A method for assisting in the selection of a cooking recipe

Abstract

A method for assisting in the selection of at least one recipe from a complete list (Ltot) of recipes, comprising: A first step of receiving (E1) at least one guest profile (P1, PN); A step of determining (E2) a restricted list (Lres) of recipes from the complete list (Ltot) of recipes; A step of transmitting (E3) to a generative language model (LLM) a textual request to generate a recommended list (Lrec) chosen from the recipes in the restricted list (Lres); A second step of receiving (E4) in response to said textual generation request, the recommended list (Lrec); A step of communicating (E5) the recommended list; A step of selecting (E6) a recipe from the recommended list (Lrec). Figure 1

Description

Title of the invention: Method for assisting in the selection of a cooking recipe technical field

[0001] The invention relates to the field of preparing a culinary dish and more particularly to a method of assisting in the selection of at least one cooking recipe from a complete list of recipes. Prior art

[0002] When a cook tries to prepare a meal, they often encounter difficulties in finding recipe ideas that will suit several people. Indeed, the chosen recipe must take into account several parameters such as allergies, intolerances, dislikes, and individual culinary preferences. Therefore, finding a recipe that pleases everyone while also meeting other criteria such as cost, time, nutritional value, or seasonality can prove to be a significant challenge.

[0003] There are applications which, using a generative language model, propose a list of recipes adapted to a list of selected ingredients and which takes into account various parameters indicated by the user.

[0004] The drawback of this type of application is that it requires a long response time and a large textual query size, also called a "prompt," meaning a significant number of "tokens" for the generative language model. Furthermore, the generative language model sometimes fails to take into account all the parameters specified by the user.

[0005] There is therefore a need in this sense for a method of assistance in a selection of a cooking recipe which is reliable, fast, and requires a limited number of tokens from the generative language model. Description of the invention

[0006] One embodiment relates to assistance in selecting at least one cooking recipe from a complete list of recipes, said method being implemented by a system comprising: - A first stage of reception, of at least one guest profile including a unique guest profile identifier, and at least one restriction data point, and at least one preference data point; - A determination step in which the system determines a restricted list of recipes from the complete list of recipes based on at least one restriction data point; - A transmission step in which the system transmits to at least one generative language model, at least one textual request to generate a recommended list of recipes chosen from the recipes in the restricted list of recipes, said textual generation request including the restricted list of recipes, and at least one preference data from at least one guest profile; - A second stage of receiving, in response to the said textual request for generation, the recommended list of recipes; - A communication step via the recommended recipe list system; - A step where a user selects at least one recipe from the recommended list of recipes for implementation.

[0007] The method according to the invention aims to help a user select at least one recipe, taking into account the dietary restrictions and preferences of each guest. In this way, the user of the method can quickly and easily determine a recipe for preparing a dish, or a set of dishes forming a complete menu, that will please all the guests.

[0008] The complete recipe list is defined as a list comprising a plurality of recipes on which no filter has been applied. Each recipe details a list of ingredients, at least one recipe parameter, and recipe instructions for preparing a dish corresponding to that recipe.

[0009] The recipe parameter is a parameter relating to the recipe. The recipe parameter is specific to each recipe. The recipe parameter can be defined for the preparation of the dish for one person, or for the entire recipe.

[0010] The method is implemented by a system that includes at least one human-machine interface and a bidirectional communication module with the generative language model. The generative language model can be integrated into the system or implemented on a server remote from the system.

[0011] The generative language model is a large language model (LLM), in particular a pre-trained generative transformer model (GPT). In other words, the generative language model is an artificial intelligence model capable of generating output text from a text query. The generative language model is typically a large language model (LLM), in particular a pre-trained generative transformer model (GPT), for example GPT-4, LaMDA, LLaMA, BLOOM, etc.

[0012] In the first step, the user enters, via the human-machine interface, at least one guest profile. The guest profile corresponds to a person to whom the dish prepared from the recipe will be served. The guest profile lists characteristics, or culinary tastes are associated with that person. Therefore, there are as many guest profiles as there are people who will be tasting the dish. Each guest profile includes at least one restriction, corresponding to the person's dislikes, and at least one preference, corresponding to the person's tastes.

[0013] At least one restriction data, and at least one preference data are received by the system.

[0014] Based on the above information, the system determines the restricted list of recipes. In other words, the system processes the complete list of recipes to remove those incompatible with at least one restriction parameter. The restricted list therefore takes into account, with certainty because it is generated by a deterministic algorithm, at least one restriction parameter. The restricted list of recipes comprises recipes selected from the complete list of recipes.

[0015] The system then transmits to the generative language model, by means of the generation textual query, this restricted list as well as at least one preference data of at least one guest profile, during the transmission step.

[0016] The textual query, generally called a "prompt," corresponds to an input text for the generative language model, defining what is being requested. The textual query is in natural language, that is, the "normal" language spoken by a human being and not a structured computer language. Natural language can be written without prior training, unlike all computer languages.

[0017] As regards the structure of the textual query, it is sufficient that the generative language model be able to understand that one wishes to obtain a recommended list of recipes from the recipes in the restricted list of recipes taking into account at least one preference data from at least one guest profile.

[0018] More specifically, the text query includes in particular a sentence defining that a desired action of the generative language model is the generation of a recommended list of recipes chosen from the recipes in the restricted list of recipes, the selection having to take into account at least one preference data of at least one guest profile.

[0019] During the second reception step, the system receives from the generative language model the response to the textual generation request consisting of the recommended list of recipes. Then, the system communicates this list to the user by means of the human-machine interface, for example via a display screen.

[0020] Finally, the user selects one of the recipes from the recommended list of recipes so that he can review it with a view to its implementation.

[0021] Thus, the method according to the invention makes it possible to select a subset of recipes from the complete list of recipes so as to improve a response time of the generative language model, while ensuring, by a deterministic algorithm, that all restriction data have been taken into account.

[0022] Furthermore, creating the restricted list of recipes allows the generative language model to process only a portion of the complete list of recipes. This is particularly advantageous because it limits the size of the text query, i.e., the "prompt," and therefore the number of "tokens," i.e., units of information used by the LLM, and thus the time / cost required to obtain a response to the text query.

[0023] Finally, the process also makes it possible to limit the risk of hallucination of the generative language model so that the recommended list corresponds to the context of realization of the recipe specified by the user.

[0024] The object of this presentation may also have one or more of the following characteristics taken alone or in combination.

[0025] In some embodiments, the first step also includes receiving at least one selection criterion for a recipe parameter, the system determining the restricted list of recipes also based on at least one selection criterion.

[0026] The selection criterion for the recipe parameter is a criterion defining the recipe parameter. In other words, the user wants the recipe parameter to satisfy the selection criterion. There are as many selection criteria as there are recipe parameters considered by the user.

[0027] In the first step, the user also specifies at least one selection criterion for the recipe parameter that is transmitted to the system. The system can then, using its deterministic algorithm, select the recipes from the restricted list of recipes based on this selection criterion, that is, the recipes from the complete list of recipes whose recipe parameter satisfies the selection criterion.

[0028] The context in which the recipe is carried out is therefore specified and taken into account in a certain way.

[0029] The restricted list of recipes takes into account more parameters so that the response time and reliability of the response of the generative language model is improved.

[0030] The selection criterion can be a minimum value, a maximum value, or an interval.

[0031] In some embodiments, the first step also includes the upon receipt of at least one criterion for choosing a menu parameter, the system determines the restricted list of recipes based also on at least one menu criterion.

[0032] The menu parameter selection criterion is a criterion defining the menu parameter that the user wishes to implement. There are as many selection criteria as there are menu parameters considered by the user.

[0033] The menu parameter is a parameter that applies to all the dishes in the menu taken as a whole. The menu parameter can be defined for the creation of the menu. For example, a menu parameter is chosen from: a total number of calories in the menu, or a general theme of the menu,

[0034] The theme of the menu can be a menu in which each recipe includes an ingredient from the sea, from a particular country, or in which each recipe can be easily transported for the purpose of preparing a picnic.

[0035] In certain embodiments, at least one recipe parameter is chosen from: an estimated time to prepare the recipe, an estimated budget for the recipe, an estimated number of calories in the recipe, a type of recipe, the seasonality of the ingredients in the recipe

[0036] The type of recipe corresponds to a category of a function of the dish obtained when preparing the recipe in the meal, such as for example a starter, a main course, a side dish, a dessert, an amuse-bouche...

[0037] For each recipe parameter, a selection criterion can be applied. Examples are given in the table below:

[0038] [Table 1] Recipe parameters Example selection criteria Estimated preparation time Between 15 min and 45 min Estimated budget Less than €7 / person Number of calories Less than 300 kcal / person Recipe type Starter

[0039] In some embodiments, the restricted list of recipes includes a number of recipe restrictions.

[0040] In other words, after application of the criteria, in particular the restriction data, the number of recipes in the restricted list is limited to the number of restrictions.

[0041] In some embodiments, the number of restrictions is specified by the user.

[0042] In some embodiments, the number of restrictions is fixed and predetermined.

[0043] In some embodiments, the number of restrictions is between 50 and 1000 recipes, preferably between 100 and 500 recipes, for example 150 recipes.

[0044] In some embodiments, the restricted list of recipes includes the number of recipe restrictions, said recipes being chosen from among the most popular recipes.

[0045] In some embodiments, at least one restriction data point is chosen from: a type of diet, a food, a food family, an allergy.

[0046] The restriction data lists the person's dislikes, intolerances or allergies.

[0047] Thus the restriction data makes it possible to take into account characteristics that the recipe which will be selected by the user must not have.

[0048] In some embodiments, at least one preference data item includes at least one preference recipe.

[0049] The preference data corresponds to the person's tastes. The preference data corresponds to at least, and preferably a plurality, of recipes that appeal to said person. The preference data can therefore be a list of recipes that the person likes.

[0050] The generative language model must then process all of this preference data to determine the recommended list of recipes.

[0051] In some embodiments, the at least one generation text query further includes a sentence defining that a desired action of the generative language model is to write an explanatory text, for each recipe in the recommended list of recipes, of the choice of said recipe in relation to at least one preference data of at least one guest profile.

[0052] The text query may also specify that for each recipe in the recommended list of recipes, the generative language model must indicate in a text why this recipe was selected with regard to at least one preference data point.

[0053] In the case where a plurality of guest profiles have been entered, the explanatory text indicates why this recipe was selected with regard to each preference data of each profile.

[0054] The explanatory text thus allows the user to understand the choice that has been made.

[0055] In some embodiments, at least one textual generation query further includes a sentence defining that the response to the textual generation query must be in a user-structured file in a predefined markup language.

[0056] The file is said to be "structured" in that it is formatted according to an expected model, i.e. "empty" or skeleton file.

[0057] In some embodiments, the predefined markup language is JavaScript Object Notation.

[0058] JavaScript Object Notation (JSON) is a textual data format derived from the object notation of the JavaScript language, but other languages ​​could be used, for example, those based on XML or HTML. By definition, any file in a markup language such as JSON is structured.

[0059] In some embodiments, at least one generation text query further includes a sentence defining that a desired action of the generative language model is a proposal of music and / or an accompanying drink for each recipe in the recommended recipe list.

[0060] Thus, the user can accompany the dish served with a drink or music.

[0061] In some embodiments, the system is chosen from: a mobile terminal, a cooking appliance, a set comprising a mobile terminal and a cooking appliance.

[0062] In the case of the assembly, some steps can be performed by the mobile terminal while others are performed by the cooking appliance. For example, the first reception step, the determination step, and the transmission step can be performed by the mobile terminal, while the second reception step, the communication step, and the selection step are performed by the cooking appliance.

[0063] In some embodiments, the process includes an implementation step in which the system is configured to assist the user in implementing the selected recipe.

[0064] The assistance provided during the implementation stage may be programming a cooking appliance or following the recipe step by step.

[0065] In some embodiments, the process includes a recording step in which the recipe selected from the recommended list of recipes is recorded and then passed to at least one generative language model during a new generation text query.

[0066] Thus a history of the selected recipes is made and transmitted to the generative language model so as to improve the selection of the recommended list in a subsequent iteration.

[0067] In some embodiments, at least one textual generation query is of the form:

[0068] “P(Pref 1,..., PrefN, Lres)

[0069] Lres: DOC

[0070] Prefl: DOCPrefl

[0071]

[0072] PrefN: DOCPrefN»

[0073] Where Prefl, PrefN is a reference to at least one preference data from guest profile 1 to N; Lres is a reference to the restricted list of recipes; P(...) the generation text query; DOC is a text of the recipes from the restricted list of recipes in a text file; DOCPrefl, DOCPrefN is a text of at least one preference data from guest profile 1 to N. Brief description of the drawings

[0074] The invention will be better understood from the following description, which relates to one or more embodiments according to the present invention, given by way of non-limiting examples and explained with reference to the accompanying schematic drawings, in which:

[0075] [Fig-1] is a sequence diagram of the process according to the invention;

[0076] [Fig.2] is a representation of the method according to the invention implemented by a mobile terminal

[0077] [Fig.3] is a representation of the process according to the invention implemented by a cooking appliance

[0078] [Fig.4] is a representation of the method according to the invention implemented by an assembly comprising a mobile terminal and a cooking appliance. Description of embodiments

[0079] Only the elements necessary for understanding the invention have been shown. To facilitate reading the drawings, the same elements bear the same reference numerals from one figure to another.

[0080] The present invention relates to a method 100 for assisting in the selection of at least one cooking recipe from a complete list of recipes Ltot, as shown in [Fig. 1] to 4. The method 100 according to the invention is intended to help a user Usr / A select at least one cooking recipe by taking into account restrictions Resl, ResN and culinary preferences Prefl, PrefN of each guest 1, N. In this way, the user Usr / A of the method can quickly and easily determine a cooking recipe that will prepare a dish pleasing to all the guests 1, N.

[0081] The method 100 is implemented by a Syst system which includes at least one human-machine interface and a bidirectional communication module with the LLM generative language model. The generative language model can be integrated into the Syst system or implemented on a server remote from the system.

[0082] The Syst system is chosen from a mobile terminal, as illustrated in Fig. 2, a cooking appliance, as illustrated in [Fig. 3], or a set comprising a mobile terminal and a cooking appliance, as illustrated in [Fig. 4].

[0083] The LLM generative language model is a large language model, in particular a pre-trained generative transformer (GPT) type model. In other words, the LLM generative language model is an artificial intelligence model capable of generating output text from a text query. The LLM generative language model is typically a large language model, in particular a pre-trained generative transformer (GPT) type model, for example GPT-4, LaMDA, LLaMA, BLOOM, etc.

[0084] During a first acceptance step El of the process, the user Usr / A enters, via the human-machine interface, at least one guest profile PI, PN. In the figures illustrated, the guest profiles PI, PN are numbered from 1 to N. Alternatively, each guest 1, N enters their own guest profile PI, PN.

[0085] Each guest profile PI, PN corresponds to a person 1, N to whom the dish prepared from the recipe will be served. The guest profile PI, PN therefore includes a unique identifier. The guest profile PI, PN lists characteristics, or culinary tastes, associated with this person. There are thus as many guest profiles PI, PN as there are people 1, N who are to eat the dish. The guest profile PI, PN includes at least one restriction data point, Resl, ResN, and at least one preference data point, Prefl, PrefN.

[0086] More specifically, the restriction data Resl, ResN lists the dislikes, intolerances, or allergies of person 1, N. At least one restriction data item Resl, ResN can be chosen from: a type of diet, a food, a food group, or an allergy.

[0087] Thus the restriction data Resl, ResN allows us to take into account characteristics that the recipe which will be selected by the user Usr / A must not have.

[0088] The preference data Prefl, PrefN corresponds to the tastes of person 1, N. The preference data Prefl, PrefN corresponds to at least, and preferably a plurality, of recipes that appeal to said person. The preference data Prefl, PrefN can therefore be a list of recipes that the person likes.

[0089] At least one restriction data Resl, ResN, and at least one preference data Prefl, PrefN are received by the Syst system.

[0090] In some embodiments, the user Usr / A also specifies at least one CS selection criterion of a recipe parameter which is transmitted to the Syst system.

[0091] The recipe parameter is a parameter relating to the recipe as a whole. The recipe parameter is specific to each recipe. The recipe parameter can be defined for preparing the dish for one person, or for the entire recipe.

[0092] At least one recipe parameter is chosen from, for example: an estimated time to prepare the recipe, an estimated budget for the recipe, an estimated number of calories in the recipe, a recipe type. The recipe type corresponds to a category of a function of the dish obtained during the preparation of the recipe in the meal, such as for example an appetizer, a main course, a side dish, a dessert, an amuse-bouche...

[0093] The CS selection criterion for the recipe parameter is a criterion defining the recipe parameter. In other words, the user Usr / A wants the chosen recipe parameter to satisfy the CS selection criterion. There are as many CS selection criteria as there are recipe parameters considered by the user Usr / A.

[0094] The CS selection criterion can be a minimum, maximum, or range value of the recipe parameter.

[0095] For each recipe parameter, a CS selection criterion can be applied. Examples are given in the table below:

[0096] [Table 2] Recipe parameters Example selection criteria Estimated preparation time Between 15 min and 45 min Estimated budget Less than €7 / person Number of calories Less than 300 kcal / person Recipe type Starter

[0097] In some embodiments, the user Usr / A also specifies at least one criterion for choosing a menu parameter,

[0098] The menu parameter selection criterion is a criterion defining the menu parameter that the user wishes to implement. There are as many selection criteria as there are menu parameters considered by the user Usr / A.

[0099] The menu parameter is a parameter that applies to all the dishes in the menu taken as a whole. The menu parameter can be defined for the creation of the menu. For example, a menu parameter is chosen from: a total number of calories in the menu, or a general theme of the menu,

[0100] The theme of the menu can be a menu in which each recipe includes an ingredient from the sea, from a particular country, or in which each recipe can be easily transported for the purpose of preparing a picnic.

[0101] The process 100 then includes a determination step E2 in which the system Syst determines a restricted list Lres of recipes based on a complete list Ltot of recipes, and at least one restriction data point Resl, ResN. According to embodiments, the determination step E2 can also determine the restricted list Lres of recipes based also on at least one selection criterion CS and / or at least one menu criterion.

[0102] The complete list Ltot of recipes is defined as a list comprising a plurality of recipes on which no filter has been applied. Each recipe details a list of ingredients, at least one recipe parameter, and recipe instructions for preparing a dish corresponding to that recipe.

[0103] The Syst system processes the complete list Ltot of recipes in such a way as to remove from the complete list Ltot of recipes those recipes incompatible with at least one restriction data point Resl, ResN. The Syst system can also remove from the complete list Ltot recipes that do not meet the selection criterion CS and / or the choice criterion.

[0104] The restricted list Lres therefore takes into account, with certainty because it is generated by a deterministic algorithm, at least one restriction parameter Resl, ResN, and optionally the selection criterion CS and / or the choice criterion. The restricted list Lres of recipes includes recipes selected from the complete list Ltot of recipes.

[0105] In some embodiments, the restricted list of recipes includes a number of recipe restrictions.

[0106] In other words, after application of the criteria associated with at least one restriction data Resl, ResN, and selection criteria CS and / or choice criteria, the number of recipes in the restricted list Lres is limited to the number of restrictions.

[0107] In some embodiments, the number of restrictions is indicated by the user Usr / A, for example during the first reception step.

[0108] In some embodiments, the number of restrictions is fixed and predetermined.

[0109] In some embodiments, the number of restrictions is between 50 and 1000 recipes, preferably between 100 and 500 recipes, for example 150 recipes.

[0110] In some embodiments, the restricted list of recipes includes the number of recipe restrictions, said recipes being chosen from among the most popular recipes.

[0111] A context for the realization of the recipe is therefore specified and taken into account in a certain way through the restricted list of recipes.

[0112] The method 100 then includes a transmission step E3 in which the system Syst transmits to at least one generative language model LLM, at least one textual request for generating a recommended list Lrec of recipes chosen from the recipes in the restricted list Lres of recipes, said textual generation request comprising the restricted list Lres of recipes, and at least one preference data Prefl, Pref N of at least one guest profile PI, PN.

[0113] The textual query, generally called a "prompt," corresponds to an input text for the generative language model, defining what is being asked of it. The textual query is in natural language, that is, the "normal" language spoken by a being Natural language is human and not a structured computer language. It can be written without prior learning, unlike all computer languages.

[0114] As regards the structure of the textual query, it is sufficient that the generative language model be able to understand that one wishes to obtain a recommended list Lrec of recipes from the recipes in the restricted list Lres of recipes taking into account at least one preference data Prefl, PrefN of at least one guest profile PI, PN.

[0115] More specifically, the text query includes in particular a sentence defining that a desired action of the generative language model LLM is the generation of a recommended list Lrec of recipes chosen from the recipes in the restricted list Lres of recipes, the selection having to take into account at least one preference data Prefl, PrefN of at least one guest profile PI, PN.

[0116] The textual generation query is, for example, of the form:

[0117] “P(Prefl,..., PrefN, Lres)

[0118] Lres: DOC

[0119] Prefl: DOCPrefl

[0120]

[0121] PrefN: DOCPrefN »

[0122] Where Prefl, PrefN is a reference to at least one preference data from guest profile 1 to N; Lres is a reference to the restricted list of recipes; P(...) the generation text query; DOC is a text of the recipes from the restricted list of recipes in a text file; DOCPrefl, DOCPrefN is a text of at least one preference data from guest profile 1 to N.

[0123] In some embodiments, the at least one generation text query further includes a sentence defining that a desired action of the LLM generative language model is to write an explanatory text, for each recipe in the recommended list Lrec of recipes, of the choice of said recipe in relation to at least one preference data Prefl, PrefN of at least one guest profile PI, PN.

[0124] The explanatory text indicates to the user Usr / A why this recipe was selected with regard to at least one preference data Prefl, PrefN of at least one guest profile PI, PN.

[0125] The explanatory text thus enables the user Usr / A to understand the choice of recipes from the recommended list Lrec that has been made.

[0126] In some embodiments, at least one generation text query further includes a sentence defining that a desired action of the LLM generative language model is a proposal of an accompanying music and / or drink for each recipe in the recommended Lrec recipe list.

[0127] Thus, the user can accompany the dish served with a drink or music.

[0128] In some embodiments, at least one textual generation query further includes a sentence defining that the response to the textual generation query must be in a user-structured file in a predefined markup language.

[0129] The file is said to be "structured" in that it is formatted according to an expected model, i.e. "empty" or skeleton file.

[0130] In some embodiments, the predefined markup language is JavaScript Object Notation.

[0131] JavaScript Object Notation (JSON) is a textual data format derived from the object notation of the JavaScript language, but other languages ​​could be used, for example, those based on XML or HTML. By definition, any file in a markup language such as JSON is structured.

[0132] Thus, assuming that the expected Lrec recommended list is in a file recommended_list.txt, that is, a user-structured file in a predefined markup language, one can, for example, use the following textual query, with comments and explanations that are not part of this query in bold:

[0133] "Can you make a recommendation from a list of five recipes from the List Restricted recipe selection so that all guests eat together, and explain why the recommendation will appeal to different guests, taking into account each guest's preferences. Guest PI has Preference Data 1, ..., and Guest PN has Preference Data N. The response must be in the form of a JSON file containing only a recipe reference, its title, and the reason for the selection. [Phrase P defining the desired action]

[0134] Restricted Recipe List: Restricted_List.txt [to be incorporated]

[0135] Preference data: DOCPrefl.txt [to be incorporated]

[0136] Preference DataN: DOCPrefN.txt [to be incorporated]

[0137] Of course, the number of recipes in the recommended list Lrec of recipes may be different, or defined by the user Usr / A.

[0138] The method 100 then comprises a second reception step E4, in which the Syst system receives from the LLM generative language model the response to the textual generation request consisting of the recommended list of recipes. Then, the Syst system communicates this list to the user Usr / A by means of the human-machine interface, for example via a display screen, during the communication step E5.

[0139] Finally, the user Usr / A selects one of the recipes from the recommended list Lrec of recipes so that he can review it for implementation purposes during a selection step E6.

[0140] Optionally, method 100 includes a recording step in which the selected recipe is recorded and then passed to at least one LLM generative language model during a subsequent textual generation query. This creates a history of the selected recipes, thereby improving the selection of the recommended list (Lrec) in a subsequent iteration.

[0141] Optionally, process 100 includes an implementation step in which the Syst system is configured to assist the user Usr / A in implementing the selected recipe.

[0142] The assistance provided during the implementation stage may be programming a cooking appliance or following the recipe step by step.

[0143] The method 100 according to the invention makes it possible to select a subset of recipes from the complete list Ltot of recipes so as to improve a response time of the generative language model LLM, while ensuring, by a deterministic algorithm, that the set of restriction data Resl, ResN have been taken into account.

[0144] Furthermore, the creation of the restricted list Lres of recipes allows the generative language model LLM to process only a portion of the complete list Ltot of recipes. This is particularly advantageous because it limits the size of the textual query, i.e., the "prompt," and therefore the number of "tokens," i.e., units of information used by the generative language model LLM, and thus the time / cost to obtain a response to the textual query.

[0145] Finally, the process also makes it possible to limit a risk of hallucination of the generative language model LLM so that the recommended list Lrec corresponds to the context of realization of the recipe specified by the user Usr / A.

[0146] Figure 2 illustrates the method 100 according to the invention implemented by a mobile application. The user Usr / A enters the profiles of guests 1, N in the mobile application, and more specifically in a front-end application, then the determination step E2 is performed by a back-end application of the mobile application. The back-end application may be linked to an integrated or remote database and designated in the figure by the term "Database".

[0147] Figure 3 illustrates the method 100 according to the invention implemented by a Product IHM cooking appliance. The user Usr / A enters the guest profiles 1, N into the cooking appliance, specifically via a human-machine interface, and then the determination step E2 is performed by an integrated back-end application of the cooking appliance. The integrated back-end application can be linked to a database that is either embedded in the cooking appliance or hosted on an external "cloud" server.

[0148] Figure 4 illustrates the method 100 according to the invention implemented by the assembly comprising a mobile application and a cooking appliance. In this embodiment, the guest profiles PI and PN are populated via a front-end application of the mobile application and then transferred via a back-end application to an embedded or remote database, which performs the determination step. This database transmits the elements of the transmission step E3 either directly to the generative language model or via an integrated back-end application of the cooking appliance. The second reception step E4 is performed by the integrated back-end application of the cooking appliance. Finally, the communication step E5 is carried out on a Product HMI (Human-Machine Interface).

[0149] Although the present invention has been described with reference to specific embodiments, it is evident that modifications and changes can be made to these examples without departing from the general scope of the invention as defined by the claims. In particular, individual features of the various embodiments illustrated / mentioned can be combined in additional embodiments. Therefore, the description and drawings should be considered in an illustrative rather than a restrictive sense.

[0150] It is also evident that all the characteristics described with reference to a process are transposable, alone or in combination, to a device, and conversely, all the characteristics described with reference to a device are transposable, alone or in combination, to a process.

Claims

Demands

1. Method of assistance (100) in selecting at least one cooking recipe from a complete list (Ltot) of recipes, said method (100) being implemented by a system (Syst), comprising: - A first receiving step (E1), of at least one guest profile (PI, PN) comprising a unique identifier of the guest profile (PI, PN), and at least one restriction data (Resl, ResN), and at least one preference data (Prefl, PrefN); - A determination step (E2) in which the system (Syst) determines a restricted list (Lres) of recipes from the complete list (Ltot) of recipes according to the at least one restriction data (Resl, ResN);- A transmission step (E3) in which the system (Syst) transmits to at least one generative language model (LLM) at least one textual request to generate a recommended list (Lrec) of recipes chosen from the recipes in the restricted list (Lres) of recipes, said textual generation request comprising the restricted list (Lres) of recipes, and at least one preference data point (Prefl, PrefN) from at least one guest profile (PI, PN); - A second reception step (E4) in response to said textual generation request, of the recommended list (Lrec) of recipes; - A communication step (E5) by the system (Syst) of the recommended list (Lrec) of recipes; - A selection step (E6) by a user (Usr / A) of at least one recipe from the recommended list (Lrec) of recipes for implementation.

2. Method (100) according to claim 1, wherein the first step (El) also includes receiving at least one selection criterion (CS) of a recipe parameter, the system (Syst) determining the restricted list (Lres) of recipes also according to at least one selection criterion (CS).

3. A method (100) according to claim 2, wherein at least one recipe parameter is chosen from: an estimated duration of recipe execution, estimated budget for the recipe, estimated number of calories in the recipe, type of recipe, seasonality of the ingredients in the recipe.

4. Method (100) according to any one of the preceding claims, wherein at least one restriction data point (Resl, ResN) is chosen from: a type of diet, a food, a food family, an allergy.

5. Method (100) according to any one of the preceding claims, wherein at least one preference data point (Prefl, PrefN) comprises at least one preference recipe.

6. Method (100) according to any one of the preceding claims, wherein the at least one generation text query further comprises a sentence defining that a desired action of the generative language model (LLM) is to write an explanatory text, for each recipe in the recommended list (Lrec) of recipes, of the choice of said recipe in relation to the at least one preference data (Prefl, PrefN) of the at least one guest profile (PI, PN).

7. Method (100) according to any one of the preceding claims, wherein at least one text generation query further comprises a sentence defining that the response to the text generation query must be in a user-structured file in a predefined markup language.

8. Method (100) according to any one of the preceding claims, wherein the system (Syst) is selected from: a mobile terminal, a cooking appliance, an assembly comprising a mobile terminal and a cooking appliance.

9. A method (100) according to any one of the preceding claims, comprising an implementation step in which the system is configured to assist the user (Usr / A) in an implementation of the selected recipe.

10. A method (100) according to any one of the preceding claims, comprising a registration step in which the recipe selected from the recommended list of recipes is registered and then passed to at least one generative language model (LLM) during a new generation text query.

11. A method (100) according to any one of the preceding claims, wherein at least one textual generation query is of the form: « P(Pref 1,..., PrefN, Lres) Lres: DOC Prefl: DOCPrefl PrefN: DOCPrefN Where Prefl, PrefN is a reference to at least one preference data from guest profile 1 to N; Lres is a reference to the shortlist of recipes; P(...) the generation text query; DOC is a text of the recipes from the shortlist of recipes in a text file; DOCPrefl, DOCPrefN is a text of at least one preference data from guest profile 1 to N.

Images