Method and apparatus for recipe optimization
Patent Information
- Application Number
- KR1020230100593
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2023-08-01
- Publication Date
- 2026-09-22
- Estimated Expiration
- 2043-08-01
Smart Images

Figure 112023084899872-PAT00001_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to a method and apparatus for training a recipe optimization model. Background Technology
[0002] The recipe research and development process requires conducting repeated experiments to derive a desired recipe, and there is a problem in that it is difficult to respond immediately to market needs due to the time required for these experiments and evaluations.
[0003] Today, recipe development technologies utilizing artificial intelligence and big data are being actively researched to address these issues. However, most of these technologies that aid in recipe development are based on static data analysis. These methods analyze past data to identify general patterns and recommend recipes to users based on these patterns. Yet, these approaches fail to consider dynamic factors and have limitations in providing personalized recommendations tailored to a user's specific tastes or situations.
[0004] According to Korean Patent Publication No. 101687515, a method for modifying a recipe based on evaluation information is disclosed, which modifies the recipe of a food product based on evaluation index information and provides it to a seller.
[0005] Meanwhile, although the present disclosure is derived at least based on the technical background examined above, the technical problem or objective of the present disclosure is not limited to solving the problems or disadvantages examined above. That is, in addition to the technical problems examined above, the present disclosure can solve various technical problems related to the contents described below. The problem to be solved
[0007] The objective of the present invention is to provide a recipe optimization model to provide a user-customized optimized recipe.
[0008] However, the technical problems that this embodiment aims to solve are not limited to the technical problems described above, and other technical problems may exist. means of solving the problem
[0009] A method for training a recipe optimization model, performed by a computing device comprising at least one processor according to one embodiment of the present disclosure for realizing the aforementioned task, may include the step of training the recipe optimization model with a learning set obtained based on recipe improvement information and user evaluation information for a recipe to derive a recipe optimized for user evaluation corresponding to recipe improvement.
[0010] In one embodiment, the recipe optimization model may be reinforced learning to find an optimal action policy that determines which action to take for a given recipe state to receive the maximum reward, based on action information related to the transition of at least one recipe state, recipe improvement information according to the action, and user evaluation information related to the reward according to the action.
[0011] In one embodiment, the recipe optimization model is trained to determine an action based on mapping information when determining an action, and the mapping information may include the relationship between the combination of ingredients and functional changes of the recipe.
[0012] In one embodiment, the recipe improvement information includes one or more changes to information regarding the type of ingredient, ingredient components, ingredient function, amount of ingredient, cooking conditions, and recipe reviews for each recipe, and the user evaluation information may include one or more information regarding user behavior and user sensory evaluation.
[0013] In one embodiment, the step of training a recipe optimization model with a learning set obtained based on recipe improvement information and user evaluation information for a recipe to derive a recipe optimized for user evaluation corresponding to the recipe improvement may include the step of obtaining an improved recipe for a target recipe based on the recipe optimization model.
[0014] In one embodiment, the step of training a recipe optimization model with a training set obtained based on recipe improvement information and user evaluation information for a recipe to derive a recipe optimized for user evaluation corresponding to the recipe improvement may include the step of updating the recipe optimization model with a training set obtained based on recipe improvement information and user evaluation information for a target recipe.
[0015] In one embodiment, the step of training a recipe optimization model with a learning set obtained based on recipe improvement information and user evaluation information for a recipe to derive a recipe optimized for user evaluation corresponding to the recipe improvement may include: a step of updating the recipe optimization model with a learning set obtained based on recipe improvement information and user evaluation information for a first target recipe; and a step of obtaining an improved recipe for a second target recipe based on the updated recipe optimization model.
[0016] In one embodiment, the step of training a recipe optimization model with a learning set obtained based on recipe improvement information and user evaluation information for a recipe to derive a recipe optimized for user evaluation corresponding to the recipe improvement may include: a step of obtaining an improved recipe for a target recipe based on the recipe optimization model; a step of updating the recipe optimization model with a learning set obtained based on recipe improvement information and user evaluation information for the improved recipe; and a step of searching for recipe improvement for the improved recipe based on the updated recipe optimization model to obtain an additional improved recipe for the improved recipe.
[0017] In one embodiment, the recipe optimization model is pre-trained to form initial knowledge based on mapping information, and the mapping information may include information regarding the relationship between the combination of ingredients and functional changes of the recipe.
[0018] In one embodiment, user evaluation information for the recipe may be associated with mapping information including the relationship between the ingredient combination and functional change of the recipe.
[0019] In one embodiment, the step of training a recipe optimization model with a training set obtained based on recipe improvement information and user evaluation information for a recipe to derive a recipe optimized for user evaluation corresponding to the recipe improvement may include the step of training the recipe optimization model with a training set obtained based on recipe improvement information for a recipe improved based on user requirements for a basic recipe and user evaluation information for the improved recipe.
[0020] A computer program stored in a computer-readable storage medium according to one embodiment of the present disclosure for realizing the aforementioned task, wherein the computer program performs operations for optimizing a recipe when executed on at least one processor of a device, and the operations may include: training a recipe optimization model with a learning set obtained based on recipe improvement information and user evaluation information for a recipe to derive a recipe optimized for user evaluation corresponding to recipe improvement.
[0021] A recipe optimization device according to one embodiment of the present disclosure for realizing the aforementioned task comprises, as a computing device, at least one processor that executes at least one program instruction; and the at least one processor may be configured to train a recipe optimization model with a learning set obtained based on recipe improvement information and user evaluation information for a recipe to derive a recipe optimized for user evaluation corresponding to recipe improvement.
[0022] A method for optimizing a recipe, performed by a computing device comprising at least one processor according to one embodiment of the present disclosure for realizing the aforementioned task, comprises the step of obtaining an improved recipe for a collected target recipe based on a recipe optimization model, wherein the recipe optimization model may correspond to a model trained with a learning set obtained based on recipe improvement information and user evaluation information for a recipe to derive a recipe optimized for user evaluation corresponding to recipe improvement. Effects of the invention
[0024] According to the present invention, the following effects are achieved.
[0025] A model trained through a method for training a recipe optimization model according to some embodiments of the present disclosure can identify the best action to achieve a given goal, which is to recommend the recipe most suitable for the user. That is, the method for training a recipe optimization model can demonstrate superior performance compared to existing methods by providing user-customized recipes to enhance the user experience.
[0026] Specifically, since the learning model learns and adapts in real time through user feedback or behavior, it possesses dynamic adaptability to respond to users' changing tastes and requirements.
[0027] Second, since the learning model considers not only short-term but also long-term rewards, it goes beyond simply recommending recipes that satisfy the user once to taking long-term benefits into account, and enables recommendations that can be extended to consider additional factors such as user health, variety, and food inventory.
[0028] Third, since learning models learn through interaction with the environment, they can learn and reflect various environmental conditions (e.g., availability of ingredients, seasons, etc.) and the resulting changes.
[0029] However, the effects of the present invention are not limited to the effects described above, and unmentioned effects will also be clearly understood by those skilled in the art from this specification and the attached drawings. Brief explanation of the drawing
[0031] To enable an understanding of the characteristics mentioned in this disclosure through a detailed description together with the embodiments presented below, some embodiments are depicted in the accompanying drawings, and the same reference numerals may be used to refer to the same or similar functions in the drawings. However, the accompanying drawings are not intended to limit the scope of this disclosure but merely to represent typical embodiments, and it should be noted that other embodiments capable of producing the same effect exist. FIG. 1 is a block diagram of a computing device in which various aspects of the present disclosure can be implemented. Figure 2 is a conceptual diagram of an agent and an environment explaining a reinforcement learning method. FIG. 3 is a flowchart illustrating an example of a recipe optimization method according to one embodiment of the present disclosure. Specific details for implementing the invention
[0032] Hereinafter, various embodiments and approaches are presented with various specific details based on the reference drawings for the sake of understanding, but these details are not strictly necessary and will be recognized by those skilled in the art. The following description and the accompanying drawings describe specific examples of a series of approaches in detail, but it should be kept in mind that these approaches are merely examples and various methods may be applied through various principles. Accordingly, these descriptions are intended to encompass various approaches and their corresponding components or equivalents. Terms such as "embodiment," "example," "aspect," and "example" as used herein do not imply that a particular approach or design is superior to others.
[0033] In addition, various approaches and features will be described through systems comprising one or more devices, terminals, servers, devices, components, modules, etc. It should be understood and recognized that various systems may include additional devices, terminals, servers, devices, components, modules, etc., or may not include all elements discussed in the drawings.
[0034] As used herein, terms such as "computer program," "component," "module," and "system" are used interchangeably and refer to computer-related entities, hardware, firmware, software, combinations of hardware and software, or executions of software. For example, a component may be, but is not limited to, a procedure executed on a processor, a processor, an object, an execution thread, a program, a computer, etc. An application program executed on a computing device and the computing device itself may also be a component. One or more components may reside within a processor or an execution thread, some components may be located on a single computer, and some may be distributed across multiple computers.
[0035] Additionally, these components may be executed from various computer-readable media having various data structures stored therein. The components may communicate through local and / or remote processing, for example, according to a signal having one or more data packets (e.g., data from one component interacting with another component in a local or distributed system, and / or data transmitted through a network such as the Internet to another system via a signal).
[0036] Hereinafter, identical or similar components are assigned the same reference numeral regardless of drawing symbols, and redundant descriptions thereof are omitted. Furthermore, in describing the embodiments disclosed in this specification, detailed descriptions of related prior art are omitted if it is determined that such detailed descriptions may obscure the essence of the embodiments disclosed in this specification. Additionally, the attached drawings are intended only to facilitate understanding of the embodiments disclosed in this specification, and the technical concept disclosed in this specification is not limited by the attached drawings.
[0037] Furthermore, these elements each contain various internal data structures and can operate on various computer-readable media. These elements can communicate through local or remote processing, for example, via signals containing one or more data packets (e.g., data and / or signals transmitted from a component of a local or distributed system to another system or a network such as the Internet).
[0038] Next, identical or similar elements are assigned the same reference number regardless of symbols in the drawings, and redundant descriptions are omitted. Additionally, when describing the embodiments disclosed in this specification, if it is determined that a detailed description of related technology could obscure the gist of this specification, such description may be omitted. Furthermore, the attached drawings are intended to facilitate understanding of the embodiments disclosed in this specification, and the technical concepts disclosed in this specification are not limited by the attached drawings.
[0039] The terms used herein are for the purpose of describing exemplary embodiments and are not intended to limit this disclosure. In this specification, the singular form may include the plural form unless specifically stated otherwise. The terms “comprising” or “comprising” do not exclude the presence or addition of one or more other elements in addition to the mentioned elements.
[0040] Although terms such as "first," "second," etc., are used to describe various parts or elements, these parts or elements are not limited by these terms. These terms are used merely to distinguish one part or element from another. Accordingly, the first part or element mentioned below may be the second part or element within the technical concept of this disclosure.
[0041] Unless otherwise defined, all terms used herein (including technical and scientific terms) may be used in the meaning generally known in the technical field to which this disclosure pertains. Terms defined in commonly used dictionaries are not to be over-interpreted unless specifically defined otherwise.
[0042] Furthermore, the word "or" implies an inclusive "or," not an exclusive one. In other words, unless specifically designated or clear from the context, the statement "X uses A or B" implies one of the implied substitutions. That is, whether X uses only A, only B, or both A and B, the expression "X uses A or B" may apply to all of these cases. In this specification, the term "and / or" refers to all possible combinations of one or more of the listed items.
[0043] Additionally, the terms "information" and "data" may often be used interchangeably in this specification.
[0044] The suffixes "module" and "part" used in the following description are for the convenience of document creation and do not have any special meaning or role in themselves.
[0045] The purpose and effects of the present disclosure, and the technical elements for achieving them, will become clear by referring to the exemplary embodiments described below in detail together with the accompanying drawings. In describing the present disclosure, specific descriptions of functions or configurations will be omitted if it is determined that such descriptions would unnecessarily obscure the main contents of the present disclosure. Furthermore, the terms described below have been defined considering their functions in the present disclosure, and these may vary depending on the intentions or practices of the user or operator.
[0046] However, the present disclosure is not limited to the embodiments disclosed below and may be implemented in various different forms. These embodiments are provided to fully explain the present disclosure and to fully inform those skilled in the art of the scope of the disclosure, and it is obvious that the present disclosure is defined by the claims.
[0047] Therefore, the definition must be determined based on the entire contents of this specification.
[0048] The scope of rights for each step in the claims of this disclosure is determined by the function and feature specified in each step, and is not affected by the order in which the steps are described within the claims unless the order of the steps is specified. For example, in a claim describing steps including Step A and Step B, even if Step A is described before Step B, the scope of rights is not limited by the requirement that Step A be performed before Step B.
[0049] FIG. 1 is a block diagram of a computing device in which various forms of the present disclosure can be implemented.
[0050] Referring to FIG. 1, the computing device (100) may include a processor (110), a communication unit (120), and a memory (130). However, since these components are not essential to the implementation of the computing device (100), the computing device (100) may have more or fewer components than those listed above.
[0051] The computing device (100) of the present disclosure may be a server for operating or managing a recipe optimization system, but this is only a limited example.
[0052] The computing device (100) may include any type of computer system or computer device, such as, for example, a microprocessor, a mainframe computer, a digital processor, a portable device and a device controller, but this is only a limited example.
[0053] The processor (110) of the computing device (100) generally controls the overall operation of the computing device (100). The processor (110) can provide or process appropriate information or functions to the user by processing signals, data, information, etc. that are input or output through components included in the computing device (100) or by running applications stored in memory (130).
[0054] Additionally, the processor (110) can control at least some of the components of the computing device (100) to run an application stored in memory (130). Furthermore, the processor (110) can operate at least two or more of the components included in the computing device (100) in combination with each other to run the application.
[0055] The processor (110) of the computing device (100) may include one or more cores and may have a processor that performs data analysis or deep learning, such as a central processing unit (CPU), a general purpose graphics processing unit (GPGPU), or a tensor processing unit (TPU). The processor (110) may read a computer program stored in memory (130) and perform data processing for machine learning for some embodiments according to the present disclosure. According to some embodiments of the present disclosure, the processor (110) may perform calculations necessary for neural network learning. The processor (110) may perform calculations for neural network learning, including processing input data in deep learning (DL), feature extraction, error calculation, and updating the weights of the neural network using backpropagation. At least one of the CPU, GPGPU, and TPU of the processor (110) may be responsible for the learning processing of the network function. For example, a CPU and a GPGPU can process network function training and data classification together. Additionally, in some embodiments of the present disclosure, processors of multiple computing devices can be used simultaneously to process network function training and data classification. A computer program executed on a computing device according to the present disclosure may be a program executable on a CPU, a GPGPU, or a TPU.
[0056] According to some embodiments of the present disclosure, a processor (110) of a computing device (100) can learn a recipe optimization model to obtain an optimized recipe.
[0057] Specifically, the processor (110) of the computing device (100) can obtain improvement information and user evaluation information for each of at least one recipe.
[0058] Additionally, the processor (110) can train a recipe optimization model using a learning set obtained based on recipe improvement information and user evaluation information for the recipe, so as to derive a recipe optimized for user evaluation corresponding to recipe improvement. Through this, the processor (110) can obtain information regarding the optimized recipe during the training process. However, it is not limited to this.
[0059] For example, the above recipe improvement information may include improvements or improved results in which one or more of the information regarding ingredient types, ingredient components, ingredient functions, ingredient quantities, cooking conditions, and recipe reviews for each recipe have been changed, and the above user evaluation information may include one or more of the information regarding user behavior and user sensory evaluation, provided, but not limited thereto.
[0060] Meanwhile, the processor (110) of the computing device (100) may provide information regarding the optimized recipe to the user. For example, the processor (110) may control the communication unit (120) to transmit information regarding the optimized and improved recipe to a user terminal or server. Here, the recipe improvement information regarding the improved recipe may include information regarding the action itself, such as the improved recipe state, which is the recipe state itself changed to conform to previous user feedback, the type of ingredient to be changed, and the amount of ingredient. However, it is not limited thereto.
[0061] Accordingly, the recipe optimization model of the present disclosure learns to take action to change the recipe to match user feedback and provides better recipes over time.
[0062] Hereinafter, a method for the computing device (100) of the present disclosure to learn a recipe optimization model and obtain an optimized improved recipe will be described with reference to FIGS. 2 and 3 relating to some embodiments of the present disclosure.
[0063] The communication unit (120) of the computing device (100) may have one or more modules that enable communication between the computing device and a user terminal, and between the computing device and a server. Furthermore, the communication unit (120) may include modules that connect the computing device (100) to one or more networks.
[0064] According to a specific embodiment of the present disclosure, the communication unit (120) can transmit optimized recipe information to a user terminal or server.
[0065] A network that enables communication between a computing device (100) and a user terminal, and between a computing device and servers, can use various wired communication systems, such as a public switchboard network (PSTN), various DSL technologies (xDSL, RADSL, MDSL, VDSL, UADSL, HDSL), and a LAN.
[0066] In addition, these networks can use various wireless communication systems, such as CDMA, TDMA, FDMA, OFDMA, SCFDMA, etc.
[0067] According to the present disclosure, a network may have various forms of communication, such as wired or wireless, and may be composed of various communication networks such as LAN, WAN, etc. Additionally, such a network may be the Internet (WWW) or may utilize short-range wireless transmission technologies such as infrared (IrDA) or Bluetooth.
[0068] The technologies mentioned in this specification may be applied to other networks in addition to the networks described above.
[0069] The memory (130) of the computing device (100) can store a program for the operation of the processor (110) or temporarily or permanently store input / output data. This memory (130) may include various types of storage media, such as flash memory, hard disk, multimedia card micro, card type memory (SD or XD memory, etc.), RAM, SRAM, ROM, EEPROM, PROM, magnetic memory, magnetic disk, optical disk, etc.
[0070] According to a specific embodiment of the present disclosure, the memory (130) may store a neural network or a neural network structure that forms the learning model of the present disclosure.
[0071] The present disclosure emphasizes that embodiments may be implemented as independent software modules. Each of these modules may be used to realize the various functions and operations described herein. Software code written as a software application using a programming language may be stored in the memory (130) of the computing device (100) and may be executed by the processor (110) of the computing device (100).
[0072] This specification specifies that machine learning models, deep learning-based models, computational models, neural networks, network functions, deep neural networks, and neural networks are interchangeable. A neural network is generally composed of a set of computational units called nodes, which may also be referred to as neurons. Such a neural network consists of at least one node, and the nodes may be interconnected through links.
[0073] In a neural network, nodes connected by links can form a relative relationship between input and output nodes. In this context, the concepts of input and output nodes are relative, meaning the same node can serve as both an input and an output node depending on the relationship. In the relationships between linked nodes, the data value of the output node is determined based on the data of the input node, and the connecting links can carry weights.
[0074] Neural networks establish relationships between input and output nodes through nodes connected via links. The characteristics of a neural network are determined by the number of nodes and links, the relationships between nodes and links, and the values of the weights assigned to the links. Two neural networks with the same number of nodes and links, differing only in the weight values of their links, can be considered different neural networks.
[0075] A neural network consists of a set of one or more nodes, and subsets of nodes within the neural network form layers. In this case, a layer is determined based on its distance from an initial input node, and distance is defined as the minimum number of links that must be traversed to reach that node from the initial input node. However, this definition of a layer is for the sake of convenience of explanation, and the degree of a layer within a neural network can be defined in other ways. For example, a layer of nodes can be defined based on its distance from a final output node.
[0076] The initial input node refers to a node through which data enters directly or a node that does not have links with other nodes. Conversely, the final output node refers to a node that does not have an output node, and the hidden node constitutes the remaining part of the neural network excluding these two nodes. One embodiment of the present disclosure may include a neural network in which the number of nodes in the input layer and the output layer are equal, and the number of nodes decreases and then increases again during the transition to the hidden layer. In other embodiments, the number of input nodes may be less than or greater than the number of output nodes, and may exhibit a pattern of decreasing or increasing the number of nodes, respectively. Additionally, a neural network in the form of a combination of several embodiments is also possible.
[0077] Deep neural networks include multiple hidden layers, through which they can understand the potential structure of data. There are various types of deep neural networks, and these are merely limited examples and do not depart from the scope of the present disclosure. In one embodiment of the present disclosure, the network function may include an autoencoder. The autoencoder generates output data similar to the input data and may include an odd number of hidden layers. Such an autoencoder performs non-linear dimensionality reduction through dimensionality reduction and symmetric expansion.
[0078] In addition, neural networks can be trained using methods such as supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning. This refers to the process by which a neural network acquires necessary knowledge.
[0079] A neural network is a system that learns by reducing its own errors. In this process, training data is repeatedly input into the neural network, and based on this, the error between the network's output and the target value is calculated. This error is backpropagated from the output layer to the input layer in a direction that reduces the error, and during this process, the weights of each node are modified.
[0080] Supervised learning uses labeled training data, whereas unsupervised learning uses unlabeled data. In supervised learning, the training data is labeled with classification labels, and the error is calculated by comparing this with the output of the neural network. In unsupervised learning, the error is calculated by comparing the output of the neural network with the input training data.
[0081] The calculated error is backpropagated from the network's output to the input, and during this process, the weights of each node are updated. The weight updates are determined by the learning rate, which is adjusted during the neural network's training cycle. By using a fast learning rate early in the training process to increase efficiency, and a low learning rate later in the training process to improve accuracy, the neural network can achieve greater accuracy.
[0082] Since the data used to train a neural network may be a subset of real data, the error on the training data may decrease, but the error on real data may increase. This phenomenon is called overfitting, which is a situation where a neural network learns excessively from the training data, degrading its performance on real data.
[0083] There are various optimization techniques to prevent overfitting. For example, methods include increasing the amount of training data, applying regularization, using dropout to disable some nodes in the network during training, or utilizing batch normalization layers. All of these methods are used to prevent excessive training on the training data and to improve performance on real-world data.
[0084] FIG. 3 is a flowchart illustrating an example of a method for training a recipe optimization model according to some embodiments of the present disclosure.
[0085] A method for training a recipe optimization model performed by a processor (110) of a computing device (100) described in some embodiments of the present disclosure may include steps S110, S120, and S130 performed by said processor (110).
[0086] Specifically, the above method may include a step of pre-learning to form initial knowledge based on mapping information (S100), a step of collecting recipe improvement information and user evaluation information for a target recipe (S110), a step of training a recipe optimization model with a learning set obtained based on the recipe improvement information and user evaluation information for the target recipe (S120), and a step of obtaining an improved recipe for the target recipe for the target recipe based on the recipe optimization model (S130).
[0087] In some embodiments of the present disclosure, in the step (S110) of collecting recipe improvement information and user evaluation information for a target recipe, the recipe improvement information may be recipe improvement information for a recipe that has been improved based on user requirements to a basic recipe, which is the target recipe.
[0088] In some embodiments of the present disclosure, the step (S120) of training a recipe optimization model with a training set obtained based on recipe improvement information and user evaluation information for the target recipe may include the step of training the recipe optimization model with a training set obtained based on recipe improvement information for a recipe improved based on user requirements for a basic recipe and user evaluation information for the improved recipe. For example, when training with a reinforcement learning algorithm, the basic recipe and its improved recipe may represent the initial starting point by determining the initial state and the result state according to the action.
[0089] In some embodiments of the present disclosure, the processor (110) may perform the steps of collecting user evaluation information for the improvement recipe in relation to the step S130 (S131), updating the recipe optimization model with a learning set obtained based on recipe improvement information and user evaluation information for the improvement recipe (S132), and obtaining an additional improvement recipe for the improvement recipe based on the updated recipe optimization model (S133).
[0090] That is, the processor (110) can train a recipe optimization model based on recipe information regarding at least one recipe, then input a target recipe that needs improvement into the recipe optimization model and collect user evaluation information of the improved recipe based on the output recipe improvement information, update the recipe optimization model with a learning set obtained based on the recipe improvement information and user evaluation information of the target recipe, and then obtain an additional improvement recipe for the current improvement recipe that needs further improvement using the updated recipe optimization model.
[0091] Here, the improved recipes found by the current recipe optimization model are recipes that have not yet received user evaluations and represent the scenarios that the model needs to learn. If necessary, user evaluation information regarding the current improved recipes can be received and provided to the recipe optimization model as a training set to update the model, and the updated recipe optimization model will reflect the user feedback in the form of user evaluation information. Since the updated recipe optimization model reflects user preferences compared to the previous model, it can provide additionally optimized improved recipes. Through this update, the current recipe optimization model becomes able to adapt to a new user environment different from the existing one. Therefore, by utilizing the updated recipe optimization model, it becomes possible to obtain customized improved recipes that directly reflect highly experiential and subjective user sensory evaluation elements, such as a specific user's taste or aroma.
[0092] In some embodiments of the present disclosure, the process of updating the recipe optimization model by performing step S132 in relation to S130 may be repeated to perform user-customized optimization. This update cycle may involve performing the step of updating the recipe optimization model based on the current recipe improvement information and the user evaluation information for the current improvement recipe whenever user evaluation information for the current improvement recipe is collected, updating the recipe optimization model after collecting user evaluation information according to predetermined criteria, or analyzing the user evaluation of the recipe and stopping the repetition of updates if it meets the predetermined criteria.
[0093] FIG. 6 is a drawing for explaining recipe improvement information and user evaluation information according to one embodiment of the present disclosure.
[0094] In one embodiment, the recipe is written as a method for cooking, preparing, or manufacturing according to specific conditions using a combination of ingredients that uses a certain amount of a specific type of ingredient, and may include user feedback regarding the method, and is not limited to formats such as text, images, videos, or specific values.
[0095] In one embodiment, the recipe may be limited according to the requirements of a specific user. For example, as a recipe for a desired result, it may be limited by specifying specific types of ingredients or ingredient components, etc., by the user.
[0096] Here, the user is not limited to humans or animals. If the user is a human, the person capable of evaluating the recipe may be an individual human, such as a researcher, customer, or company representative, or a group of such individuals, but is not limited thereto.
[0097] In one embodiment, regarding the recipe improvement information of the recipe, user evaluation information may be collected for each individual recipe, but user evaluation information for all possible combinations of recipe states may not be accumulated.
[0098] In one embodiment, the recipe improvement information may refer to changing one or more of the attribute information representing the state of a recipe, such as the type of ingredient, ingredient components, ingredient function, ingredient quantity, ingredient price, cooking conditions, and recipe reviews for each recipe, or information regarding the changed recipe. The user evaluation information may include one or more of the information regarding user behavior and user sensory evaluation as information based on user feedback, but is not limited thereto.
[0099] For example, ingredient types refer to the types of each ingredient that make up a recipe and may be expressed through text, symbols, images, videos, etc., but are not limited to these.
[0100] The quantity of ingredients may be a value expressing the combination of each ingredient constituting the recipe, such as weight, volume, content, or mixing ratio, but is not limited to this.
[0101] Ingredient functions may be mapped to individual ingredient types in a recipe. Ingredient functions may refer to information regarding correlations concerning functions that are proportional to or not proportional to the amount of ingredients, depending on the individual ingredients or the components they contain. For example, if the ingredient function is one that affects sensory qualities, strawberries may affect the red color, influence sweetness and sourness, and provide a distinctive aroma called strawberry scent.
[0102] Ingredient prices can be used to calculate the price of the combination of all ingredients constituting a recipe based on the quantity and type of each ingredient, using quantitative and qualitative information regarding individual ingredients.
[0103] Ingredient composition refers to information regarding the individual components of a specific type of material, and may be, for example, numerical information regarding moisture, fat, carbohydrates, protein, amino acids, vitamins, or specific components related to taste or aroma.
[0104] Cooking conditions are information regarding the cooking process of a recipe and may include cooking time, cooking temperature, cooking method, cooking sequence, cooking tools, etc. Cooking conditions have a correlation with other attributes of the recipe described above and can be mapped to them. For example, onions have a strong pungent taste before heating, but after heating, the pungent taste weakens and sweetness is added; therefore, heating can be used as a cooking condition, or heating temperature and time can be used as cooking conditions to map to a specific type of ingredient, such as an onion, and to the function of the ingredient. However, it is not limited to this.
[0105] User sensory evaluation may be an evaluation performed through the senses of the color, taste, aroma, texture, etc. of the result of providing the user with an actual recipe. Unlike user behavior evaluation, specific subjective criteria may be directly expressed, and for example, in the case of taste, it may include an evaluation of the type of taste and its intensity, such as sweetness, saltiness, spiciness, bitterness, sweetness, astringency, and sourness. However, it is not limited to this.
[0106] Recipe reviews are collected from various sources and may be written by existing users based on their experience after directly using the recipe; these reviews include the users' subjective reactions and may include evaluations of various factors (cooking difficulty, taste, ingredient costs, etc.). Recipe reviews may be provided as evaluative opinions such as sentences or words regarding the recipe, or as multiple-choice survey responses, and can be processed into an appropriate form to be linked with other information through text mining, etc. For example, a text opinion in a recipe review stating "It is too bland" can be transformed into the format "The saltiness is weak." However, it is not limited to this.
[0107] User behavior indicates the degree to which a specific user prefers a particular recipe and can be measured based on behavioral data such as user selection, clicks, and purchases. However, it is not limited to this.
[0108] For example, the relationship between a specific recipe and a user may be as follows. This describes a case where a recipe provided to a specific user is a recipe for making "matcha-flavored milk." The recipe may have the following recipe information: types of ingredients such as "milk, matcha powder, sugar"; quantities of ingredients such as "milk 89g (89%), matcha powder 1g (1%), sugar 10g (10%)"; ingredient composition such as "carbohydrate 10g, protein 10g, moisture 70g, solids 5g, fat 5g"; ingredient prices such as "milk 890 won, matcha powder 10 won, sugar 10 won"; ingredient functions such as "milk: tasteless, white 5, matcha powder: astringent 5, green 7, sugar: sweet 9, colorless"; and cooking conditions such as "add matcha powder and sugar to milk and mix." Regarding such recipes, users may provide feedback such as "the astringent taste is too strong" or "please increase the sugar by 1%," or users may perform the action of selecting the recipe. Alternatively, users may be grouped according to similar tastes, and user feedback regarding recipes provided to them in common may be aggregated and used as user evaluation information for the recipe. However, this is not limited thereto.
[0109] Information regarding such recipes may be collected through analysis via experiments, measured using measuring equipment, obtained through providers or specialized materials, or acquired through user interfaces, and may involve appropriate preprocessing, normalization, or outlier removal, but is not limited thereto.
[0110] According to some embodiments of the present disclosure, the recipe optimization model may be a learning model that has been reinforced to find an optimal action policy to determine which action to take on a given recipe state to receive the maximum reward, based on information related to the recipe state for at least one recipe, action information related to the transition of the recipe state, recipe improvement information based on the action, and user evaluation information related to the reward based on the action.
[0111] Hereinafter, according to some embodiments of the present disclosure, when the recipe optimization model is a learning model trained through reinforcement learning, the interaction between the agent (310) and the environment (330) is described with reference to FIG. 2.
[0112] The environment (330) refers to the overall context of recipe optimization and may include user preferences, availability of specific ingredients, seasonal factors, cooking factors, etc. However, it is not limited thereto.
[0113] The environment (330) provides the agent (310) with information regarding the recipe, such as various ingredient compositions, ingredient quantities, ingredient components, cooking conditions, user behavior, and user sensory evaluation. The state (state, S_t) of the recipe is based on the recipe's attribute information, and the reward (reward, R_t) is based on user evaluation information. Regarding this state (S_t) of the recipe, the agent (310) determines an action (action, A_t) to improve the recipe. When the determined action is transmitted to the environment (330), the agent (310) receives the recipe improvement information (state, S_t+1), which is the changed state, and the reward (reward, R_t) from the environment (330).
[0114] Specifically, first, the agent (310) determines an action based on the current state provided by the environment (330). This state information may include current attribute information of the recipe (ingredient composition, amount of ingredients, ingredient components, cooking conditions, etc.) and user evaluation or feedback information (user behavior, user sensory evaluation, etc.). The agent (310) determines an action based on this state information, and this action may be an action of changing or modifying the attributes of the recipe, for example, changing the amount of ingredients or substituting them with other ingredients.
[0115] Next, when the agent (310) applies the action (A_t) it has decided upon to the environment (330), the environment (330) can provide the agent (310) with the recipe improvement information (S_t+1) in a changed state. Additionally, if the agent (310) receives feedback from a user regarding the recipe improvement information resulting from its action, the feedback is delivered to the agent (310) as a reward (R_t) based on user evaluation information regarding this feedback. This reward plays an important role in the agent (310) determining future actions (A_t+1).
[0116] Accordingly, the interaction between the agent (310) and the environment (330) includes providing state information, determining actions, transmitting changed state information, and transmitting rewards, through which the agent determines actions in a direction that optimizes the recipe.
[0117] In some embodiments of the present disclosure, in a recipe optimization system trained with a recipe optimization model, input data and output data can be defined mainly as follows.
[0118] Here, the input data consists of recipe attribute information for the current recipe, recipe improvement information for the current recipe, and user rating information. This serves as the basis for the agent to decide on actions.
[0119] During the training process of a reinforcement learning model, recipe user evaluation information can be treated as a reward and serve as one of the input data. During the utilization process of the reinforcement learning model, recipe attribute information can serve as input data representing the current state, as well as output data representing the state resulting from the modified recipe as a consequence of the agent's actions. In this way, the reinforcement learning process and its utilization process operate in a manner where input and output data interact with each other.
[0120] The agent (310) performing recipe optimization decides what action to take in a given state and proceeds with learning based on the reward obtained as a result. Through this, the agent (310) learns to act in a direction that obtains a reward and recommends better recipes over time.
[0121] For example, the types and quantities of ingredients in a recipe can be viewed as a state. Here, the state refers to the ratio of each ingredient in a specific recipe. For example, '250g of chicken' could be a state.
[0122] Actions can be viewed as changing the mixing ratio of specific ingredients, and specifically, they may mean increasing or decreasing the amount of specific ingredients. For example, an action could be 'increase chicken by 50g,' and through this action, the state transitions to another state (e.g., '300g chicken').
[0123] According to some embodiments of the present disclosure, such actions may be limited by predetermined criteria. For example, the action may refer to predetermined criteria for the unit and range of changes in the mixing ratio of specific ingredients. For example, the ratio range of each ingredient may be defined in advance, such as setting the unit for increasing or decreasing chicken meat to 50g, and the action may not deviate from this range when a random action is selected. This prevents the generation of unrealistic recipes, such as ice cream containing 80% sugar. These criteria may utilize ingredient prices, ingredient functions, cooking conditions, etc., included in the attribute information of the recipe, or utilize separately collected databases, such as the Ministry of Food and Drug Safety database or a standard database based on research papers.
[0124] Specifically, a recipe optimization model can use user evaluation information regarding a recipe as a reward for reinforcement learning. This reward may refer to evaluating whether the recipe is suitable for the environment, how satisfactory it is to the user, or how much it can be improved by modifying the recipe.
[0125] Here, the result based on the recipe improvement information of the recipe is the state after taking a specific action on the current state, and when the action is taken, a new recipe state can be created as a result of the action, such as, for example, '300g of chicken'.
[0126] If the environment (330) knows the reward function and the transition probability distribution function, this reinforcement learning is called "model-based" reinforcement learning. Conversely, if the agent (310) does not know the reward function and the transition probability distribution function of the environment (330), it is called "model-free" reinforcement learning.
[0127] The agent (310) aims to learn a policy that maximizes the cumulative reward value in a given environment (330). This policy refers to a set of rules that determine the probability that the agent (310) takes a specific action in a specific state.
[0128] According to the present disclosure, the environment (330) may include a model capable of providing a state and a reward, which may exist in the form of a simulator within a computing device (100) or may receive recipe information from a user interface to calculate the state and the reward.
[0129] The method by which an agent decides its action can be based on one of three approaches: value-based, policy-based, or a combination of both. Value-based methods are those based on value functions, such as Q-learning and DQN, while policy-based methods, like Policy Gradient, decide actions based on the final return and a policy function without a value function. Methods based on both value and policy determine the agent's action by having the policy function decide the action and the value function evaluate the action; the Soft Actor-Critic algorithm falls into this category.
[0130] For example, in the case of Q-learning, the TD method is used to update Q-values, and the update formula is as follows.
[0131] Q(s, a) ηa) + α * [r + γ * max_a' Q(s', a') - Q(s, a)]
[0132] Here, s is the current state, a is the selected action, α is the learning rate, and r is the immediate reward received.
[0133] γ is a discount factor that determines how important future rewards are considered, max_a' Q(s', a') is the maximum value among all possible actions a' for the next state s', and Q(s, a) represents the previous Q-value for the current state and action. The reinforcement learning algorithm described above is merely an example and can be selected or combined as needed.
[0134] According to some embodiments of the present disclosure, user evaluation information for the recipe may be obtained through user feedback related to the mapping information. For example, it may be related to one or more attribute information representing the current state of the recipe, and this relationship is referenced as mapping information. For example, user evaluation information may be obtained as user feedback regarding sensory qualities such as taste, aroma, and texture throughout the recipe, and may be viewed as related to the overall material function regarding the combination of ingredients in the recipe. User feedback may relate to actions that change the state, such as changing the type of ingredient by replacing existing ingredients with specific ingredients, adjusting the amount of specific ingredients, or changing the combination thereof, or to changes in the state resulting from performing such actions. For example, it may be instructions to directly adjust the type or amount of ingredients, such as "Please add salt," "Please replace sugar with stevia," or "Please reduce chili oil by 1%." Alternatively, if the user feedback is related to a function such as taste, the material-function mapping information plays an important role in the interpretation of user feedback and the calculation of compensation. That is, the relationship between the user evaluation information of the recipe and the recipe information can be explained by the material-function mapping information. By using mapping information, specific materials related to user feedback can be identified, and the range of actions can be limited by making changes to those materials (e.g., adjusting the amount). This enables the reinforcement learning agent to perform more efficient learning from its initial state.
[0135] According to some embodiments of the present disclosure, the process (110) may perform a process of preprocessing such user feedback into user evaluation information based on mapping information. For example, user feedback such as "The sweetness is too strong. Sweetness is 8. Please reduce the sugar" can be produced as user evaluation information that performs optimal actions, such as replacing the ingredient or adjusting the amount of the ingredient, regarding the ingredient type related to the ingredient function of sweetness. That is, the processor (110) can interpret the user's feedback in an appropriate form using natural language processing (NLP), etc., and, for example, can extract key information such as 'sweetness', 'strong', '8', 'sugar', and 'reduce' from the feedback such as "The sweetness is too strong. Sweetness is 8. Please reduce the sugar". Through this process, the user feedback can be converted into a reward that is used for agent learning and recipe optimization.
[0136] According to some embodiments of the present disclosure, the recipe optimization model can be trained using the mapping information described above. That is, the recipe optimization model can be pre-trained with mapping information regarding ingredient functions according to ingredient combinations, and then trained using user evaluation information and recipe improvement information obtained through user feedback related to the mapping information as a training set. The trained recipe optimization model can determine an action based on the mapping information regarding the input current state to obtain an improved recipe. Since the recipe optimization model training using this mapping information is pre-trained and initialized based on functional changes that change according to ingredient combinations, the mapped ingredient combinations and corresponding functional changes using the recipe attribute information, as well as user evaluation information regarding user feedback, can be directly reflected in the reinforcement learning stage.
[0137] According to some embodiments of the present disclosure, the method for calculating rewards for training a recipe optimization model may vary depending on the characteristics and objectives of the problem. For example, if a user provides feedback as a number (e.g., "Taste rating 8"), this score may be used as the reward, and the agent may prefer that recipe as the feedback score for the recipe increases. Alternatively, if a user provides feedback by comparing it to a previous recipe (e.g., "This recipe is better than the previous recipe"), this information may be used as a relative reward. For example, if the current recipe has a higher rating than the previous recipe, the reward may be set higher, and if the opposite is true, the reward may be set lower. Preferably, user feedback may be combined with ingredient-function mapping information so that if the feedback provided by the user is related to a specific ingredient or function, the reward may be adjusted to reflect changes in that ingredient or function, thereby enabling the agent to learn user preferences more accurately and improve recipes. However, this is not limited thereto.
[0138] In other words, a reinforcement learning-based recipe optimization model utilizing ingredient-function mapping information as described above can better leverage information regarding the taste, nutritional content, and cooking methods of food based on ingredient characteristics compared to methods that do not use mapping information. This enables the model to make more sophisticated predictions and decisions, and strengthens its inference capabilities regarding new ingredients or recipes, which can aid in predicting the characteristics of unknown ingredients or new recipes lacking user feedback. Furthermore, using ingredient-function mapping information allows the model to learn knowledge about the interactions and characteristics between ingredients more quickly, thereby improving overall learning efficiency. Similarly, if user feedback is related to the aforementioned mapping information, it enables the model to reflect user preferences or needs more accurately.
[0139] Therefore, a reinforcement learning-based recipe optimization model utilizing material feature mapping information has advantages in various processes, such as optimized recipe generation, user feedback processing, model training, and inference.
[0140] In some embodiments of the present disclosure, the processor (110) may perform a step (S100) of initialization or pre-training to form initial knowledge based on mapping information in relation to the step S110. Such mapping information may be collected in a database in advance and may be updated by reflecting user feedback or external databases such as the Ministry of Food and Drug Safety database or a standard database based on papers. For example, S100 may acquire mapping information using information regarding the type of material, the amount of material, and the function of the material, and pre-train a recipe optimization model to form initial knowledge using this as pre-training data. The pre-training method does not distinguish between supervised learning, unsupervised learning, reinforcement learning, etc.
[0141] Here, the meaning of forming initial knowledge based on the above mapping information may vary depending on the specific learning method of the learning model. If the recipe optimization model adopts a reinforcement learning approach, the above mapping information can directly or indirectly represent the relationship between actions regarding the recipe and the corresponding rewards. By utilizing this, the initial values of each state and action can be estimated using the mapping information, thereby enabling the agent to perform more effective exploration when starting learning. For example, in the case of Q-learning, this means setting the Q-value to a non-zero value based on the above mapping information.
[0142] Alternatively, there may be cases where the prior learning is supervised learning. Supervised learning can be performed using material type information or combinations of material type and quantity information as input data, and material function information corresponding to those combinations as output data. In other words, supervised prior learning may mean obtaining a training set labeled with combinations of material type and quantity and material function information based on mapping information, and then performing prior learning on this. However, it is not limited to this.
[0143] When a recipe optimization model performs reinforcement learning after such prior training, it can be helpful in implementing the recipe optimization model because the pre-initialized knowledge is utilized during the reinforcement learning search process, even if user feedback is insufficient. Alternatively, while exploration can be performed by referring to this mapping information initially, it is also possible to use an epsilon greedy method, in which the reinforcement learning agent learns based on its own experience while also adopting a method of finding new possibilities by performing exploration with a certain probability. However, it is not limited to this.
[0144] According to some embodiments of the present disclosure, initial recipe improvement information, which is a recipe improved based on user requirements for a basic recipe, can be obtained based on a recipe optimization model learned through a step (S100) of initializing or pre-learning to form initial knowledge based on mapping information.
[0145] According to some embodiments of the present disclosure, when the recipe optimization model is a learning model trained through reinforcement learning, the step (S110) of collecting recipe improvement information and user evaluation information regarding at least one recipe may include an initialization step (S100) for reinforcement learning.
[0146] In this initialization phase, recipe improvement information and user evaluation information regarding at least one recipe are separated into the state and action of each recipe, and the reward for all possible states and actions may be set to 0 to allow the user to explore the environment by acting randomly at first, or the initial reward value may be set based on user evaluation information regarding recipes already collected to induce the user to prioritize the use of recipes with good user feedback from the early stages of learning, or a type of model-based reinforcement learning may be performed by using prior knowledge such as the mapping information described below as a model to limit the range to recipe states containing ingredients related to user feedback and actions to change those ingredients using the mapping information. However, this is not limited thereto.
[0147] According to some embodiments of the present disclosure, if the recipe optimization model is a learning model trained through reinforcement learning, the processor (110) may determine an action based on material-function mapping information, which is information regarding the relationship between the combination of ingredients in a recipe and changes in ingredient functions, in relation to step S130. The mapping information is information indicating how the type of each ingredient, the amount of ingredients, cooking conditions, etc., affect the taste, aroma, texture, etc., of the entire recipe, and may include information such as, for example, at what weight ratio sugar provides a specific sweetness intensity.
[0148] In one embodiment, in the strategy for determining an action regarding the current state of the recipe optimization model, mapping information may be applied to both the exploration strategy, which attempts new things, and the exploitation strategy, which attempts to make an experience-based selection. For example, when the exploration strategy transitions to a random state, mapping information may be referenced to restrict the search for the random state to an action that transitions to a state among the current states that has a material function related to previous user feedback and a mapped material type, or in the exploitation strategy, mapping information may be used as a virtual reward for an inexperienced state and action to initialize.
[0149] According to some embodiments of the present disclosure, if the recipe optimization model is a learning model trained through reinforcement learning, the processor (110), in relation to step S120, can use mapping information to consider how an agent affects the overall functional characteristics of the recipe (e.g., taste, nutritional value, texture, etc.) when finding an action policy that matches user feedback. In other words, the agent can decide how to combine and change an ingredient based on the mapping information. A reward based on user evaluation information can be viewed as an indicator of user satisfaction appearing in the feedback. For example, feedback such as "It is too sweet" implies that the high intensity of sweetness has lowered user satisfaction. Therefore, to reflect this as negative feedback in the action policy, the agent can use mapping information to find the type of ingredient that has the ingredient function of sweetness, and can learn a related action policy to replace the ingredient or reduce the amount of the ingredient. If, upon receiving such user feedback, the action of increasing the amount of sugar receives a negative reward, the agent will not prefer the action of increasing the amount of sugar in the future. However, this is not limited thereto.
[0150] Hereinafter, a method is disclosed in which a processor (110) obtains an optimized recipe using a recipe optimization model learned using a learning method according to some embodiment of the present disclosure, and then provides it using a user interface.
[0151] The processor (110) provides the acquired optimization recipe to the user through a user interface and can provide the type of ingredient, amount of ingredient, ingredient composition, ingredient price, and ingredient function using attribute information regarding the optimization recipe, and can process and provide this information using an external database. For example, the processor (110) can generate a report including nutritional content, environmental friendliness score, stability-related information, and ingredient price information by referring to the attribute information of the acquired optimization recipe and an external database, and provide it to the user.
[0152] Hereinafter, an embodiment is disclosed in which a processor (110) performs recipe optimization using a recipe optimization model learned using a learning method according to some embodiments of the present disclosure.
[0153] The processor (110) can obtain recipe information indicating the state of a target recipe that requires improvement, and perform recipe optimization on a new recipe based on a recipe optimization model. Here, the recipe optimization model may correspond to a model trained based on recipe improvement information and user evaluation information regarding at least one existing recipe to derive a recipe optimized for user evaluation corresponding to a recipe change.
[0154] At this time, the recipe optimization model may preferably be implemented using a reinforcement learning method pre-trained with information regarding the relationship between the ingredient combination and functional change of the above-described recipe, but is not limited thereto and various learning methods may be used.
[0155] According to some embodiments of the present disclosure, the process (110) may perform optimization of a new recipe by utilizing a learned recipe optimization model. A new recipe is input into its current state using an already learned model, and as a result, an optimal action is derived to obtain an improved recipe, where no additional learning occurs, and the learned model merely derives the optimal action for the current state.
[0156] Meanwhile, the process (110) can evaluate the performance of the learned model by inputting attribute information of a test recipe to test the current recipe optimization model and receiving a user evaluation of the state output by the model. At this time, it is desirable that the test recipe be different from the existing set of learned recipes.
[0157] Hereinafter, a specific embodiment of the present disclosure will be described.
[0158] For example, there may be a cycle in which 250g of chicken is provided to the user as an improved recipe (S_2) of 200g of chicken, which is a basic recipe (S_1), and a reinforcement learning model is trained with a reward (R_1) based on user feedback regarding increasing the amount of chicken by 50g (action A_1) in the basic recipe state (S_1), thereby continuously providing the user with an improved recipe (S_t+1) through the process of obtaining a new improved recipe (S_t) for the previous recipe state.
[0159] State S_1: Chicken 200g, Action A_1: Increase chicken amount by 50g, receive user feedback and reward R_1: Positive feedback (+1), New State S_2: Chicken 250g.
[0160] State S_2: Chicken 250g, Action A_2: Increase chicken amount by 50g, receive user feedback and reward R_2: Neutral feedback (0), New State S_3: Chicken 300g.
[0161] State S_3: Chicken 300g, Action A_3: Decrease chicken amount by 50g, receive user feedback and reward R_3: Positive feedback (+1), New State S_4: Chicken 250g.
[0162] State S_4: Chicken 250g, Action A_4: Decrease chicken amount by 50g, receive user feedback and reward R_4: Negative feedback (-1), New State S_5: Chicken 200g.
[0163] State S_5: Chicken 200g, Action A_5: Increase chicken amount by 50g, receive user feedback and reward R_5: Positive feedback (+1), New State S_6: Chicken 250g.
[0164] When the above process was performed using Q-learning among reinforcement learning algorithms, the following update formula was used, and the calculation was performed assuming α=0.5 and γ=0.9.
[0165] Q(s_t, a_t) ηa_t) + α * [r_t+1 + γ * max_a Q(s_t+1, a) - Q(s_t, a_t)]
[0166] Q(S_1, A_1) = 0.5, Q(S_2, A_2) = 0, Q(S_3, A_3) = 0.5, Q(S_4, A_4) = -0.5, Q(S_5, A_5) = 0.5
[0167] If the training of the reinforcement learning model in the above learning process is terminated according to predetermined conditions such as performance or convergence, the recipe of 250g chicken can be obtained as the optimized recipe based on the recipe optimization model.
[0168] Although it has been described that the present disclosure may generally be executed by a computer device, those skilled in the art will recognize that the present disclosure may be implemented in combination with computer-executable instructions and other program modules that can operate on one or more computers. It may also be implemented by a combination of hardware and software.
[0169] A program module generally includes functions, programs, components, data structures, etc., that perform specific tasks or implement specific abstract data types. Furthermore, those skilled in the art will be aware that the method presented in this disclosure can be executed in various computer system configurations, such as single-processor or multi-processor computer systems, minicomputers, mainframe computers, personal computers, portable computing devices, microprocessor-based or programmable consumer electronics.
[0170] The embodiments described in this disclosure may also be executed in a distributed computing environment where some operations are performed by remote processing devices connected via a communication network. In this distributed computing environment, program modules may be located in both local memory storage and remote memory storage.
[0171] Computers generally include various computer-readable media. Media accessible to a computer may be considered computer-readable media, and these media include volatile and non-volatile media, transient and non-transient media, and removable and fixed media. For example, computer-readable media may include computer-readable storage media and computer-readable transmission media. Computer-readable storage media are media implemented using various methods and technologies for storing information, and these media may include RAM, ROM, EEPROM, flash memory, CD-ROM, digital video disc (DVD), magnetic cassette, magnetic tape, magnetic disk storage device, etc.
[0172] Computer-readable transmission media are media that implement computer-readable instructions, data structures, program modules, etc., via a carrier wave or other transmission mechanism, and include all information transmission media. The term "modulated data signal" refers to a signal in which information is encoded by setting or changing one or more of the characteristics of the signal. For example, computer-readable transmission media include wired networks, direct connection lines, acoustic, RF, infrared, and other wireless media. Any combination of these media may be included within the scope of computer-readable transmission media.
[0173] An example of an environment (1100) according to the present disclosure includes a computer (1102), which has a processing unit (1104), a system memory (1106), and a system bus (1108). Here, the system bus (1108) serves to connect various system components, particularly the system memory (1106), to the processing unit (1104). The processing unit (1104) can be any of various commercial processors, and a dual processor or other multiprocessor architecture can also be used as the processing unit (1104).
[0175] The system bus (1108) can be interconnected with other elements through various forms of local buses, such as memory buses, peripheral buses, and various types of commercial bus architectures. The system memory (1106) includes read-only memory (ROM) (1110) and random access memory (RAM) (1112). The BIOS is stored in the ROM, EPROM, EEPROM, etc., as non-volatile memory (1110), and the BIOS is responsible for basic routines for exchanging information between components of the computer (1102). The RAM (1112) may also include static RAM for data caching, such as high-speed RAM.
[0176] This computer (1102) also includes an internal hard disk drive (HDD) (1114), a magnetic floppy disk drive (FDD) (1116), and an optical disk drive (1120). These can each be connected to the system bus (1108) via a hard disk drive interface (1124), a magnetic disk drive interface (1126), and an optical drive interface (1128). Some of these interfaces use USB or IEEE 1394 interface technology for external drives.
[0177] The computer-readable media associated with these drives provide data, data structures, computer-executable instructions, etc., in the form of non-volatile storage. From the perspective of the computer (1102), these drives and media serve to store all data in a suitable digital format. Removable optical media such as HDDs, removable magnetic disks, CDs, or DVDs may be examples of the computer-readable media of the present disclosure, but other types of media such as Zip drives, magnetic cassettes, flash memory cards, cartridges, etc. may also be used. These media may contain computer-executable instructions necessary to execute the method of the present disclosure.
[0178] Various program modules, an operating system (1130), one or more application programs (1132), other program modules (1134), program data (1136), etc. may be stored in the drive and RAM (1112). Some or all of these may be cached in RAM (1112). The present disclosure may be implemented in various commercial operating systems or combinations thereof.
[0179] The user can input commands or information into the computer (1102) through input devices such as a keyboard (1138) and a mouse (1140). Other input devices such as a microphone, IR remote control, joystick, game pad, stylus pen, and touch screen may also be used. These input devices are primarily connected to the system bus (1108) via an input device interface (1142), which is connected to the processing unit (1104). This connection can be made through various interfaces such as a parallel port, an IEEE 1394 serial port, a game port, a USB port, and an IR interface.
[0180] The computer (1102) can be connected to a monitor (1144) or other display device via an interface such as a video adapter (1146) through a system bus (1108). In addition, it can have various peripheral output devices such as speakers and printers built in.
[0181] This computer (1102) can be connected via wired or wireless communication by forming a network with remote computer(s) (1148). The remote computer(s) (1148) can take various forms, such as workstations, personal computers, laptops, gaming devices, peer devices, etc., and may include some or all of the components of the computer (1102). However, for simplicity, only the memory storage device (1150) is shown. This connection can be formed as a network such as a LAN (1152) or WAN (1154), which is a networking environment commonly used in offices or companies. Such a network can form a corporate network such as an intranet and, through it, connect to a global computer network, for example, the Internet.
[0182] In a LAN, a computer (1102) may be connected to a local network (1152) via a wired or wireless communication network interface or adapter (1156). The adapter (1156) enables wired or wireless communication to the LAN (1152), and the LAN (1152) includes a wireless access point installed to communicate with the wireless adapter (1156). In a WAN, the computer (1102) is connected to another computing device on the WAN (1154) via a modem (1158) or establishes communication over the Internet. This modem (1158) may be internal or external and may be wired or wireless, and is connected to the system bus (1108) via a serial port interface (1142). In a network environment, the program module or part described in the computer (1102) may be stored in a remote memory / storage device (1150). The described network connection is an example, and other methods of establishing communication links between computers are also possible.
[0183] The computer (1102) can communicate with various wireless devices such as printers, scanners, desktop and / or portable computers, PDAs, communication satellites, wireless tags, etc. This includes at least Wi-Fi and Bluetooth wireless technologies. Thus, the communication may follow a standard network structure or be ad hoc communication between at least two devices.
[0184] Wi-Fi is a technology that enables wireless access to the Internet. It is a technology similar to cellular phones that allows computers to transmit and receive data anywhere, indoors or outdoors. Wi-Fi networks use IEEE 802.11 (a, b, g, and others) wireless technology to provide secure, reliable, and fast wireless connections. Wi-Fi can be used to connect computers to each other, to the Internet, or to wired networks (using IEEE 802.3 or Ethernet). Wi-Fi networks operate in the unlicensed 2.4 and 5 GHz wireless bands and can operate at data speeds of 11 Mbps (802.11a) or 54 Mbps (802.11b), or on products that support both bands.
[0185] As is generally known in the technical field of the present disclosure, data, instructions, commands, information, signals, bits, symbols, chips, etc., may be expressed in various forms such as voltage, current, electromagnetic waves, magnetic fields, optical fields, particles, etc.
[0186] A person skilled in the art to which this disclosure is made will recognize that, in connection with the implementation examples described herein, various logic blocks, modules, processors, means, circuits, and algorithm steps may be implemented by electronic hardware, various forms of programs (referred to herein as software for convenience), design code, or a combination thereof. To clarify this interaction between hardware and software, various exemplary components, blocks, modules, circuits, and steps have been previously described for their functions. Whether these functions are implemented in hardware or software depends on the design constraints imposed on the specific application and the overall system. A person skilled in the art to which this disclosure is made may implement the functions described for each specific application in various ways, but such implementation decisions should not be construed as exceeding the scope of this disclosure.
[0187] The various implementations proposed in this disclosure may be realized as methodologies, devices, or products utilizing standard programming and / or engineering techniques. The term “product” includes computer programs, carriers, or media accessible on any computer-readable storage device. For example, computer-readable storage media include, but are not limited to, magnetic storage devices (e.g., hard disks, floppy disks, magnetic tapes, etc.), optical discs (e.g., CDs, DVDs, etc.), smart cards, and flash memory devices (e.g., EEPROMs, cards, sticks, key drives, etc.). Additionally, the various storage media mentioned herein include one or more devices and / or other machine-readable media for storing information.
[0188] It should be noted that the specific order or hierarchy of the steps in the presented processes is merely part of an exemplary approach. It should be noted that, depending on design priorities, the specific order or hierarchy of the steps in the processes may be rearranged within the scope of this disclosure. The appended method claims provide elements of various steps in a sample order, but are not limited to the specific order or hierarchy presented.
[0189] The description of the presented embodiments helps a person skilled in the art to use or practice the present disclosure. Various variations of these embodiments will be apparent to a person skilled in the art, and the general principles defined in the present disclosure may be applied to other embodiments beyond the scope of the present disclosure. Accordingly, the present disclosure is not limited to the embodiments presented herein but should be interpreted to the broadest possible extent in which the principles and novel features presented herein are consistently applied.
Claims
Claim 1 A method for training a recipe optimization model and providing an improved recipe, performed by a computing device comprising at least one processor, comprising: a step of pre-training the recipe optimization model to form initial knowledge based on mapping information including the relationship between the ingredient combination and functional change of the recipe; a step of training the recipe optimization model with a training set obtained based on recipe improvement information and user evaluation information; wherein the training step comprises reinforcing the model to find an optimal action policy to determine which action to take for a given recipe state to receive the maximum reward based on action information related to the transition of at least one recipe state, recipe improvement information according to the action, and user evaluation information related to the reward according to the action; wherein the reinforcement learning is characterized by being trained to determine the action based on the mapping information when determining the action. Claim 2 delete Claim 3 delete Claim 4 A method according to claim 1, wherein the recipe improvement information includes one or more changes to information regarding ingredient type, ingredient composition, ingredient function, ingredient quantity, cooking conditions, and recipe reviews for each recipe, and the user evaluation information includes one or more information regarding user behavior and user sensory evaluation. Claim 5 A method according to claim 1, wherein the step of training a recipe optimization model with a learning set obtained based on recipe improvement information and user evaluation information for a recipe to derive a recipe optimized for user evaluation corresponding to the recipe improvement includes the step of obtaining an improved recipe of a target recipe based on the recipe optimization model. Claim 6 A method according to claim 1, wherein the step of training a recipe optimization model with a learning set obtained based on recipe improvement information and user evaluation information for a recipe to derive a recipe optimized for user evaluation corresponding to the recipe improvement includes the step of updating the recipe optimization model with a learning set obtained based on recipe improvement information and user evaluation information for a target recipe. Claim 7 A method according to claim 1, wherein the step of training a recipe optimization model with a learning set obtained based on recipe improvement information and user evaluation information for a recipe to derive a recipe optimized for user evaluation corresponding to the recipe improvement comprises: a step of updating the recipe optimization model with a learning set obtained based on recipe improvement information and user evaluation information for a first target recipe; and a step of obtaining an improved recipe for a second target recipe based on the updated recipe optimization model. Claim 8 A method according to claim 1, wherein the step of training a recipe optimization model with a learning set obtained based on recipe improvement information and user evaluation information for a recipe to derive a recipe optimized for user evaluation corresponding to the recipe improvement comprises: a step of obtaining an improved recipe for a target recipe based on the recipe optimization model; a step of updating the recipe optimization model with a learning set obtained based on recipe improvement information and user evaluation information for the improved recipe; and a step of obtaining an additional improved recipe for the improved recipe based on the updated recipe optimization model. Claim 9 delete Claim 10 A method according to claim 1, wherein user evaluation information for the recipe is related to mapping information including the relationship between the combination of ingredients and functional changes of the recipe. Claim 11 A method according to claim 1, wherein the step of training a recipe optimization model with a training set obtained based on recipe improvement information and user evaluation information for a recipe to derive a recipe optimized for user evaluation corresponding to the recipe improvement includes the step of training the recipe optimization model with a training set obtained based on recipe improvement information for a recipe improved based on user requirements for a basic recipe and user evaluation information for the improved recipe. Claim 12 A computer program stored on a computer-readable storage medium, wherein the computer program, when executed on at least one processor of a device, performs operations to train a recipe optimization model and provide an improved recipe, the operations comprising: a pre-training operation to form initial knowledge based on mapping information including the relationship between the ingredient combination and functional change of a recipe; a training operation to train the recipe optimization model with a training set obtained based on recipe improvement information and user evaluation information; in the training operation, a reinforcement training operation to find an optimal action policy to determine which action to take for a given recipe state to receive maximum reward based on action information related to the transition of at least one recipe state, recipe improvement information according to the action, and user evaluation information related to the reward according to the action; and in the reinforcement training, a training operation to train the recipe optimization model to determine the action based on the mapping information when determining the action. Claim 13 A computing device comprising: at least one processor that executes at least one program instruction; and a memory coupled to the at least one processor; wherein the memory stores a program instruction for training a recipe optimization model and providing an improved recipe, and the at least one processor, by executing the program instruction: pre-training the recipe optimization model to form initial knowledge based on mapping information including the relationship between the ingredient combination and functional change of a recipe; training the recipe optimization model with a training set obtained based on recipe improvement information and user evaluation information; and, in the training process, reinforcing the recipe optimization model to find an optimal action policy to determine which action to take to receive the maximum reward based on action information related to the transition of at least one recipe state, recipe improvement information according to the action, and user evaluation information related to the reward according to the action, and, during the reinforcement training, training the recipe optimization model to determine the action based on the mapping information when determining the action. Claim 14 A method for optimizing a recipe, performed by a computing device comprising at least one processor, comprising the step of obtaining an improved recipe for a collected target recipe based on a recipe optimization model, wherein the recipe optimization model is pre-trained to form initial knowledge based on mapping information including the relationship between the ingredient combination and functional change of the recipe, and is trained with a learning set obtained based on recipe improvement information and user evaluation information, and is reinforced learning to find an optimal action policy that determines which action to take for a given recipe state to receive the maximum reward based on action information related to the transition of at least one recipe state, recipe improvement information according to said action, and user evaluation information related to the reward according to said action, and corresponds to a model trained such that when determining an action in said reinforcement learning, said action is determined based on said mapping information.
Citation Information
Patent Citations
Cooking support service providing method and system
JP6615163B2
Method for providing customized recipe service
KR1020170137345A
System and method for recommending recipe
KR1020230039012A
Artificial intelligence (AI)-based contactless QSC inspection solution system
WO2022055030A1
