Qualified ketogenic diet evaluation method and application thereof

Abstract

The invention discloses a qualified ketogenic diet evaluation method and application thereof, and belongs to the technical field of healthy diet evaluation.The method comprises the steps that diet data recorded by multiple users are obtained and preprocessed to obtain a data set, a heterogeneous graph of the data set is constructed, and a ketogenic diet evaluation result is obtained; a graph neural network is adopted to carry out characterization analysis on the composition to obtain a knowledge graph corresponding to the recipe information, a message propagation network is adopted to carry out optimization analysis on the knowledge graph according to the recipe information to obtain a user health model, and qualified ketogenic diet evaluation is carried out on each user based on the user health model and a preset evaluation index to obtain an evaluation result; through multi-dimensional data collection and analysis, the evaluation quality and accuracy of the ketogenic diet are improved, personalized healthy diet recommendation is provided for the user, and thus powerful support is provided for reasonable application and effective management of the ketogenic diet.

Description

Technical Field

[0001] This invention belongs to the field of healthy diet assessment technology, and in particular relates to a method for evaluating a qualified ketogenic diet and its application. Background Technology

[0002] The ketogenic diet therapy (KDT) is a formulated diet characterized by high fat, very low carbohydrates, and moderate protein and other nutrient intake. In recent years, it has gained widespread attention in areas such as weight management, diabetes control, and as an adjunct to epilepsy treatment. It achieves various health benefits by altering the body's energy metabolism pathways, shifting the body from relying on carbohydrates for energy to relying on fat. However, the effectiveness and safety of the KDT are highly dependent on the accuracy of its implementation and individual adaptation. Accurate implementation of the KDT requires flexible control of the proportions of fat, protein, and carbohydrates, while monitoring the body's metabolic response. Individual differences, such as metabolic rate, underlying health conditions, and lifestyle, also affect the effectiveness and safety of the KDT. Currently, methods for assessing the quality of KDT implementation have significant limitations. Common evaluation methods mainly focus on single indicators, and most rely on manual surveys or sampling for recording or assessment, such as blood ketone levels, failing to comprehensively consider the multifaceted impact of the KDT on individual growth and development, psychological state, nutritional balance, and potential risks. This single-indicator evaluation method is difficult to accurately reflect the overall implementation effect and potential problems of the KDT, hindering scientific dietary adjustments and risk management. Therefore, there is an urgent need to provide a qualified KDT evaluation method and its application to solve the aforementioned technical problems. Summary of the Invention In view of this, the present invention provides a qualified ketogenic diet evaluation method and its application. Through multi-dimensional data collection and analysis, it improves the evaluation quality of the ketogenic diet and provides users with personalized healthy diet recommendations to solve the above-mentioned technical problems. The specific technical solution adopted is as follows.

[0003] On the one hand, the present invention provides a method for evaluating a qualified ketogenic diet, comprising the following steps: The diet data of multiple users is obtained and preprocessed to obtain a dataset, wherein the diet data includes user information and recipe information; A heterogeneous graph of the dataset is constructed, and a graph neural network is used to perform representation analysis on the heterogeneous graph to obtain a knowledge graph corresponding to the recipe information. The heterogeneous graph includes users, recipes, ingredients, and calories. Based on the recipe information, a message propagation network is used to optimize and analyze the knowledge graph to obtain a user health model. The message propagation network includes an embedding layer, a higher-order propagation layer, and a prediction layer. The embedding layer is used to provide embedded representations for all recipes, users, and relationships. The higher-order information propagation layer includes information propagation and information aggregation, used to propagate and aggregate information between different entities. The prediction layer is used to rank the importance between entities and users and predict the matching score between recipes and users. The evaluation results are obtained by evaluating each user's ketogenic diet based on the user health model and preset evaluation indicators. The preset evaluation indicators include growth and development information, psychological state information, food diversity information, metabolic index information, and adverse reaction information.

[0004] As a preferred embodiment of the above technical solution, constructing the heterogeneous graph of the dataset includes: Preset a user-recipe pair Recipe i has a set of ingredients. Image features and calorie factor A heterogeneous graph is constructed to analyze the complex relationships between users, recipes, ingredients, and calories. The heterogeneous graph is used to predict the probability of user u interacting with recipe i. The graph structure allows the model to capture different types of nodes and various types of edges between nodes. The self-supervised ingredient prediction module is used to analyze the relationships between ingredients and predict the recipe set for recipe i. The knowledge graph is obtained by learning representations of the masked components, wherein a multi-head attention model is used to recover components given the surrounding context. The masked components.

[0005] As a preferred embodiment of the above technical solution, constructing the heterogeneous graph of the dataset includes: Preset Represents a set of users The set of recipes is represented by the user-recipe interaction matrix. ,in, This indicates whether user u interacts with recipe i, and each recipe i has associated image features. Ingredient Set and heat factor Ingredient Set It is a multi-hot encoded vector. If recipe i contains ingredient k, then the corresponding... Otherwise, it is 0; heat factor It is a single-heat encoded vector used to represent the calorie level of recipe i, where These represent the quantities of different calorie levels, and the corresponding graph neural network model is: Input: User-recipe interaction matrix Y, recipe ingredient matrix Each row corresponds to a recipe, and the number of columns represents the total number of possible ingredients. (Recipe image feature matrix) Recipe calorie information matrix ; Output: Interactive function , used to predict the probability of user u interacting with recipe i.

[0006] As a preferred embodiment of the above technical solution, a heterogeneous graph is constructed to analyze the complex relationships between users, recipes, ingredients, and calories, including: If recipe i contains ingredients If its calorie attribute is c, then connect an ingredient node and a calorie node to the recipe node; If user u selects recipe i, then there is an edge between the recipe node and the user node; Use a single embedding To represent each node, where Let d represent the node, d represent the embedding size, and let d represent the one-hot encoding of user u. Projecting it into the following embedded expression is: (1) in, It is the user's embedding, If the learnable embedding matrix is ​​the matrix of all users, then all nodes maintain an embedding matrix represented as follows: ,in, , , and These represent the number of users, recipes, ingredients, and calorie content, respectively.

[0007] As a preferred embodiment of the above technical solution, a graph neural network is used to perform characterization analysis on the heterogeneous graph to obtain a knowledge graph corresponding to the recipe information, including: Preset calorie perception user representation Recipe image features extracted by pre-trained network and embedding of target components The image features are converted into d-dimensional embeddings using a mapping layer, and the corresponding expression is: (2) Wherein, image embedding is denoted as , , For the parameters of the mapping layer, a feedforward network is used to calculate the attention weights of each component: (3) in, , , These are the parameters to be learned. Tanh is used as the non-linear activation function, and the embeddings of the components are fused through attention weights to obtain personalized component representations. (4) Based on the user's preference for different recipe information, the characteristics of the ingredients are integrated. Then, the characterization of the different ingredients in the recipe. , , , A feedforward network is used to calculate the attention weights for different components: (5) Where q represents a component feature, and component-level attention is used to aggregate the representations of different components into a comprehensive recipe feature; , , These are the parameters to be learned; By integrating different component representations through attention weights, we obtain the user-oriented summative recipe features: (6) in, This indicates that the personalized elements focus on the user.

[0008] As a preferred embodiment of the above technical solution, a user health model is obtained by optimizing and analyzing the knowledge graph using a message propagation network based on the recipe information, including: We employ high-order connectivity of knowledge graphs for high-order feature propagation, and utilize the local connectivity of the graph structure and the neighbor information of nodes to capture the relationships and features between nodes. We aggregate neighbor information through multi-layer GCN stacking to obtain a global graph structure representation. The corresponding computational expression is: (7) in, Represents a non-linear activation function. This represents the input features of the k-th layer. This represents the output feature of the k-th layer. The degree matrix represents the degree of the self-linked matrix. Represents a linear transformation matrix. Where A represents the adjacency matrix of the target entity. Represents the identity matrix; Set all sampling neighbor sizes to Where n represents the number of samples, the expression for the attention weight between the head entity h and the tail entity t connected by the relation r is: (8) Normalize formula (3) using the softmax function: (9) in, Indicates attention weights. This represents the activation function. It is a weight vector. It is a weight matrix. Indicates a connection operation. Represents a user entity vector. Represents a relation vector; target node The representation is based on the aggregation of the user and its neighbor nodes, and the corresponding calculation expression is: (10) in, express The neighborhood aggregation is represented as, Represents the entity vector of neighboring nodes. Represents the vector of the user entity connected to it; Based on the user's self-network To represent the embedded representation of the user's domain, yes Standardized weights: (11).

[0009] As a preferred embodiment of the above technical solution, the target entity o and its neighboring entities are... Combining them to perform a nonlinear transformation, the corresponding expression is: (12) in, Representation of the neighborhood of entities and users, concat means join, concat means connection. Represents a nonlinear transformation. Indicate target The final representation of the domain; After the recipe information is propagated through h layers and messages are aggregated, the new entity representations of the higher-order neighbors can be obtained. The expression for the interaction probability is derived using the inner product form: (13) in, This represents the final user u obtained after aggregating the neighborhood of user interactions. This represents the final entity obtained through h-order propagation. The representation, PRE, is derived from the user representation and the final prediction function. for The preference probability, and the corresponding loss function is: (14) in, Represents cross-entropy loss, This represents the positive sample set. Indicates a negative sample. This represents the L2 regularizer. This represents the Lagrange coefficient.

[0010] As a preferred embodiment of the above technical solution, an evaluation result is obtained by assessing each user's eligibility for a qualified ketogenic diet based on the user health model and preset evaluation indicators, including: Hit Ratio@K is used to evaluate the recommendation hit rate of the user health model, where K represents the top K items in the KDT table. The hit rate of each user is accumulated, and the accumulated result is divided by the total number of users to obtain the value of Hit Ratio@K. The more items on the ketogenic diet chart that match the user's preferences, the higher the Hit Ratio@K value and the higher the hit rate. Conversely, a lower hit rate indicates fewer items in the assessment results that meet the qualified KDT. The corresponding expression is: (15) in, This represents a list of eligible KDT items of length K returned by the user health model. This represents the list of items that users actually interacted with in the test set, where N is the total number of users.

[0011] As a preferred embodiment of the above technical solution, the growth and development information is used to assess the impact of the ketogenic diet on the user's normal growth and development or the maintenance of healthy body composition; the psychological state information is used to assess the user's mood, appetite control, and satisfaction with the diet during the implementation of the ketogenic diet; the food diversity information is used to assess the richness of the variety of foods consumed daily, whether bowel movements are regular and normal, and the overall nutritional balance of the diet; the metabolic indicators are used to assess biochemical markers of ketosis, including blood ketone levels, urine ketone levels, and glucose / ketone ratio; and the adverse reaction information is used to assess whether significant adverse symptoms or health problems, constipation, indigestion, risk or occurrence of gallstones, and negative nitrogen balance occur during the implementation of the ketogenic diet.

[0012] On the other hand, the present invention also provides the application of the evaluation method described above in the evaluation and analysis of a qualified ketogenic diet.

[0013] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a method for evaluating a qualified ketogenic diet and its application. It involves acquiring dietary data recorded by multiple users, preprocessing the data to obtain a dataset, constructing a heterogeneous graph of the dataset, and using a graph neural network to represent and analyze the graph to obtain a knowledge graph corresponding to the recipe information. Based on the recipe information, a message propagation network is used to optimize and analyze the knowledge graph to obtain a user health model. Based on the user health model and preset evaluation indicators, a qualified ketogenic diet evaluation is conducted for each user to obtain the evaluation result. Through multi-dimensional data collection and analysis, the evaluation quality and accuracy of the ketogenic diet are improved, providing users with personalized healthy diet recommendations, thereby providing strong support for the rational application and effective management of the ketogenic diet. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart of the qualified ketogenic diet evaluation method provided by the present invention; Figure 2 A flowchart of an evaluation method according to another embodiment of the present invention; Figure 3 This is a structural block diagram of the comprehensive score in the evaluation method provided by the present invention. Detailed Implementation

[0016] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0017] See Figure 1 This invention provides a method for evaluating a qualified ketogenic diet, comprising the following steps: S1: Obtain dietary data from multiple user records and preprocess the dietary data to obtain a dataset, wherein the dietary data includes user information and recipe information; S2: Construct a heterogeneous graph of the dataset, and use a graph neural network to perform representation analysis on the heterogeneous graph to obtain a knowledge graph corresponding to the recipe information, wherein the heterogeneous graph includes users, recipes, ingredients and calories; S3: Based on the recipe information, the knowledge graph is optimized and analyzed using a message propagation network to obtain a user health model. The message propagation network includes an embedding layer, a higher-order propagation layer, and a prediction layer. The embedding layer provides embedded representations for all recipes, users, and relationships. The higher-order propagation layer includes information propagation and information aggregation, used to propagate and aggregate information between different entities. The prediction layer is used to rank the importance between entities and users and predict the matching score between recipes and users. S4: Based on the user health model and preset evaluation indicators, a qualified ketogenic diet evaluation is conducted for each user to obtain the evaluation results. The preset evaluation indicators include growth and development information, psychological state information, food diversity information, metabolic indicator information, and adverse reaction information.

[0018] In this embodiment, constructing the heterogeneous graph of the dataset includes: pre-setting a user-recipe pair. Recipe i has a set of ingredients. Image features and calorie factor A heterogeneous graph is constructed to analyze the complex relationships between users, recipes, ingredients, and calories. This graph is used to predict the probability of user u interacting with recipe i. The graph structure allows the model to capture different types of nodes and various types of edges between them. A self-supervised ingredient prediction module is used to analyze the relationships between ingredients by predicting the recipe set for recipe i. The knowledge graph is obtained by learning representations of the masked components, wherein a multi-head attention model is used to recover components given the surrounding context. The masked components. The growth and development information is used to assess the impact of the ketogenic diet on the user's normal growth and development or maintenance of healthy body composition; the psychological state information is used to assess mood, appetite control, and satisfaction with the diet during the implementation of the ketogenic diet; the food diversity information is used to assess the richness of daily food intake, regularity of bowel movements, and overall nutritional balance of the diet; the metabolic indicators are used to assess biochemical markers of ketosis, including blood ketone levels, urine ketone levels, and glucose / ketone ratio; the adverse reaction information is used to assess whether significant adverse symptoms or health problems, constipation, indigestion, risk or occurrence of stones, and negative nitrogen balance occur during the implementation of the ketogenic diet.

[0019] It should be noted that constructing the heterogeneous graph of the dataset includes: Preset Represents a set of users The set of recipes is represented by the user-recipe interaction matrix. ,in, This indicates whether user u interacts with recipe i, and each recipe i has associated image features. Ingredient Set and heat factor Ingredient Set It is a multi-hot encoded vector. If recipe i contains ingredient k, then the corresponding... Otherwise, it is 0; heat factor It is a single-heat encoded vector used to represent the calorie level of recipe i, where These represent the quantities of different calorie levels, and the corresponding graph neural network model is: Input: User-recipe interaction matrix Y, recipe ingredient matrix Each row corresponds to a recipe, and the number of columns represents the total number of possible ingredients. (Recipe image feature matrix) Recipe calorie information matrix ; Output: Interactive function , used to predict the probability of user u interacting with recipe i.

[0020] This approach involves constructing a heterogeneous graph to explicitly explore the complex relationships between users, recipes, ingredients, and calories. This graph structure allows the model to capture different types of nodes (users, recipes, ingredients) and various types of edges between them. The connectivity of the graph nodes can be leveraged to explore potential relationships by propagating embedding vectors from calories to users, with recipes acting as bridges. Graph Neural Networks (GNNs) can propagate and update node embedding vectors within the graph, enabling the model to learn complex interaction patterns between nodes. Information from calorie nodes can propagate to recipe nodes via edge connections, and then to user nodes. By considering user, recipe, ingredient, and calorie nodes simultaneously in the graph, the model can more comprehensively capture user dietary behavior. For example, the model can identify which ingredient or recipe features are associated with a user's calorie awareness, thus providing more personalized recommendations. By acquiring dietary data from multiple users and preprocessing the data to obtain a dataset, a heterogeneous graph of the dataset is constructed. A graph neural network is then used to represent and analyze the graph to obtain a knowledge graph corresponding to the recipe information. Based on the recipe information, a message propagation network is used to optimize and analyze the knowledge graph to obtain a user health model. Based on the user health model and preset evaluation indicators, each user is evaluated for a qualified ketogenic diet, and the evaluation results are obtained. Through multi-dimensional data collection and analysis, the evaluation quality and accuracy of the ketogenic diet are improved, providing users with personalized healthy diet recommendations, thereby providing strong support for the rational application and effective management of the ketogenic diet.

[0021] Optionally, a heterogeneous graph can be constructed to analyze the complex relationships between users, recipes, ingredients, and calories, including: If recipe i contains ingredients If its calorie attribute is c, then connect an ingredient node and a calorie node to the recipe node; If user u selects recipe i, then there is an edge between the recipe node and the user node; Use a single embedding To represent each node, where Let d represent the node, d represent the embedding size, and let d represent the one-hot encoding of user u. Projecting it into the following embedded expression is: (1) in, It is the user's embedding, If the learnable embedding matrix is ​​the matrix of all users, then all nodes maintain an embedding matrix represented as follows: ,in, , , and These represent the number of users, recipes, ingredients, and calorie content, respectively.

[0022] In this embodiment, a graph neural network is used to perform characterization analysis on the heterogeneous graph to obtain the knowledge graph corresponding to the recipe information, including: Preset calorie perception user representation Recipe image features extracted by pre-trained network and embedding of target components The image features are converted into d-dimensional embeddings using a mapping layer, and the corresponding expression is: (2) Wherein, image embedding is denoted as , , For the parameters of the mapping layer, a feedforward network is used to calculate the attention weights of each component: (3) in, , , These are the parameters to be learned. Tanh is used as the non-linear activation function, and the embeddings of the components are fused through attention weights to obtain personalized component representations. (4) Based on the user's preference for different recipe information, the characteristics of the ingredients are integrated. Then, the characterization of the different ingredients in the recipe. , , , A feedforward network is used to calculate the attention weights for different components: (5) Where q represents a component feature, and component-level attention is used to aggregate the representations of different components into a comprehensive recipe feature; , , These are the parameters to be learned; By integrating different component representations through attention weights, we obtain the user-oriented summative recipe features: (6) in, This indicates that the personalized elements focus on the user.

[0023] It should be noted that the user health model is obtained by optimizing and analyzing the knowledge graph using a message propagation network based on the recipe information, including: We employ high-order connectivity of knowledge graphs for high-order feature propagation, and utilize the local connectivity of the graph structure and the neighbor information of nodes to capture the relationships and features between nodes. We aggregate neighbor information through multi-layer GCN stacking to obtain a global graph structure representation. The corresponding computational expression is: (7) in, Represents a non-linear activation function. This represents the input features of the k-th layer. This represents the output feature of the k-th layer. The degree matrix represents the degree of the self-linked matrix. Represents a linear transformation matrix. Where A represents the adjacency matrix of the target entity. Represents the identity matrix; Set all sampling neighbor sizes to Where n represents the number of samples, the expression for the attention weight between the head entity h and the tail entity t connected by the relation r is: (8) Normalize formula (3) using the softmax function: (9) in, Indicates attention weights. This represents the activation function. It is a weight vector. It is a weight matrix. Indicates a connection operation. Represents a user entity vector. Represents a relation vector; target node The representation is based on the aggregation of the user and its neighbor nodes, and the corresponding calculation expression is: (10) in, express The neighborhood aggregation is represented as, Represents the entity vector of neighboring nodes. Represents the vector of the user entity connected to it; Based on the user's self-network To represent the embedded representation of the user's domain, yes Standardized weights: (11).

[0024] Among them, calorie-aware user representation learning learns calorie-aware user representations through the message propagation mechanism of graph neural networks, allowing the model to capture users' preferences for different calorie levels. The knowledge graph message propagation algorithm mainly consists of three parts: embedding, a higher-order information propagation layer, and a prediction layer. The main purpose of the embedding layer is to convert high-dimensional features, such as user, recipe, and attribute information, into low-dimensional continuous vector representations. This conversion helps the model learn these features more effectively, capturing the potential relationships between users and items more effectively, and improving the accuracy and efficiency of recommendations. The higher-order propagation layer mainly includes information propagation and information aggregation. Information propagation obtains the higher-order structural features and semantic information of entities by analyzing their higher-order propagation attributes; information aggregation aggregates the importance of information from different entities to the user for recommendation purposes. Knowledge graphs have rich entity and semantic relationships; simply acquiring directly connected entities is insufficient to improve recommendation performance. Therefore, it is necessary to utilize the higher-order connectivity of knowledge graphs for higher-order feature propagation. Graph convolutional neural networks utilize the local connectivity of the graph structure and the neighbor information of nodes to capture the relationships and features between nodes. By stacking multiple layers of GCNs, a wider range of neighbor information can be aggregated step by step, thereby obtaining a more global graph structure representation.

[0025] Optionally, the target entity o and its neighboring entities can be... Combining them to perform a nonlinear transformation, the corresponding expression is: (12) in, Representation of the neighborhood of entities and users, concat means join, concat means connection. Represents a nonlinear transformation. Indicate target The final representation of the domain; After the recipe information is propagated through h layers and messages are aggregated, the new entity representations of the higher-order neighbors can be obtained. The expression for the interaction probability is derived using the inner product form: (13) in, This represents the final user u obtained after aggregating the neighborhood of user interactions. This represents the final entity obtained through h-order propagation. The representation, PRE, is derived from the user representation and the final prediction function. for The preference probability, and the corresponding loss function is: (14) in, Represents cross-entropy loss, This represents the positive sample set. Indicates a negative sample. This represents the L2 regularizer. This represents the Lagrange coefficient.

[0026] In this embodiment, an evaluation result is obtained by assessing each user's eligibility for a qualified ketogenic diet based on the user health model and preset evaluation indicators, including: Hit Ratio@K is used to evaluate the recommendation hit rate of the user health model, where K represents the top K items of the ketogenic diet table. The hit rate of each user is accumulated, and the accumulated result is divided by the total number of users to obtain the value of Hit Ratio@K. The more items on the ketogenic diet chart that match the user's preferences, the higher the Hit Ratio@K value and the higher the hit rate. Conversely, a lower hit rate indicates fewer items on the chart that match the criteria for a qualified ketogenic diet. The corresponding expression is: (15) in, This represents a list of eligible ketogenic diet items of length K returned by the user's health model. This represents the list of items that users actually interacted with in the test set, where N is the total number of users. This improves the accuracy of dietary health recommendations for users and enhances the accuracy and effectiveness of evaluating a qualified ketogenic diet.

[0027] In one feasible embodiment, the present invention also provides the application of the evaluation method described above in the evaluation and analysis of a qualified ketogenic diet. By establishing a comprehensive multi-dimensional evaluation index system and corresponding scoring standards, the implementation of the ketogenic diet is quantitatively evaluated, thereby objectively determining whether it is "qualified," and guiding practical applications based on the evaluation results. The evaluation index system includes the following five key dimensions: Growth and development, used to assess the effects of the ketogenic diet on normal growth and development in children / adolescents or on maintaining healthy body composition in adults; Psychological state is used to assess mood, appetite control, and satisfaction with food during the KDT (Knowledge, Technology, and Diet) program. Food diversity is used to assess the richness of the variety of foods consumed daily, whether bowel movements are regular and normal, and the overall nutritional balance of the diet. Metabolic indicators are biochemical markers used to assess ketogenic status, including blood ketone levels (target range is usually 1.2-4.9 mmol / L), urine ketone levels (qualitative detection target is usually ≥+++ (3+ above)) and glucose / ketone ratio (the molar concentration ratio of blood glucose to blood ketone, target range is usually 1:1-2:1). Adverse reactions are used to assess whether significant adverse symptoms or health problems occur during KDT, with a focus on constipation, indigestion, risk or occurrence of gallstones, and negative nitrogen balance.

[0028] The evaluation process includes the following steps: First, data collection is conducted by the participants, parents, doctors, nurses, and nutritionists. The system collects relevant information from the participants, including but not limited to detailed dietary records (for at least 3-7 consecutive days), body measurement data (such as weight, height / length, waist circumference, body fat percentage, etc.), self-assessment questionnaires or interview records of psychological state, food type records, bowel movement records, relevant laboratory test reports, adverse reaction symptom records, and family history information. Next, indicator evaluation is performed. Based on the collected data, each of the five defined dimensions is evaluated and analyzed. Finally, a comprehensive score is calculated. A preset scoring standard is used to quantify the performance of each dimension. The scoring standard is a total of 100 points, with each of the five dimensions accounting for 20 points. Each dimension includes several specific scoring points. The scores of each dimension are summarized to calculate the overall score, and a passing score is set (e.g., an overall score ≥ 80 points, or all dimensions meeting the minimum requirement and the overall score meeting the standard). Reaching this passing score indicates that the KDT implementation at this stage is considered "qualified."

[0029] The detailed scoring criteria are as follows: (1) Growth and development dimension (20 points): assess the trend of weight / BMI change and the growth rate of height / body length according to age and gender. If the deviation from the normal range is within the expected range, points will be deducted according to the degree. (2) Psychological state dimension (20 points): assess mood, appetite, and satisfaction (whether one is willing to share food with others). Each item has a set level corresponding to a different score. (3) Food diversity dimension (20 points): including food types (calculate the average number of food types per meal. If the target value is ≥8 types, full marks will be given. If it is insufficient, points will be deducted proportionally), bowel regularity (maintaining 1-2 natural bowel movements per day will not result in a deduction of points. If constipation occurs, points will be deducted according to the degree) and nutrition. Balance (assess whether the proportion of macronutrients meets the requirements of ketosis and whether the intake of micronutrients is sufficient; if not, points will be deducted for each item); (4) Metabolic indicators (20 points): blood ketones within the target range (1.2-4.9 mmol / L) are full marks; if lower or higher, points will be deducted according to the degree; urine ketones ≥+++ are full marks; if lower, points will be deducted according to the degree; sugar / ketone ratio within the target range is full marks; if deviated, points will be deducted according to the degree; (5) Adverse reactions (20 points): stone-related situations are deducted according to whether stones have occurred and whether there are high-risk factors; negative nitrogen balance is deducted according to blood test results and dietary protein assessment; hyperlipidemia and bone metabolism are deducted according to the occurrence and severity.

[0030] See Figure 2 and Figure 3 The specific implementation methods include professional execution or partial self-monitoring by users under the guidance of professionals. It is recommended to conduct the initial basic assessment at the beginning of KDT (e.g., week 1-2), followed by a formal evaluation after the adaptation period (e.g., week 3-4), and then evaluations periodically (e.g., monthly) or as needed (e.g., if discomfort occurs) depending on individual circumstances. Data collection tools may include standardized diet record apps, body composition analyzers, psychological state scales, laboratory test reports, etc. Scoring can be performed by professionals (doctors, nutritionists, or nurses) based on the collected data and pre-set detailed scoring rules, or an electronic scoring system can be designed to assist in the calculation. A comprehensive multi-dimensional evaluation index system and corresponding scoring standards are established. By collecting and analyzing relevant data from subjects during KDT implementation and scoring them, a comprehensive score is obtained to objectively determine whether their KDT performance is "qualified." Furthermore, this invention applies this evaluation method to the implementation process of KDT, enabling it to serve multiple scenarios such as KDT management, health monitoring, personalized diet plan development, and dietary education and guidance, thereby providing strong support for the scientific application and effective management of KDT.

[0031] In one feasible embodiment, the qualified KDT assessment method has a wide range of applications and can be used for dietary management during the implementation of KDT, such as during the initial adaptation period and maintenance period. By applying this assessment method regularly (e.g., every 2-4 weeks or as needed), and adjusting the diet plan in a timely and scientific manner based on the assessment results (including total score, scores of each dimension and specific issues), such as adjusting the proportion of macronutrients, increasing the types of specific foods, supplementing electrolytes / vitamins, and strategies to deal with adverse reactions, the diet plan can be kept in a "qualified" state and its safety and effectiveness can be optimized.

[0032] This evaluation method can also serve as a routine health monitoring tool for KDT populations. Through systematic assessment, it can identify potential growth and development problems, psychological distress, nutritional deficiencies, metabolic abnormalities, or adverse reaction risks early on, providing a basis for timely interventions (such as medical consultation, nutritional supplementation, program adjustments, or even dietary suspension) to prevent health risks. When designing KDT programs for individuals, or adjusting programs based on evaluation results, the evaluation results (especially weaknesses and individual response characteristics) can serve as input, helping to tailor personalized KDT programs that better suit their physiological characteristics, lifestyle habits, health goals, and tolerance, thereby improving the feasibility and success rate of the program.

[0033] This evaluation method, along with its specific evaluation dimensions and standards, can also serve as the core content for scientific nutrition education for KDT practitioners (and their families / caregivers). By explaining the meaning and achievement requirements of each indicator, it guides them to correctly understand KDT principles, master food selection skills, identify adverse reaction signals, and conduct self-monitoring (such as recording diet, bowel movements, and mood), thereby improving their compliance and self-management abilities. The evaluation results themselves are also an important feedback and guidance tool. The following are the rating results for different users: After four weeks of KDT (Knowledge, Nutrition, and Diet) training, Dong scored 85 points (passing), but only 15 points in the "Food Diversity" dimension (main deduction point: an average of only 3 types of food per meal). Based on this, the nutritionist guided Dong to increase the variety of vegetables, try different fat sources (such as avocados, nuts, and different types of oils), and enrich her diet. Mr. Li scored 12 points in the "Adverse Reactions" dimension, with deductions for frequent bloating (indigestion) and a family history of kidney stones (a risk factor for kidney stones). The doctor recommended adjusting his fat intake (separate meals, choosing easily digestible fats), ensuring adequate water intake, and regularly monitoring relevant indicators. Zhu's initial assessment showed low psychological satisfaction with the extremely high-fat diet, accompanied by mild nausea. The nutritionist developed a "mild ketogenic diet" plan for him (slightly higher carbohydrates, slightly lower fat content), and increased protein intake to meet his satiety needs; Wang's evaluation results showed a low score in "metabolic indicators" (blood ketones 0.8 mmol / L). The analysis suggests that the reason may be excessive intake of hidden carbohydrates. He was educated on how to carefully read food labels, avoid hidden carbohydrates, and was shown the correct way to use the blood ketone meter.

[0034] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limitations; therefore, other examples of exemplary embodiments may have different values.

[0035] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0036] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for evaluating a qualified ketogenic diet, characterized in that, Includes the following steps: The diet data of multiple users is obtained and preprocessed to obtain a dataset, wherein the diet data includes user information and recipe information; A heterogeneous graph of the dataset is constructed, and a graph neural network is used to perform representation analysis on the heterogeneous graph to obtain a knowledge graph corresponding to the recipe information. The heterogeneous graph includes users, recipes, ingredients, and calories. Based on the recipe information, a message propagation network is used to optimize and analyze the knowledge graph to obtain a user health model. The message propagation network includes an embedding layer, a higher-order propagation layer, and a prediction layer. The embedding layer is used to provide embedded representations for all recipes, users, and relationships. The higher-order information propagation layer includes information propagation and information aggregation, used to propagate and aggregate information between different entities. The prediction layer is used to rank the importance between entities and users and predict the matching score between recipes and users. The evaluation results are obtained by evaluating each user's ketogenic diet based on the user health model and preset evaluation indicators. The preset evaluation indicators include growth and development information, psychological state information, food diversity information, metabolic index information, and adverse reaction information.

2. The method for evaluating a qualified ketogenic diet according to claim 1, characterized in that, Constructing a heterogeneous graph of the dataset includes: Preset a user-recipe pair Recipe i has a set of ingredients. Image features and calorie factor A heterogeneous graph is constructed to analyze the complex relationships between users, recipes, ingredients, and calories. The heterogeneous graph is used to predict the probability of user u interacting with recipe i. The graph structure allows the model to capture different types of nodes and various types of edges between nodes. The self-supervised ingredient prediction module is used to analyze the relationships between ingredients and predict the recipe set for recipe i. The knowledge graph is obtained by learning representations of the masked components, wherein a multi-head attention model is used to recover components given the surrounding context. The masked components.

3. The method for evaluating a qualified ketogenic diet according to claim 2, characterized in that, Constructing a heterogeneous graph of the dataset includes: Preset Represents a set of users The set of recipes is represented by the user-recipe interaction matrix. ,in, This indicates whether user u interacts with recipe i, and each recipe i has associated image features. Ingredient Set and heat factor Ingredient Set It is a multi-hot encoded vector. If recipe i contains ingredient k, then the corresponding... Otherwise, it is 0; heat factor It is a single-heat encoded vector used to represent the calorie level of recipe i, where These represent the quantities of different calorie levels, and the corresponding graph neural network model is: Input: User-recipe interaction matrix Y, recipe ingredient matrix Each row corresponds to a recipe, and the number of columns represents the total number of possible ingredients. (Recipe image feature matrix) Recipe calorie information matrix ; Output: Interactive function , used to predict the probability of user u interacting with recipe i.

4. The method for evaluating a qualified ketogenic diet according to claim 2, characterized in that, Construct a heterogeneous graph to analyze the complex relationships between users, recipes, ingredients, and calories, including: If recipe i contains ingredients If its calorie attribute is c, then connect an ingredient node and a calorie node to the recipe node; If user u selects recipe i, then there is an edge between the recipe node and the user node; Use a single embedding To represent each node, where Let d represent the node, d represent the embedding size, and let d represent the one-hot encoding of user u. Projecting it into the following embedded expression is: （1） in, It is the user's embedding, If the learnable embedding matrix is ​​the matrix of all users, then all nodes maintain an embedding matrix represented as follows: ,in, , , and These represent the number of users, recipes, ingredients, and calorie content, respectively.

5. The method for evaluating a qualified ketogenic diet according to claim 2, characterized in that, A graph neural network is used to represent and analyze the heterogeneous graph to obtain a knowledge graph corresponding to the recipe information, including: Preset calorie perception user representation Recipe image features extracted by pre-trained network and embedding of target components The image features are converted into d-dimensional embeddings using a mapping layer, and the corresponding expression is: （2） Wherein, image embedding is denoted as , , For the parameters of the mapping layer, a feedforward network is used to calculate the attention weights of each component: （3） in, , , These are the parameters to be learned. Tanh is used as the non-linear activation function, and the embeddings of the components are fused through attention weights to obtain personalized component representations. （4） Based on the user's preference for different recipe information, the characteristics of the ingredients are integrated. Then, the characterization of the different ingredients in the recipe. , , , A feedforward network is used to calculate the attention weights for different components: （5） Where q represents a component feature, and component-level attention is used to aggregate the representations of different components into a comprehensive recipe feature; , , These are the parameters to be learned; By integrating different component representations through attention weights, we obtain the user-oriented summative recipe features: （6） in, This indicates that the personalized elements focus on the user.

6. The method for evaluating a qualified ketogenic diet according to claim 1, characterized in that, Based on the recipe information, a message propagation network is used to optimize and analyze the knowledge graph to obtain a user health model, including: We employ high-order connectivity of knowledge graphs for high-order feature propagation, and utilize the local connectivity of the graph structure and the neighbor information of nodes to capture the relationships and features between nodes. We aggregate neighbor information through multi-layer GCN stacking to obtain a global graph structure representation. The corresponding computational expression is: （7） in, Represents a non-linear activation function. This represents the input features of the k-th layer. This represents the output feature of the k-th layer. The degree matrix represents the degree of the self-linked matrix. Represents a linear transformation matrix. Where A represents the adjacency matrix of the target entity. Represents the identity matrix; Set all sampling neighbor sizes to Where n represents the number of samples, the expression for the attention weight between the head entity h and the tail entity t connected by the relation r is: （8） Normalize formula (3) using the softmax function: （9） in, Indicates attention weights. This represents the activation function. It is a weight vector. It is a weight matrix. Indicates a connection operation. Represents a user entity vector. Represents a relation vector; target node The representation is based on the aggregation of the user and its neighbor nodes, and the corresponding calculation expression is: （10） in, express The neighborhood aggregation is represented as, Represents the entity vector of neighboring nodes. Represents the vector of the user entity connected to it; Based on the user's self-network To represent the embedded representation of the user's domain, yes Standardized weights: （11）。 7. The method for evaluating a qualified ketogenic diet according to claim 6, characterized in that, Also includes: Connect the target entity o and its neighboring entities Combining them to perform a nonlinear transformation, the corresponding expression is: （12） in, Representation of the neighborhood of entities and users, concat means join, concat means connection. Represents a nonlinear transformation. Indicate target The final representation of the domain; After the recipe information is propagated through h layers and messages are aggregated, the new entity representations of the higher-order neighbors can be obtained. The expression for the interaction probability is derived using the inner product form: （13） in, This represents the final user u obtained after aggregating the neighborhood of user interactions. This represents the final entity obtained through h-order propagation. The representation, PRE, is derived from the user representation and the final prediction function. for The preference probability, and the corresponding loss function is: （14） in, Represents cross-entropy loss, This represents the positive sample set. Indicates a negative sample. This represents the L2 regularizer. This represents the Lagrange coefficient.

8. The method for evaluating a qualified ketogenic diet according to claim 1, characterized in that, Based on the user health model and preset evaluation indicators, an evaluation of each user's eligibility for a qualified ketogenic diet is conducted to obtain evaluation results, including: Hit Ratio@K is used to evaluate the recommendation hit rate of the user health model, where K represents the top K items of the ketogenic diet table. The hit rate of each user is accumulated, and the accumulated result is divided by the total number of users to obtain the value of Hit Ratio@K. The more items on the ketogenic diet chart that match the user's preferences, the higher the Hit Ratio@K value and the higher the hit rate. Conversely, a lower hit rate indicates fewer items on the chart that match the criteria for a qualified ketogenic diet. The corresponding expression is: （15） in, This represents a list of eligible ketogenic diet items of length K returned by the user's health model. This represents the list of items that users actually interacted with in the test set, where N is the total number of users.

9. The method for evaluating a qualified ketogenic diet according to claim 1, characterized in that, The growth and development information is used to assess the impact of the ketogenic diet on the user's normal growth and development or to maintain a healthy body composition. The psychological state information is used to assess mood, appetite control, and satisfaction with the diet during the implementation of the ketogenic diet. The food diversity information is used to assess the richness of the variety of foods consumed daily, whether bowel movements are regular and normal, and the overall nutritional balance of the diet. The metabolic indicators are used to assess biochemical markers of ketosis, including blood ketone levels, urine ketone levels, and glucose / ketone ratio. The adverse reaction information is used to assess whether significant adverse symptoms or health problems, constipation, indigestion, risk or occurrence of stones, and negative nitrogen balance occur during the implementation of the ketogenic diet.

10. The application of a qualified ketogenic diet evaluation method according to any one of claims 1-9 in the evaluation and analysis of qualified ketogenic diets.

Images