required is provided on the screen displayed in step (S55).

[0090] When the user selects to perform recipe information optimization analysis (S56; YES), the present evaluation device 10 (display processing unit 25) extracts factor information indicating a causal relationship or correlation with some of the evaluation items among the multifaceted evaluation items using existing analysis methods such as causal inference, feature selection methods, etc. Then, the present evaluation device 10 displays information indicating the causal relationship or correlation between each factor of the factor information and the evaluation item on the user terminal 15 (S57). 5 on the user terminal 15, or may display a list of factor information that is causally or correlatively related to the evaluation items together with feature importance in a table format on the user terminal 15. Furthermore, the factor information may display only information indicated in the recipe information (production steps, preparations, etc.) as factors, or may display information not included in the recipe information as factors.

[0091] By displaying the results of such causal or correlation analysis, it is possible to demonstrate the reliability of the evaluation results displayed in step (S55), and to clearly show the user the relationship between the preparation steps and ingredients included in the recipe information and each evaluation item, which makes it easier for the user to judge the quality of the recipe information presented in step (S55), and also leads to increasing the user's interest in cooking. Furthermore, when information not included in the recipe information is presented as a factor, such factor information can make the user aware of ingredients, seasonings, or cooking steps that are missing from the presented recipe information. However, a user who does not require such detailed analysis can select not to perform recipe information optimization analysis (S56; NO).

[0092] 6, the evaluation device 10 provides a menu on the screen displayed in step S55 that allows the user to change the recipe information. When the user performs an operation to change the recipe in accordance with the menu (S58; YES), the evaluation device 10 changes the recipe information generated in step S52 in response to the user operation (S59). This change may be to replace or delete some of the preparations indicated in the recipe information, to add new preparations (ingredients, seasonings, etc.), or to change the cooking steps. The user can also change the recipe information by referring to the factor information displayed in step (S57). For example, the original recipe information did not include "garlic" as a preparation, but by referring to the fact that "garlic" was presented in step (S57) as a factor that has a causal or correlative relationship with the evaluation item of "rich flavor," the user performs a change operation to add garlic as a preparation.

[0093] When the recipe information is changed in step (S59), the evaluation device 10 acquires change specification information from the changed recipe information and re-estimates the evaluation values ​​of the multifaceted evaluation items based on the change specification information (S54). In estimating the evaluation values ​​based on the change specification information, a trained model of a type that estimates evaluation values ​​from specification information may be used, or a set of a trained model of a type that outputs review information from specification information and a trained model of a type that estimates evaluation values ​​from review information may be used. Thereafter, the process from step (S55) onward is executed in the same manner as described above. In this case, in step (S55), the present evaluation device 10 may cause the user terminal 15 to display information corresponding to the evaluation values ​​of the multifaceted evaluation items before and after the change, which are output based on the recipe information before and after the change, so that the information can be compared.

[0094] By comparing the evaluation results of the multiple evaluation items before and after the change in this display, the user can easily understand how the evaluation results have changed due to the recipe change and can judge whether the change was good or bad. Furthermore, the user can obtain a better recipe by changing the recipe information while referring to the information corresponding to the evaluation values ​​of the multiple evaluation items.

[0095] In addition, in a mode in which the recipe information output from the LLM is optimized by expanding and enhancing the input prompt, the evaluation method may include steps as shown in FIG. Fig. 7 is a flowchart showing a specific example of a recipe information generation process in the recipe evaluation method (this evaluation method) according to this embodiment. In Fig. 7, the same reference numerals (S51 and S52) as in Fig. 6 are used for the processes shown in Fig. 6, and any of the processes shown in Fig. 6 may be executed as the processes shown in dashed lines.

[0096] 7, the evaluation device 10 will be described as the entity that executes each step shown in Fig. 7, but each step can also be described as being executed by one or more computers or CPUs, or each of the above-mentioned processing modules of the evaluation device 10. Since each step is similar to the processing content of each processing module of the evaluation device 10, details of each step will be omitted as appropriate.

[0097] In step (S71), the evaluation device 10 (recipe generation unit 22) queries the recipe DB 221 based on the user request information acquired in step (S51). For example, if the user request information is character string data in a sentence format, the evaluation device 10 extracts a search word from the user request information and sends a search request to the recipe DB 221 using the search word. As a result, in step (S72), the evaluation device 10 acquires from the recipe DB 221 recipe information that is highly relevant to the user's request (related recipe information).

[0098] In step (S73), the evaluation device 10 provides input data to the SLM based on the related recipe information acquired in step (S72). The input data provided to the SLM may be the related recipe information acquired in step (S72) itself, information extracted from the related recipe information, or information to which some information has been added to the related recipe information. In addition, the user request information acquired in step (S51) and the related recipe information acquired in step (S72) may be input to the SLM.

[0099] In step (S74), the evaluation device 10 generates an input prompt for the LLM. Here, the SLM may be configured to output the input prompt, or the input prompt for the LLM may be generated from one or both of the user request information (S51) and the related recipe information (S72) and output data from the SLM. The evaluation device 10 provides the input prompt acquired in step (S74) to the LLM, whereby recipe information and review information are generated by the LLM (S52).

[0100] The processing flow after step (S52) may be the processing flow after step (S53) shown in FIG. 6, and steps (S75), (S76), and (S77) are executed somewhere in that processing flow. After the recipe information is displayed in step (S55), if the evaluation device 10 (recipe generation unit 22) acquires information indicating the evaluation of one or more evaluation items among the multifaceted evaluation items of the food or drink created by the recipe of the recipe information, i.e., recipe evaluation information (S75; YES), it performs tuning of the recipe DB 221 (S76). The recipe evaluation information is acquired by input by the user for the recipe information displayed on the user terminal 15. In step (S76), the evaluation device 10 inputs the acquired recipe evaluation information into the optimization model 222, thereby acquiring the property value or increase / decrease value of the recipe DB 221 corresponding to the recipe evaluation information from the optimization model 222, and thereby updating the property of the recipe DB 221.

[0101] The evaluation device 10 (recipe generation unit 22) updates the properties of the recipe DB 221, and if the recipe information is to be regenerated (S77; YES), the process is repeated from the step (S71) of making an inquiry to the updated recipe DB 221, and the recipe information and review information are regenerated (S52). If the recipe information is not to be regenerated (S77; NO), the step (S55) in Fig. 6 may be executed, or the process may be ended.

[0102] [Variations] The above-described embodiment is merely an example, and may be partially modified as appropriate. For example, in the above embodiment, a so-called server-client configuration has been exemplified in which the evaluation device 10 displays a screen serving as a user interface on the user terminal 15, but the evaluation device 10 may also be configured to use the display device 5 and the input device 6 as a user interface. The evaluation device 10 may be a stationary computer or a portable computer.

[0103] 6 and 7 show a plurality of steps in sequence, but the execution order of each step is not limited to the illustrated example and can be changed as appropriate without departing from the spirit of the invention. Furthermore, the present evaluation method may be configured to exclude steps (S56) to (S59), or may be configured to exclude steps (S51) and (S52) in addition to these steps. In the latter case, the present evaluation device 10 may acquire specific information and review information from another computer, or may acquire specific information and review information in response to user input.

[0104] The software configuration of the evaluation device 10 in the above embodiment can also be modified as shown in FIG. 8 is a diagram conceptually illustrating an example of the software configuration of the recipe evaluation device 10 according to the modified example. In addition to the configuration of the above-described embodiment, the recipe evaluation device 10 according to the modified example further includes an ingredient information acquisition unit 61, an evaluation history acquisition unit 62, and a related information provision unit 63. The ingredient information acquisition unit 61, the evaluation history acquisition unit 62, and the related information provision unit 63 may be realized in the same manner as other processing modules.

[0105] The ingredient information acquisition unit 61 acquires ingredient supplement information including either or both of the use history information of the food and drink ingredients and the surplus information of the food and drink ingredients. For example, the ingredient information acquisition unit 61 can acquire the use history information of food ingredients by extracting ingredient information (preparation information) included in each recipe information from a database that stores the history of recipe information used by the user. This database may be stored in the evaluation device 10 or another server device. Information on surplus food ingredients may be obtained from a refrigerator system that can manage ingredients and other items stored in the refrigerator, or from information on food ingredients held by nearby residents or nearby stores that is stored on SNS (Social Networking Service) or a system used for sharing information among nearby residents, or from sale information provided by a system managed by nearby stores.

[0106] The evaluation history acquisition unit 62 acquires history information of the evaluation values ​​of the multifaceted evaluation items estimated by the estimation unit 23 as evaluation history information. The present evaluation device 10 or another server device holds a database that stores the evaluation values ​​of the multifaceted evaluation items estimated by the estimation unit 23 in chronological order, and the evaluation history acquisition unit 62 acquires the evaluation history information from this database.

[0107] In this modification, the recipe generation unit 22 generates recipe information based on the user request information and the supplemental ingredient information acquired by the ingredient information acquisition unit 61. Specifically, the recipe generation unit 22 generates recipe information by inputting the user request information and the supplemental ingredient information into a large-scale language model. For example, the supplemental ingredient information may be input to the large-scale language model as character string data such as "If possible, please use the food and drink ingredient of . . . " in addition to the user request information. Furthermore, when supplementary ingredient information including the use history information of food and beverage ingredients is acquired, the supplementary ingredient information may be input to the large-scale language model as character string data based on the time series of use of food and beverage ingredients, such as "Please do not overlap with food and beverage ingredients used within the last 3 days (within 5 days, within 1 week, etc.), in addition to the user's request information. This may be a recipe use history or a history of menu names of dishes and beverages, rather than a use history of food and beverage ingredients.

[0108] By generating recipe information using such supplementary ingredient information, surplus ingredients can be effectively utilized, and waste of such surplus ingredients can be reduced. In addition, by generating recipe information using the user's ingredient usage history, recipe information that matches the user's preferences can be generated.

[0109] In addition, the recipe generation unit 22 in this modified example can also generate recipe information that takes into account the use of food ingredients related to changes in the evaluation value of a certain multifaceted evaluation item, based on user request information, supplementary ingredient information (information on the usage history of food ingredients), and the above-mentioned evaluation history information. Specifically, the recipe generation unit 22 identifies a certain period during which the evaluation value improved or deteriorated based on the history of evaluation values ​​for a specific preference evaluation item (e.g., health status, sleep status, beauty status, mental and physical state, etc.) indicated in the evaluation history information, and identifies the food and beverage ingredients ingested during that identified period based on the history information on food and beverage ingredients. At this time, the recipe generation unit 22 may identify nutrients or food components related to (that may affect) the change in the evaluation value for the specific preference evaluation item from the information on the food and beverage ingredients ingested during the period during which the evaluation value improved or deteriorated, and identify food and beverage ingredients containing the identified nutrients or food components. The recipe generation unit 22 then generates recipes that utilize, as much as possible, or increase the amount of food and beverage ingredients identified for the improvement in the evaluation value, and generate recipes that eliminate, as much as possible, or reduce the amount of food and beverage ingredients identified for the deterioration in the evaluation value. Examples of improvements in the evaluation value of the specific preference evaluation item include improvement in constipation, improvement in skin firmness, improvement in irritability, improvement in ability to fall asleep, etc. Examples of deterioration in the evaluation value of the specific preference evaluation item include becoming constipated, losing skin firmness, feeling irritable, difficulty falling asleep, etc. However, examples of improvements and deteriorations in the evaluation value are not limited to these examples.

[0110] By generating recipe information using such evaluation history information, it is possible to appropriately support changes in health condition, sleep condition, beauty condition, mental and physical condition, etc. caused by each individual's diet.

[0111] In addition to the user request information, supplementary ingredient information, and evaluation history information, the recipe generation unit 22 may also use information about the target diners and other surrounding information to generate recipe information. Examples of this diners' information include age, sex, physique, constitution, preferences, habits, health status including physical condition, chronic illnesses, and medical history, cosmetic condition of skin, hair, etc., mental state, sleep state, and exercise state. Examples of other surrounding information include information about weather, season, location, and trends.

[0112] The related information providing unit 63 provides recipe-related store site information including one or more of information on a mail-order site for the preparations included in the recipe information, information on the restaurant corresponding to the recipe information, or information on the delivery order site of the restaurant. There are no limitations on the method for identifying such recipe-related store site information corresponding to recipe information. For example, the related information providing unit 63 can identify information on mail-order websites for each of the ingredients, seasonings, cooking utensils, etc. included as preparations in the recipe information by searching the Internet. The related information providing unit 63 also identifies information on restaurants that provide food and drink prepared according to the recipe information or similar food and drink, or information on delivery order websites for such restaurants. This identification may be performed using an Internet search or a proprietary database. Furthermore, the related information providing unit 63 may identify information on restaurants corresponding to the recipe information by using the recipe information and an estimated rating value for the recipe information.

[0113] The related information providing unit 63 may display the recipe-related store site information on the user terminal 15, or may notify the user by email or the like; the method of providing the information is not limited in any way. In this way, according to this modification, the user can obtain not only recipe information and its evaluation information, but also recipe-related store site information corresponding to the recipe information, which can be useful for preparing ingredients, seasonings, etc. Also, by obtaining information about restaurants, the user can eat the food and drink of the recipe information without having to cook it themselves.

[0114] However, the recipe evaluation device 10 according to this modified example may be configured to have the related information providing unit 63 excluding the ingredient information acquiring unit 61 and the evaluation history acquiring unit 62, or may be configured to have the ingredient information acquiring unit 61 and the evaluation history acquiring unit 62 excluding the related information providing unit 63.

[0115] Some or all of the above-described embodiments and modifications can be specified as follows: However, the above-described embodiments and modifications are not limited to the following descriptions.

[0116] <1> an acquisition means for acquiring review information of a food or drink prepared using a target recipe and specific information about the food or drink prepared using the target recipe; an estimation means for estimating evaluation values ​​of multifaceted evaluation items of the food or drink prepared using the target recipe based on the acquired review information and the identification information; A recipe evaluation device comprising: <2> The evaluation values ​​of the multifaceted evaluation items further include a time-course evaluation value indicating an evaluation over time when eating the food or drink prepared according to the target recipe. <1> The recipe evaluation device according to claim 1. <3> a display processing means for displaying, on a display unit, a time-varying evaluation video dynamically showing a change in evaluation over time during eating and drinking corresponding to the time-varying evaluation values ​​of at least some of the items when the time-varying evaluation values ​​are included for at least some of the items among the multifaceted evaluation items; Further provided with <2> The recipe evaluation device according to claim 1. <4> a display processing means for displaying, on a display unit, character string data in a description format corresponding to the evaluation values ​​of the multifaceted evaluation items estimated by the estimation means; Further provided with <1> from <3> 10. The recipe evaluation device according to claim 9, wherein <5> a request acquisition means for acquiring user request information regarding food and drink; a generation means for generating recipe information including the review information and the specific information based on the acquired user request information, using a large-scale language model trained using a recipe information group and a review information group; Further provided with the acquiring means acquires the specific information from the generated recipe information. <1> from <4> 10. The recipe evaluation device according to claim 9, wherein <6> a display processing means for displaying, on a display unit, factor information having a causal relationship or a correlation with at least some of the evaluation items in the multifaceted evaluation items, the factor information including at least some of the production steps and preparation items indicated in the generated recipe information; Further provided with <5> The recipe evaluation device according to claim 1. <7> the display processing means causes the display unit to display information indicating a causal relationship or a correlation between each factor included in the factor information and at least some of the evaluation items in the multifaceted evaluation items. <6> The recipe evaluation device according to claim 1. <8> a selection processing means for acquiring information on an evaluation item selected by a user from the multifaceted evaluation items; a display processing means for displaying, on a display unit, information on the production steps and preparation items indicated in the generated recipe information, which are in a causal or correlative relationship with the evaluation item selected by the user; Further provided with <6> or <7> The recipe evaluation device according to claim 1. <9> a display processing means for displaying the preparation items indicated in the generated recipe information on a display unit; a change processing means for changing the recipe information in response to a change operation by a user for the displayed preparation items; Further provided with The estimation means acquires the specific information from the changed recipe information, and based on the specific information, further estimates evaluation values ​​of multifaceted evaluation items of food and drink prepared using the changed recipe indicated by the recipe information; the display processing means causes the display unit to display information corresponding to the evaluation values ​​of the multifaceted evaluation items before and after the change, which are output based on the recipe information before and after the change, in a comparative manner; <6> from <8> 10. The recipe evaluation device according to claim 9, wherein <10> The generating means By querying a recipe database based on the acquired user request information, related recipe information highly relevant to the user request is acquired; generating an input prompt by providing input data based on the obtained related recipe information to a small language model; obtaining the recipe information and the review information by providing the generated input prompt to the large-scale language model; The recipe database is a Retrieval Augmented Generation (RAG) database constructed from a large number of recipe data, the small-scale language model is configured with a smaller number of parameters than the large-scale language model; <5> from <9> 10. The recipe evaluation device according to claim 9, wherein <11> a means for acquiring recipe evaluation information indicating an evaluation of one or more evaluation items among the multifaceted evaluation items of the food or drink prepared using the recipe of the output recipe information; Further provided with the recipe database stores recipe data in a configuration including a plurality of elements and properties assigned between the elements; the generation means inputs the acquired recipe evaluation information into a trained model to acquire a value or an increase / decrease value of the property in the recipe database corresponding to the recipe evaluation information, and updates the property in the recipe database with the acquired value or increase / decrease value. <10> The recipe evaluation device according to claim 1. <12> ingredient information acquisition means for acquiring ingredient supplement information including one or both of the use history information of food ingredients and the surplus information of food ingredients; Further provided with the generating means generates the recipe information based on the user request information and the acquired ingredient supplement information. <5> from <11> 10. The recipe evaluation device according to claim 9, wherein <13> ingredient information acquisition means for acquiring ingredient supplement information including consumption history information of food and drink ingredients; evaluation history acquisition means for acquiring, as evaluation history information, history information of the evaluation values ​​of the multifaceted evaluation items estimated by the estimation means; Further provided with The generating means generates the recipe information taking into consideration the use of food and beverage ingredients related to changes in the evaluation values ​​of specific multifaceted evaluation items, based on the user request information, the acquired ingredient supplementary information, and the acquired evaluation history information. <5> from <11> 10. The recipe evaluation device according to claim 9, wherein <14> a related information providing means for providing recipe-related store site information including one or more of information on a mail-order site for preparations required for the target recipe, information on a restaurant corresponding to the target recipe, or information on a food delivery ordering site for the restaurant; Further provided with <1> from <13> 10. The recipe evaluation device according to claim 9, wherein <15> 1. A recipe evaluation method executed by one or more computers, comprising: the one or more computers: acquiring review information about the food and drink prepared using the target recipe and specific information about the food and drink prepared using the target recipe; a step of estimating evaluation values ​​of multifaceted evaluation items of the food or drink prepared using the target recipe based on the acquired review information and the identification information; A recipe evaluation method that performs <16> The evaluation values ​​of the multifaceted evaluation items further include a time-course evaluation value indicating an evaluation over time when eating the food or drink prepared according to the target recipe. <15> The recipe evaluation method described in <17> a step in which, when the time-course evaluation values ​​are included for at least some of the items among the multifaceted evaluation items, the one or more computers display, on a display unit, a time-course evaluation change video that dynamically shows a time-course change in evaluation during eating and drinking corresponding to the time-course evaluation values ​​of the at least some of the items; Run the following again: <16> The recipe evaluation method described in <18> a step of causing the one or more computers to display, on a display unit, character string data in a descriptive format corresponding to the evaluation values ​​of the multifaceted evaluation items; Run the following again: <15> from <17> 10. A recipe evaluation method according to claim 9, wherein <19> the one or more computers: acquiring user request information regarding food and beverages; generating recipe information including the review information and the specific information based on the acquired user request information using a large-scale language model trained using a recipe information group and a review information group; and further performing a step of acquiring the specific information from the generated recipe information. <15> from <18> 10. A recipe evaluation method according to claim 9, wherein <20> a step of displaying, on a display unit, factor information having a causal relationship or correlation with at least some of the evaluation items in the multifaceted evaluation items, the factor information including at least some of the creation steps and preparation items indicated in the generated recipe information, by the one or more computers; Run the following again: <19> The recipe evaluation method described in <21> a step of causing the one or more computers to display, on the display unit, information indicating a causal relationship or a correlation between each factor included in the factor information and at least some of the evaluation items in the multifaceted evaluation items; Run the following again: <20> The recipe evaluation method described in <22> a step in which the one or more computers acquire information on evaluation items selected by a user from the multifaceted evaluation items; a step of displaying, on a display unit, information on the production steps and preparation items indicated in the generated recipe information, which are in a causal or correlative relationship with the evaluation item selected by the user; Run the following again: <20> or <21> The recipe evaluation method described in <23> the one or more computers: a step of displaying the preparation items indicated by the generated recipe information on a display unit; changing the recipe information in response to a user's change operation on the displayed preparation items; obtaining the specific information from the modified recipe information; a step of further estimating evaluation values ​​of multifaceted evaluation items of the food or drink prepared using the modified recipe indicated by the modified recipe information based on the acquired specific information; a step of displaying information corresponding to the evaluation values ​​of the multifaceted evaluation items before and after the change, which is output based on the recipe information before and after the change, on a display unit so as to be comparable; Run the following again: <20> from <22> 10. A recipe evaluation method according to claim 9, wherein <24> the one or more computers: a step of acquiring related recipe information that is highly relevant to the user's request by querying a recipe database based on the acquired user's request information; generating an input prompt by providing input data based on the obtained related recipe information to a small language model; Further execute In the step of generating the recipe information and the review information, the recipe information and the review information are obtained by providing the generated input prompt to the large-scale language model; The recipe database is a Retrieval Augmented Generation (RAG) database constructed from a large number of recipe data, the small-scale language model is configured with a smaller number of parameters than the large-scale language model; <19> from <23> 10. A recipe evaluation method according to claim 9, wherein <25> the recipe database stores recipe data in a configuration including a plurality of elements and properties assigned between the elements; the one or more computers: a step of acquiring recipe evaluation information indicating an evaluation of one or more evaluation items among the multifaceted evaluation items of the food or drink prepared using the recipe of the output recipe information; a step of inputting the acquired recipe evaluation information into a trained model to acquire the value or increase / decrease of the property in the recipe database corresponding to the recipe evaluation information; updating the property in the recipe database with the obtained value or increment; Run the following again: <24> The recipe evaluation method described in <26> the one or more computers: Further performing a step of acquiring ingredient supplement information including one or both of the consumption history information of the food ingredient and the surplus information of the food ingredient; The recipe information is generated based on the user request information and the acquired ingredient supplement information. <19> from <25> 10. A recipe evaluation method according to claim 9, wherein <27> the one or more computers: acquiring ingredient supplement information including consumption history information of food and drink ingredients; a step of acquiring history information of the estimated evaluation values ​​of the multifaceted evaluation items as evaluation history information; Further execute The recipe information is generated based on the user request information, the acquired ingredient supplement information, and the acquired evaluation history information, taking into consideration the use of food and beverage ingredients related to changes in evaluation values ​​of specific multifaceted evaluation items. <19> from <25> 10. A recipe evaluation method according to claim 9, wherein <28> the one or more computers: providing recipe-related store site information including one or more of information on a mail-order site for preparations required for the target recipe, information on a restaurant corresponding to the target recipe, or information on a food delivery ordering site for the restaurant; Run the following again: <15> from <27> 10. A recipe evaluation method according to claim 9, wherein <29> A computer program stored in one or more memories and executed by one or more processors, <15> from <28> 2. A computer program causing the one or more computers, each having the one or more memories and the one or more processors, to execute the recipe evaluation method according to any one of claims 1 to 11. [Explanation of symbols]

[0117] 1 CPU 2. Memory 3 Input / Output Interface 4 Communication Unit 5 Display device 6 Input Devices 10 Recipe evaluation device (this evaluation device) 11. Communications Network 15 User terminal 21 Request Acquisition Department 22 Recipe generation section 23 Estimation part 25 Display processing section 26 Operation processing section 27 Change Processing Section 61 Material information acquisition department 62 Evaluation history acquisition unit 63 Related Information Department 221 Recipe Database (DB) 222 Preference-specific optimization model (optimization model)

Claims

1. an acquisition means for acquiring review information of a food or drink prepared using a target recipe and specific information about the food or drink prepared using the target recipe; an estimation means for estimating evaluation values ​​of multifaceted evaluation items of the food or drink prepared using the target recipe based on the acquired review information and the identification information; A recipe evaluation device comprising:

2. The evaluation values ​​of the multifaceted evaluation items further include a time-course evaluation value indicating an evaluation over time when eating the food or drink prepared according to the target recipe. The recipe evaluation device according to claim 1 .

3. a display processing means for displaying, on a display unit, a time-varying evaluation video dynamically showing a change in evaluation over time during eating and drinking corresponding to the time-varying evaluation values ​​of at least some of the items when the time-varying evaluation values ​​are included for at least some of the items among the multifaceted evaluation items; The recipe evaluation device according to claim 2 , further comprising:

4. a display processing means for displaying, on a display unit, character string data in a description format corresponding to the evaluation values ​​of the multifaceted evaluation items estimated by the estimation means; The recipe evaluation device according to claim 1 , further comprising:

5. a request acquisition means for acquiring user request information regarding food and drink; a generation means for generating recipe information including the review information and the specific information based on the acquired user request information, using a large-scale language model trained using a recipe information group and a review information group; Further provided with the acquiring means acquires the specific information from the generated recipe information. The recipe evaluation device according to claim 1 .

6. a display processing means for displaying, on a display unit, factor information having a causal relationship or a correlation with at least some of the evaluation items in the multifaceted evaluation items, the factor information including at least some of the production steps and preparation items indicated in the generated recipe information; The recipe evaluation device according to claim 5 , further comprising:

7. the display processing means causes the display unit to display information indicating a causal relationship or a correlation between each factor included in the factor information and at least some of the evaluation items in the multifaceted evaluation items. The recipe evaluation device according to claim 6 .

8. a selection processing means for acquiring information on an evaluation item selected by a user from the multifaceted evaluation items; a display processing means for displaying, on a display unit, information on the production steps and preparation items indicated in the generated recipe information, which are in a causal or correlative relationship with the evaluation item selected by the user; The recipe evaluation device according to claim 6 , further comprising:

9. a display processing means for displaying the preparation items indicated in the generated recipe information on a display unit; a change processing means for changing the recipe information in response to a change operation by a user for the displayed preparation items; Further provided with The estimation means acquires the specific information from the changed recipe information, and based on the specific information, further estimates evaluation values ​​of multifaceted evaluation items of food and drink prepared using the changed recipe indicated by the recipe information; the display processing means causes the display unit to display information corresponding to the evaluation values ​​of the multifaceted evaluation items before and after the change, which are output based on the recipe information before and after the change, in a comparative manner; The recipe evaluation device according to claim 6 .

10. The generating means By querying a recipe database based on the acquired user request information, related recipe information highly relevant to the user request is acquired; generating an input prompt by providing input data based on the obtained related recipe information to a small language model; obtaining the recipe information and the review information by providing the generated input prompt to the large-scale language model; The recipe database is a Retrieval Augmented Generation (RAG) database constructed from a large number of recipe data, the small-scale language model is configured with a smaller number of parameters than the large-scale language model; The recipe evaluation device according to claim 5 .

11. a means for acquiring recipe evaluation information indicating an evaluation of one or more evaluation items among the multifaceted evaluation items of the food or drink prepared using the recipe of the output recipe information; Further provided with the recipe database stores recipe data in a configuration including a plurality of elements and properties assigned between the elements; the generation means inputs the acquired recipe evaluation information into a trained model to acquire a value or an increase / decrease value of the property in the recipe database corresponding to the recipe evaluation information, and updates the property in the recipe database with the acquired value or increase / decrease value. The recipe evaluation device according to claim 10 .

12. ingredient information acquisition means for acquiring ingredient supplement information including one or both of the use history information of food ingredients and the surplus information of food ingredients; Further provided with the generating means generates the recipe information based on the user request information and the acquired supplementary ingredient information. The recipe evaluation device according to claim 5 .

13. ingredient information acquisition means for acquiring ingredient supplement information including consumption history information of food and drink ingredients; evaluation history acquisition means for acquiring, as evaluation history information, history information of the evaluation values ​​of the multifaceted evaluation items estimated by the estimation means; Further provided with The generating means generates the recipe information taking into consideration the use of food and beverage ingredients related to changes in the evaluation values ​​of specific multifaceted evaluation items, based on the user request information, the acquired ingredient supplementary information, and the acquired evaluation history information. The recipe evaluation device according to claim 5 .

14. a related information providing means for providing recipe-related store site information including one or more of information on a mail-order site for preparations required for the target recipe, information on a restaurant corresponding to the target recipe, or information on a food delivery ordering site for the restaurant; The recipe evaluation device of claim 13 further comprising:

15. 1. A recipe evaluation method executed by one or more computers, comprising: the one or more computers: acquiring review information about the food and drink prepared using the target recipe and specific information about the food and drink prepared using the target recipe; a step of estimating evaluation values ​​of multifaceted evaluation items of the food or drink prepared using the target recipe based on the acquired review information and the identification information; A recipe evaluation method that performs

16. A computer program stored in one or more memories and executed by one or more processors, the computer program causing the one or more computers having the one or more memories and the one or more processors to execute the recipe evaluation method of claim 15.

Images