System and application method of nutrition recipe recommendation system based on artificial intelligence in chronic disease intervention management

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a system of a nutrition recipe recommendation system based on artificial intelligence in chronic disease intervention management and an application method. Comprising a data acquisition module, a data preprocessing and feature engineering module, a knowledge base management module, a personalized nutritional requirement modeling module, a recipe generation and optimization module, a user interface module and a model training and updating module which are in communication connection through a network. According to the method, personalized health data such as physiological indexes, chronic disease characteristics, diet preference, allergy and intolerance information of a user are collected in multiple dimensions, a nonlinear relation between the characteristics of the user and nutritional requirements is deeply mined in combination with a deep learning model, and an accurate personalized nutritional requirement model is generated according to specific illness conditions and body indexes of different chronic disease patients; therefore, recipe recommendation conforming to individual differences is provided, and the risk of aggravating the illness state due to improper diet is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular to a system and application method of a nutrition recipe recommendation system based on artificial intelligence in chronic disease intervention management. BACKGROUND

[0002] With the intensification of global population aging and the changes in lifestyle, the prevalence of chronic non-communicable diseases continues to rise, which has become a major public health problem that seriously threatens human health. In the intervention and management of chronic diseases, nutrition and diet as a basic and key link, its scientificity and individualization are directly related to the disease control, prognosis improvement and life quality improvement of patients. In recent years, the rapid development of artificial intelligence technology has provided new possibilities for personalized nutrition management of patients with chronic diseases, and intelligent nutrition recipe recommendation systems have emerged as the times require, and the demand for intelligence has become increasingly prominent.

[0003] Among them, the existing intelligent nutrition recipe recommendation system aims to provide meal suggestions for users in an automated way to assist in chronic disease intervention. These systems are usually based on pre-set nutrition databases and general dietary guidelines, trying to provide certain dietary references for users. However, in actual application, these methods often expose their inherent limitations, especially in dealing with individual needs and responding to rapidly changing knowledge systems.

[0004] The prior art shows deficiencies in individualization, failing to fully consider individual differences of users, including physical condition, disease characteristics, eating habits, food allergies or intolerance, etc.; for example, the nutritional needs of different chronic disease patients differ greatly, but existing systems often fail to provide precise and personalized nutrition recipe recommendations for each patient's specific condition and physical indicators, resulting in poor patient compliance and possible failure to achieve the expected effect of chronic disease intervention, or even exacerbation of the disease due to improper diet. In addition, nutrition knowledge and medical research results continue to develop rapidly, with new food nutrients, disease and nutrition associations, and more effective dietary intervention methods emerging, but the current recommendation method based on fixed knowledge base updates slowly and is difficult to keep up with these changes; at the same time, the existing technology lacks sufficient consideration of users' eating habits and food preferences in different regions and cultural backgrounds, such as unique dietary taboos and preferences in minority areas, making it difficult for recommended recipes to be accepted and followed by users. SUMMARY

[0005] (I) Technical problems solved

[0006] In view of the deficiencies of the prior art, the present application provides a system and application method of a nutrition recipe recommendation system based on artificial intelligence in chronic disease intervention management, which solves the problem of "poor use effect" in the above background technology.

[0007] (II) Technical solutions

[0008] To achieve the above objectives, the present invention is implemented through the following technical solution: a nutrition recipe recommendation system based on artificial intelligence in chronic disease intervention and management, comprising a data acquisition module, a data preprocessing and feature engineering module, a knowledge base management module, a personalized nutrition needs modeling module, a recipe generation and optimization module, a user interface module, and a model training and update module, wherein the modules are connected through network communication.

[0009] The data acquisition module is used to collect users' health-related data, including a biosensor interface unit, a medical record integration unit, and a user input unit;

[0010] The data preprocessing and feature engineering module is used to perform deep cleaning, normalization, feature extraction and feature transformation on the raw data to generate standardized feature vectors;

[0011] The knowledge base management module is used to collect, update, and maintain professional knowledge related to nutrition and chronic disease intervention.

[0012] The personalized nutrition needs modeling module is used to combine user-specific characteristics and professional knowledge base to construct a user-specific nutrition needs model.

[0013] The recipe generation and optimization module is used to generate candidate recipes based on the nutritional requirements model and perform multi-dimensional optimization.

[0014] The user interface module is used for interaction between the system and the user;

[0015] The model training and update module is used to continuously train and iteratively update the artificial intelligence model in the personalized nutrition needs modeling module.

[0016] Preferably, the biosensor interface unit is used to connect to wearable devices or medical-grade sensors to acquire the user's physiological indicators in real time, including blood glucose levels, blood pressure data, weight, body fat percentage, and heart rate variability; the medical record integration unit is used to import the user's chronic disease diagnosis information, past medical history, medication status, and medical examination reports from the hospital information system or personal health records; the user input unit is used to acquire the user's dietary habits and preferences, food allergy information, dietary restrictions, lifestyle, exercise intensity, personalized diet requirements, and feedback data on recommended diets.

[0017] Preferably, the data preprocessing and feature engineering module includes a data cleaning unit, a normalization unit, a feature extraction unit, and a feature transformation unit. The data cleaning unit is used to process missing values, outliers, and noise in the data. The normalization unit is used to unify features with different dimensions and numerical ranges into a specific interval. The feature extraction unit is used to mine high-level, meaningful features from the original data. The feature transformation unit is used to encode features of different modalities.

[0018] Preferably, the knowledge base management module includes a knowledge acquisition unit, a knowledge storage unit, and a knowledge update unit. The knowledge acquisition unit is used to acquire nutritional knowledge, food nutrient data, disease-nutrition association rules, and dietary intervention suggestions from authoritative nutrition journals, medical research reports, national dietary guidelines, food composition databases, and disease clinical pathways. The knowledge storage unit uses a graph database or relational database to store knowledge in a structured manner, including constructing a knowledge graph. In the knowledge graph, nodes represent food, nutrients, diseases, symptoms, and user attribute entities, and edges represent the relationships between entities. The knowledge update unit is used to update the knowledge base periodically or triggered by events.

[0019] Preferably, the personalized nutrition requirement modeling module includes a multi-layer neural network structure, which includes a deep neural network, a long short-term memory network, or a Transformer network. Its input consists of the user-standardized feature vector output by the data preprocessing and feature engineering module and the relevant knowledge entity embedding vector provided by the knowledge base management module. This is used to learn the complex nonlinear relationship between user characteristics and nutrition requirements, and to dynamically infer the specific requirements and limitations for macronutrients and micronutrients.

[0020] Preferably, the recipe generation and optimization module includes a recipe generation unit and a recipe optimization unit; the recipe generation unit is used to screen ingredient combinations from a food database based on nutritional constraints and food allergen avoidance, and combine them with a cooking knowledge base to form specific dishes and recipes; the recipe optimization unit is used to perform multi-objective optimization on candidate recipes, and the optimization objectives include nutritional balance, compliance, ingredient availability, meal preparation convenience, and cost-effectiveness, and the optimization algorithm adopts a multi-objective genetic algorithm or reinforcement learning method.

[0021] Preferably, the user interface module includes a recipe display unit, a feedback collection unit, and a health report unit; the recipe display unit is used to present recommended nutritious recipes in an interface, including the dish name, picture, cooking method, ingredients, nutritional analysis, and a description of its relevance to chronic disease management goals; the feedback collection unit is used to obtain user ratings, comments, preference tags, and actual changes in diet and health indicators for the recommended recipes; the health report unit is used to periodically generate personalized health reports, summarizing user dietary performance, changes in chronic disease indicators, nutritional intake analysis, and health recommendations.

[0022] Preferably, an application method of an artificial intelligence-based nutritional recipe recommendation system in chronic disease intervention and management includes the following steps:

[0023] S1, to obtain personalized health data of users, mainly to comprehensively and accurately collect various health-related information that reflects individual differences of users;

[0024] S2 performs multimodal feature encoding on user health data, converting the raw, heterogeneous user health data obtained in S1 into feature vector representations using artificial intelligence technology;

[0025] S3 provides access to the latest nutritional and chronic disease intervention knowledge, primarily by ensuring the scientific validity and timeliness of recommended diets and addressing the issue of slow knowledge updates in traditional systems.

[0026] S4, build a personalized nutrition needs model, which mainly focuses on the uniqueness of each user and tailors their precise nutrition needs.

[0027] S5 generates candidate recipes that meet user needs. It mainly uses the personalized nutritional needs model built by S1, as well as a huge food database and cooking knowledge base, to intelligently filter, combine and generate a series of candidate recipes that meet the user's nutritional, health and preference constraints.

[0028] S6 assesses the overall suitability of the candidate recipes, which mainly involves a multi-dimensional and detailed evaluation of the candidate recipes generated in S5 to ensure that the recommended recipes are not only nutritionally sound.

[0029] S7 recommends optimized nutritional recipes to users, mainly presenting the nutritional recipes evaluated and selected by S6 to users in a clear, intuitive and easy-to-understand way;

[0030] S8 collects user feedback and updates the model, mainly by proactively collecting users' actual experiences and health improvement effects of recommended recipes, and incorporating this feedback data into model training.

[0031] (III) Beneficial Effects

[0032] This invention provides a system and application method for an artificial intelligence-based nutritional recipe recommendation system in chronic disease intervention and management. It has the following beneficial effects:

[0033] (1) This invention collects personalized health data such as users’ physiological indicators, chronic disease characteristics, dietary preferences, allergies and intolerances from multiple dimensions, and combines deep learning models to deeply mine the nonlinear relationship between user characteristics and nutritional needs. It can generate accurate personalized nutritional needs models for the specific conditions and physical indicators of different chronic disease patients, and then provide recipe recommendations that fit individual differences, effectively improve patient compliance, ensure the effectiveness of chronic disease intervention, and reduce the risk of aggravating the condition due to improper diet.

[0034] (2) This invention uses a knowledge acquisition unit to regularly crawl the latest research results, dietary guidelines and other information from authoritative sources. Combined with the expert review and incremental learning mechanism of the knowledge update unit, it ensures that the knowledge base can integrate new food nutritional components, disease and nutrition-related knowledge in a timely manner. This solves the problem of slow updates of fixed knowledge bases in the prior art, and ensures that the recommended recipes are always based on the latest and most accurate professional knowledge, thereby improving the scientific nature and timeliness of the recommendations.

[0035] (3) This invention constructs a multilingual and multicultural food database and combines it with a cultural and regional adaptability assessment module. During the recipe generation and optimization process, it considers factors such as the food culture, customs and dietary taboos of the user's region, so that the recommended recipes are more in line with the user's cultural background and dietary traditions, thereby improving the user's acceptance and compliance with the recipes. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the system framework of the present invention;

[0037] Figure 2 This is a schematic diagram of the application method of the present invention. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Please see Figure 1 - Figure 2 This invention provides an artificial intelligence-based nutritional recipe recommendation system for chronic disease intervention and management, comprising a data acquisition module, a data preprocessing and feature engineering module, a knowledge base management module, a personalized nutritional needs modeling module, a recipe generation and optimization module, a user interface module, and a model training and updating module. These modules are connected via network communication. Specifically:

[0040] The data acquisition module is responsible for collecting various health-related data from users, including a biosensor interface unit, a medical record integration unit, and a user input unit. The biosensor interface unit connects to various wearable devices or medical-grade sensors to acquire users' physiological indicators in real time, such as blood glucose levels from a blood glucose meter, blood pressure data from a blood pressure monitor, weight and body fat percentage from a body fat scale, and heart rate variability from a heart rate wristband. This sensor data is transmitted to the system via wireless communication protocols such as Bluetooth or Wi-Fi. The medical record integration unit imports users' chronic disease diagnosis information, past medical history, medication information, and other relevant medical examination reports from hospital information systems or personal health records. This unit must strictly adhere to data privacy and security standards, ensuring encrypted data transmission and access control. The user input unit provides an interactive interface, allowing users to actively input their dietary preferences, food allergy information, lists of intolerant foods, special dietary restrictions, lifestyle habits, exercise intensity, and personalized recipe needs, such as preferred cuisine, cooking methods, and meal preparation time budget. This unit also collects user feedback data on recommended recipes, including satisfaction ratings, taste evaluations, and actual health improvement effects. All collected data will be timestamped and undergo preliminary data cleaning, such as removing duplicate data and standardizing the format, before being stored in the data storage module.

[0041] The data preprocessing and feature engineering module receives raw data from the data acquisition module and performs deep cleaning, normalization, feature extraction, and feature transformation to generate standardized feature vectors suitable for artificial intelligence model processing. This module includes a data cleaning unit, a normalization unit, a feature extraction unit, and a feature transformation unit. The data cleaning unit handles missing values, outliers, and noise. Missing value handling can employ mean imputation, median imputation, or machine learning algorithms such as K-nearest neighbor imputation. Outlier identification uses statistical methods such as Z-score detection or outlier detection based on isolated forests. Noise is eliminated through smoothing filtering or wavelet denoising techniques. The normalization unit unifies features with different dimensions and numerical ranges to a specific interval, such as zero to one, to eliminate the impact of dimensional differences on model training. Common methods include min-max normalization or Z-score normalization. The feature extraction unit mines high-level, meaningful features from the raw data. For example, for time-series blood glucose data, it can extract fluctuation range, average value, peak value, trough value, and intraday trend; for text describing dietary habits, it can perform natural language processing to extract key food preference terms and nutritional keywords. The feature transformation unit is responsible for encoding features of different modalities. For example, it converts categorical features such as chronic disease types and food allergens into unique codes or embedding codes, and converts user preference text into word vectors or sentence vectors. The standardized feature vectors output by the data preprocessing and feature engineering module will be used as input to the personalized nutrition needs modeling module.

[0042] The knowledge base management module focuses on collecting, updating, and maintaining professional knowledge related to nutrition and chronic disease intervention. It includes a knowledge acquisition unit, a knowledge storage unit, and a knowledge update unit. The knowledge acquisition unit continuously acquires the latest nutritional knowledge, food nutrient data, disease-nutrient association rules, and dietary intervention recommendations from authoritative nutrition journals, medical research reports, national dietary guidelines, food composition databases, and disease clinical pathways through web crawling technology and human expert review mechanisms. For example, when new research finds that a certain food is beneficial or harmful to a specific chronic disease, the knowledge acquisition unit will promptly capture this information. The knowledge storage unit uses a graph database or relational database to structurally store this knowledge. For example, a knowledge graph can be constructed where nodes represent entities such as food, nutrients, diseases, symptoms, and user attributes, and edges represent relationships such as "food contains nutrients," "nutrients affect diseases," and "foods prohibited by diseases." This graph can efficiently support complex queries and reasoning. The knowledge update unit is responsible for updating the knowledge base periodically or triggered by events. When new knowledge is acquired, the unit initiates a verification process where experts evaluate and confirm the new knowledge, then integrate it into the existing knowledge system and automatically update the structure and content of the knowledge graph. This mechanism ensures that the system always recommends recipes based on the latest and most accurate professional knowledge.

[0043] The personalized nutrition needs modeling module is responsible for constructing a user-specific nutrition needs model by combining user-specific characteristics and a knowledge base. This model incorporates multi-layered neural network structures, such as deep neural networks, long short-term memory networks, or multi-layered neural networks. The input consists of the user-standardized feature vector output from the data preprocessing and feature engineering module, and relevant knowledge entity embedding vectors provided by the knowledge base management module. The model learns the complex non-linear relationship between user characteristics and nutritional needs. Specifically, based on the user's chronic disease type (e.g., diabetes, hypertension), combined with physiological indicators such as blood glucose levels and blood pressure readings, as well as dietary habits and allergy information, the model dynamically infers the specific requirements and limitations for macronutrients such as carbohydrates, proteins, and fats, and micronutrients such as vitamins and minerals.

[0044] The recipe generation and optimization module, based on the user's nutritional needs model output by the personalized nutritional needs modeling module, and combined with a food database and culinary knowledge base, generates preliminary candidate recipes and performs multi-dimensional optimization, including a recipe generation unit and a recipe optimization unit. The recipe generation unit first filters out suitable ingredient combinations from a massive food database based on nutritional constraints determined by the user's nutritional needs model, such as protein intake and fat restrictions, as well as food allergen avoidance, such as avoiding seafood. Then, combining these ingredients with the culinary knowledge base, it creates specific dishes and recipes. The culinary knowledge base contains information on various dish preparation methods, seasoning usage, cooking times, etc.

[0045] The user interface module is responsible for all interactions between the system and the user. It includes a recipe display unit, a feedback collection unit, and a health report unit. The recipe display unit presents recommended nutritious recipes to users through a clear and intuitive interface, including the name of each dish, an image, detailed instructions, required ingredients and their nutritional analysis, as well as a description of its relevance to the user's chronic disease management goals. For example, it clearly indicates the carbohydrate content of a dish and its impact on blood sugar. The feedback collection unit provides convenient interaction methods, allowing users to rate recommended recipes, post comments, mark preferences or dislikes, and record the user's actual dietary habits and changes in health indicators. This feedback data is sent back to the data acquisition module and the model training and update module for continuous model improvement. The health report unit regularly generates personalized health reports, summarizing the user's dietary performance over a period of time, trends in chronic disease indicators, nutritional intake analysis, and health recommendations provided by the system.

[0046] The model training and update module is responsible for the continuous training and iterative updates of the AI ​​model in the personalized nutrition needs modeling module. It includes a model training unit and a model update unit. The model training unit receives user feedback data and new health data from the data acquisition module, as well as the latest knowledge from the knowledge base management module. This data is used to retrain the model to adapt to changes in users' health conditions, evolving dietary preferences, and the latest nutritional research findings. The training process may employ a federated learning mechanism, aggregating learning experiences from multiple users while protecting user data privacy. The model update unit is responsible for deploying the newly trained model to the system, replacing the old model. This process requires ensuring a smooth transition, avoiding service interruptions, and conducting rigorous performance testing and security verification of the new model. Through this continuous learning and update mechanism, the system can maintain the real-time, accurate, and adaptive nature of its recommendations.

[0047] Based on the above, this invention also provides a method for applying an artificial intelligence-based nutritional recipe recommendation system in chronic disease intervention and management, specifically including the following steps:

[0048] S1, acquiring personalized health data, primarily involves comprehensively and accurately collecting various health-related information reflecting individual user differences. This is achieved through multi-channel and multi-dimensional data collection, which compensates for this deficiency. Specifically, acquiring personalized health data includes the following steps:

[0049] S101 collects users' physiological and biochemical data, acquiring real-time physiological status information through intelligent biosensors or medical devices. For example, a blood glucose monitor periodically or continuously acquires the user's blood glucose levels, including fasting blood glucose, postprandial blood glucose, and random blood glucose, and records their timestamps; a smart blood pressure monitor acquires the user's systolic and diastolic blood pressure, as well as heart rate data; a smart body fat scale acquires the user's weight, body fat percentage, muscle mass, bone mass, and other body composition data; and a heart rate bracelet or smartwatch acquires the user's heart rate, heart rate variability, sleep duration, and sleep quality data. These devices typically use Bluetooth, Wi-Fi, or other short-range wireless communication technologies to automatically synchronize data to the user's application, and then transmit it to the backend data storage system through a secure encrypted channel. To ensure data accuracy, the acquisition devices need to be calibrated regularly, and users should perform measurements under standard operating procedures, such as measuring blood pressure at fixed times and in the same posture, to reduce errors.

[0050] S102: Enter the user's chronic disease diagnosis and medical history information. This is done through a structured form or medical record import interface to obtain the user's core disease information. This includes the diagnosed chronic disease type, such as type 2 diabetes, hypertension, hyperlipidemia, fatty liver, and kidney disease; the diagnosis date; current medication status, such as drug name, dosage, and usage; and related complications or accompanying symptoms. For example, for kidney disease patients, their renal function stage needs to be specifically recorded, as it directly affects protein and electrolyte intake restrictions. In addition, the user's allergy history needs to be recorded, such as peanut allergy or lactose intolerance, as well as a list of food intolerances, such as indigestion of certain grains. This information provides key contraindications and restrictions for subsequent diet recommendations. During data entry, the system performs preliminary semantic validation, such as checking if drug names exist in standard pharmacopoeias, to improve data quality.

[0051] S103 collects data on users' dietary habits and lifestyles, primarily to understand their daily dietary preferences, cooking habits, and daily routines. For example, questionnaires are used to collect user preferences for different cuisines (e.g., Sichuan, Cantonese, vegetarian), flavor preferences (e.g., mild, spicy), and food likes and dislikes. Meal frequency, meal times, and regularity are also recorded. Lifestyle data includes users' occupation, work intensity, exercise frequency, type, and intensity; this information is used to assess daily energy expenditure. For example, the energy needs of office workers and manual laborers differ significantly. Additionally, users' budget for meal preparation time is collected, such as how much time they are willing to spend preparing food each day, which affects the complexity of recommended recipes. This unstructured data undergoes preliminary analysis using natural language processing techniques to extract key semantic features.

[0052] S104 integrates and validates multi-source data, primarily responsible for consolidating data from different sources and formats, and performing rigorous data validation and cleaning. The integration process includes aligning data with different timestamps and standardizing units for data collected from different devices, such as unifying all energy units to kilojoules. Data validation includes checking the completeness, consistency, and logical rationality of the data. For example, it checks whether there are extreme deviations between a user's height and weight and their age and gender, and whether blood glucose medication and diagnostic information match. For data that is illogical or contains obvious errors, the system will mark it and prompt the user or administrator to correct it. This step ensures the accuracy and reliability of subsequent model input data, laying a solid foundation for the construction of personalized nutrition needs models.

[0053] S2 involves multimodal feature encoding of user health data. The raw, heterogeneous user health data obtained in S1 is transformed into a unified, high-dimensional, semantically meaningful feature vector representation using artificial intelligence technology. This specifically includes the following steps:

[0054] S201 performs time-series feature extraction and encoding on physiological and biochemical indicator data, employing time series analysis techniques for feature extraction. For example, for continuous blood glucose monitoring data, short-term fluctuation trends are extracted using methods such as sliding window averaging and exponential smoothing; periodic features, such as intraday blood glucose rhythm, are extracted using Fourier transform or wavelet analysis. Recurrent neural network structures, such as Long Short-Term Memory networks or gated recurrent units, are used to deeply encode time-series sequences of blood glucose, blood pressure, and heart rate, capturing their long-term dependencies and nonlinear dynamic patterns. The encoder maps the raw time-series data into a fixed-length time-series feature vector, which effectively represents the dynamic change pattern of the physiological indicator over a period of time.

[0055] S202 involves entity embedding encoding of chronic disease diagnosis and medical history information, using entity embedding techniques to encode it into continuous, low-dimensional vector representations. For example, for disease entities such as "type 2 diabetes" and "hypertension," word embedding models, such as word vectors or pre-trained models in the medical field, are used to convert them into vectors. Similar diseases or drugs will have similar vector representations in the embedding space. Furthermore, considering the relationships between these entities, such as the relationship between "insulin" and "diabetes," knowledge graph embedding techniques can be used to encode entities such as diseases, drugs, symptoms, and allergens, as well as their interrelationships, into vectors. This ensures that the vectors not only represent the entities themselves but also implicitly contain their semantic context within the medical knowledge graph.

[0056] S203 involves semantic feature encoding of dietary habits and lifestyle data using natural language processing and multi-label classification techniques. For free-text descriptions of dietary preferences, such as "prefers light Cantonese cuisine and dislikes fried food," text embedding models like Transformer Encoder convert them into semantic vectors. For multi-label data, such as "allergic to seafood" or "gluten intolerant," multi-hot encoding or multi-label embedding techniques can be used. Additionally, numerical lifestyle data such as exercise frequency and intensity, after normalization, can be directly used as dimensions of the feature vectors. This encoding process transforms vague user preference descriptions into precise numerical features that the model can understand.

[0057] S204 performs multimodal feature fusion, combining the feature vectors from different modalities generated in S201, S202, and S203 to form a unified and comprehensive user multimodal health feature vector. Fusion methods include, but are not limited to: feature concatenation (e.g., directly concatenating feature vectors from all modalities into a longer vector); attention mechanism fusion (learning the importance of features from different modalities through self-attention networks or cross-attention networks and assigning corresponding weights for weighted summation; for example, when a user has a severe history of allergies, the weight of the allergen feature vector is significantly increased); and multi-task learning framework (allowing encoders from different modalities to learn their own modal-specific representations while sharing some parameters, finally unifying them at the fusion layer). The fused feature vector comprehensively reflects the user's physiological state, disease background, dietary preferences, and lifestyle, providing rich and refined input for subsequent construction of personalized nutritional needs models.

[0058] S3, acquiring the latest nutritional and chronic disease intervention knowledge, mainly focuses on ensuring the scientific validity and timeliness of recommended diets, addressing the problem of slow knowledge updates in traditional systems. Specifically, it includes the following steps:

[0059] S301 regularly crawls and analyzes authoritative nutritional data. Using an automated web crawler, it periodically accesses and downloads the latest dietary guidelines, food nutrient databases, recommended nutrient intake standards, and the latest clinical nutrition research reports from authoritative global or national nutrition organizations such as the World Health Organization, the National Institute for Nutrition and Health of the Chinese Center for Disease Control and Prevention, the U.S. National Library of Medicine, and the European Food Safety Authority. The analysis program automatically identifies and extracts key information from the reports, such as newly discovered food bioactive components, the latest evidence linking specific nutrients to chronic diseases, and revised nutrient requirements for different population groups. For example, if the latest research indicates that a certain type of dietary fiber has new benefits for controlling type 2 diabetes, this information will be precisely extracted.

[0060] S302 integrates and updates the disease-nutrition knowledge graph, combining the nutritional data acquired in S301 with the existing disease knowledge base and dynamically updating the knowledge graph. The knowledge graph stores knowledge in triplets, such as entity-relationship-entity, for example, "food - rich in - vitamin C", "vitamin C - helps - immune function", and "hypertension - restrict - sodium intake". When new research reveals a link between disease and nutrition, such as discovering that a certain trace element has a protective effect on cardiovascular health, the knowledge graph adds corresponding entities and relationships. The update process uses incremental learning to avoid rebuilding the entire graph each time. Simultaneously, conflict detection and resolution are performed. For example, if new guidelines conflict with old guidelines, the system will mark the conflict and prompt human experts for review and decision-making to ensure the consistency and accuracy of knowledge.

[0061] S303 incorporates expert experience and clinical guidelines, combining automated data acquisition with human expert review and the input of clinical practice guidelines. A professional team of nutritionists and physicians regularly reviews the automatically acquired knowledge, corrects errors, and supplements it with tacit knowledge that the automated system struggles to capture, based on the latest clinical experience and evidence-based medicine. This includes information such as the impact of dietary habits in specific regions on chronic disease interventions. For instance, dietary intervention recommendations for some rare chronic diseases may require the experience of specialist physicians. Simultaneously, the sections on dietary interventions in nationally published clinical treatment guidelines for chronic diseases are structured and input, such as dietary management guidelines for diabetic patients and dietary guidelines for hypertension patients. These guidelines provide a fundamental compliance guarantee for recipe recommendations.

[0062] S304 aims to build a multilingual and multicultural food database, encompassing distinctive foods from around the world, along with their nutritional components and cooking methods. For example, in addition to standard grains, vegetables, and meats, it includes regionally unique bean products, spices, and local snacks. Each food item will have detailed nutritional information, such as protein, fat, carbohydrates, vitamins, minerals, and dietary fiber content, as well as common cooking methods, allergen information, and seasonal availability. The database will be regularly updated to include newly introduced or improved food varieties and categorized using cultural tags.

[0063] S4. Construct a personalized nutrition needs model. Utilizing the user's multimodal health characteristics output from S2 and the latest nutritional knowledge acquired in S3, a deep learning model is used to infer the user's specific requirements and limitations in macronutrients, micronutrients, and energy intake. This includes the following steps:

[0064] S401 Initialize the deep learning model architecture. Select and initialize a multi-layer deep learning model architecture, such as a hybrid model containing multiple feedforward neural networks, long short-term memory layers, and attention mechanisms. The model's input layer receives the user's multimodal health feature vector generated in S2, including physiological and biochemical indicators, disease embedding vectors, and semantic features related to dietary habits and lifestyles. The output layer is set to predict the requirements or limits of various nutrients, such as target energy intake, carbohydrate intake range, protein intake, fat intake, upper and lower limits of sodium intake, upper or lower limits of potassium intake, calcium intake, and recommended amounts of various vitamins and minerals. The model's intermediate layers learn the complex mapping relationship between input features and output nutritional requirements through non-linear activation functions, such as modified linear units.

[0065] S402 integrates user features with knowledge graph embeddings, fusing the user's multimodal health feature vectors with the embedding vectors of user-related knowledge entities in the knowledge graph constructed in S3. For example, the user's disease embedding vector interacts with the "disease-nutrient" relationship embeddings in the knowledge graph to activate nutrient restriction knowledge related to that disease. This fusion can be achieved through an attention mechanism, allowing the model to automatically learn which knowledge fragments are most important for predicting the current user's nutritional needs. For example, for a hypertensive patient, the model will give higher attention weight to knowledge related to "sodium intake limits." This deep fusion enables the model to not only learn from the user's own data but also to use rich professional knowledge for reasoning.

[0066] S403 predicts nutrient requirements based on chronic disease type and physiological indicators. The model uses inference based on the user's chronic disease type and real-time physiological indicators to predict the specific range of nutrient requirements for various nutrients. For example, for patients with type 2 diabetes, the model predicts their appropriate daily carbohydrate intake and its distribution between meals based on their current blood glucose level, historical glycated hemoglobin data, and exercise intensity, and may recommend low-glycemic index carbohydrates. For patients with hyperlipidemia, the model predicts the upper limit of their saturated fat and cholesterol intake. The prediction results are given in range form, such as a carbohydrate intake range of 200 to 250 grams, to provide some flexibility. The model also considers the synergistic effects between nutrients, such as vitamin D and calcium absorption.

[0067] S404 adjusts nutrient weights based on dietary preferences and lifestyle. Building upon the predictions in S403, it further fine-tunes nutrient requirements by incorporating the user's dietary preferences and lifestyle. For example, if a user has a strong preference for a vegetarian diet, the model will adjust protein sources, recommending more plant-based protein and potentially increasing the recommended intake of certain micronutrients such as iron and zinc to compensate for the lower absorption rates of plant-based foods. If a user has a high-intensity exercise routine, the model will appropriately increase their energy and protein intake requirements. This adjustment is achieved through an internal weight adaptive mechanism, ensuring that the final nutritional requirements are both scientifically sound and aligned with the user's actual lifestyle.

[0068] S405 outputs a personalized nutrient requirement vector. After the aforementioned multi-layer processing, the output layer of the deep learning model generates the personalized nutrient requirement vector. This vector describes the current user's specific requirements, recommended ranges, or limits for energy, macronutrients such as carbohydrates, protein, and fat, and micronutrients such as vitamins and minerals, as well as avoidance requirements for specific food components. For example, the vector might contain details such as "daily energy intake of 2000 calories," "carbohydrates accounting for 45% to 50% of total energy," "sodium intake below 2.3 grams," and "avoiding foods containing gluten." This vector serves as a key input to the recipe generation and optimization module, guiding subsequent recipe construction.

[0069] S5 generates candidate recipes that meet user needs. Based on the personalized nutritional needs model built in S4, as well as a vast food database and culinary knowledge base, it intelligently filters, combines, and generates a series of candidate recipes that meet the user's nutritional, health, and preference constraints. Specifically, it includes the following steps:

[0070] S501, filtering a list of ingredients that meet nutritional requirements: First, based on the personalized nutrient requirement vector output in S104, the system filters basic ingredients that meet nutritional requirements from a multilingual and multicultural food database. For example, if the user needs a high-protein, low-carbohydrate diet, the system will prioritize ingredients such as chicken breast, fish, tofu, eggs, and dark leafy greens, while excluding high-starch foods such as rice and noodles. Simultaneously, the system strictly adheres to the user's allergen and contraindicated food list to ensure that the selected ingredients do not contain any components that may cause adverse reactions. For each selected ingredient, the system obtains its detailed nutritional data as the basis for subsequent combinations.

[0071] S502: Based on the cooking knowledge base, a preliminary dish combination is generated. After obtaining the ingredient list, the system combines the selected ingredients with the cooking knowledge base to generate a preliminary dish combination. The cooking knowledge base stores a large amount of information on recipes, cooking methods, and seasoning usage. The system prioritizes dishes that are simple to prepare and take little time, based on the user's cooking habits and preparation time budget. For example, if the user prefers a light flavor and has limited preparation time, the system will tend to recommend dishes such as steamed fish and cold vegetable salad. In addition, the system also considers the incompatibilities or synergies between ingredients, such as avoiding combining incompatible foods or pairing foods that help with nutrient absorption.

[0072] S503 constructs a multi-meal meal plan throughout the day, further integrating it into a complete meal plan that meets the user's needs for three or more meals a day. The system considers the balanced distribution of nutrients throughout the day, such as the energy and nutrient ratios of breakfast, lunch, and dinner, avoiding any meal being over- or under-nourished. For example, it will allocate more carbohydrates to breakfast and lunch to meet the energy needs of daytime activities, while dinner will focus on light, easily digestible protein and vegetables. Simultaneously, the system also considers the intervals between meals to ensure the user eats regularly. For patients with chronic diseases who need snacks, such as diabetic patients, the system will generate additional snack suggestions that meet their blood sugar control needs.

[0073] S504 generates a diverse list of candidate recipes. To increase user flexibility and meet diverse needs, multiple candidate recipes are generated that meet the same nutritional requirements but differ in ingredients, dishes, or cooking methods. The diverse generation strategy can be based on advanced artificial intelligence technologies such as random sampling, constraint-based search, or generative adversarial networks, ensuring that the generated recipes are sufficiently rich and attractive while meeting core nutritional requirements. Ultimately, a candidate recipe list containing multiple complete daytime meal plans is formed for subsequent evaluation and optimization.

[0074] S6 assesses the overall adaptability of the candidate recipes, primarily through a multi-dimensional and refined evaluation of the candidate recipes generated in S105. This ensures that the recommended recipes are not only nutritionally sound but also perform well in terms of user compliance, ingredient availability, cost-effectiveness, and cultural adaptability. Specifically, this includes the following steps:

[0075] S601 assesses the nutritional balance and compliance with chronic disease indicators. The system calculates the nutrient content of each candidate recipe, including energy, macronutrients such as carbohydrates, protein, fat, and dietary fiber, and micronutrients such as vitamins and minerals. This content is then compared to the personalized nutrient requirement vector generated in S4. The system evaluates whether the recipe meets the user's minimum requirements without exceeding the maximum limits. Simultaneously, the system simulates the impact of the recipe on the user's chronic disease indicators. For example, for diabetic patients, the system predicts the theoretical peak and fluctuation trend of postprandial blood glucose based on the types and amounts of carbohydrates in the recipe, combined with the user's insulin sensitivity, to ensure that the recipe does not cause drastic fluctuations in blood glucose. For patients with hyperlipidemia, the system assesses whether the saturated fat and cholesterol content meets the target for lowering blood lipids.

[0076] S602, User Compliance and Taste Preference Assessment, evaluates user acceptance of candidate recipes by analyzing user dietary habits and taste preference characteristics coded in S2. For example, if a user explicitly states they dislike spicy food, the system will lower the rating of recipes containing chili peppers. If historical user feedback data shows an aversion to a certain ingredient, such as cilantro, the rating of recipes containing cilantro will decrease significantly.

[0077] S603, Ingredient Availability and Cost-Effectiveness Assessment: This assesses the availability and affordability of ingredients required for candidate recipes in the user's region. The system integrates data from major local supermarkets and farmers' markets to obtain real-time inventory and price information for ingredients. It also evaluates whether any recipes contain rare or difficult-to-procure ingredients, or whether the total cost of ingredients exceeds the user's budget.

[0078] S604, Cooking Ease of Use and Meal Preparation Time Assessment, evaluates the complexity of candidate recipes based on user lifestyle data coded in S2, particularly their meal preparation time budget. The system analyzes the cooking steps, required time, and necessary utensils for each dish, then calculates the total meal preparation time for all recipes throughout the day. If a user's meal preparation time budget is 30 minutes, but the total meal preparation time for a particular recipe exceeds one hour, the cooking ease of use score for that recipe will be very low. This assessment aims to meet users' needs for an efficient lifestyle and reduce adherence barriers caused by cumbersome cooking processes.

[0079] S605, Cultural and Regional Adaptability Assessment, primarily considers the dietary culture and customs of the user's region to ensure that recommended recipes are culturally appropriate. For example, in northern China, recipes may include a greater proportion of noodles; in southern China, rice and soups may be recommended more often. For ethnic minority users, the system will avoid recommending foods that are forbidden to their diet and prioritize dishes that conform to their traditional dietary culture.

[0080] S606, Generate a Comprehensive Recipe Score. Through the multi-dimensional evaluation described above, the system calculates a comprehensive score for each candidate recipe. This comprehensive score is the final result obtained by weighted summation of all evaluation sub-items or other multi-objective optimization algorithms. The weights can be dynamically adjusted based on the priorities set by the user in their personalized needs input in S1. For example, for a cost-sensitive user, cost-effectiveness will have a higher weight. Finally, the recipe with the highest comprehensive score or several top-scoring recipes are recommended.

[0081] S7 recommends optimized nutritional recipes to users, primarily presenting the recipes evaluated and selected in S106 in a clear, intuitive, and easy-to-understand manner. The success of this step lies not only in providing the recipes themselves, but also in offering detailed explanations and health guidance to help users understand and effectively follow them. Specifically, the optimized nutritional recipes are presented in a user-friendly interface through the recipe display unit of the user interface module. This includes high-resolution images of each dish, its name, main ingredients, detailed cooking steps, and estimated cooking time. The recipes are clearly divided by meal type, such as breakfast, lunch, dinner, and snacks. Visual design enhances the user's intuitive experience and appetite, increasing acceptance.

[0082] S8, which collects user feedback and updates the model, is a key component for achieving adaptive learning and continuous optimization, solving the problems of poor knowledge updates and adaptability in traditional systems. By proactively collecting users' actual experiences and health benefits of recommended recipes and integrating this feedback data into model training, the system can continuously evolve, providing more accurate and user-relevant recommendations. Specifically, the feedback collection unit in the user interface module obtains direct user evaluations of recommended recipes. This includes ratings for each dish (e.g., one to five stars), taste evaluations, and difficulty ratings. Users can also input free text comments to express their satisfaction with the recipes, suggestions for improvement, or problems encountered.

[0083] Based on the above information and practical scenarios, the specific usage methods are as follows:

[0084] 1. When a new user registers or an existing user logs into the system, the data collection module acquires their personalized health data, including real-time physiological and biochemical indicators such as blood sugar and blood pressure, chronic disease diagnoses and medical history information such as diabetes and allergens, and detailed dietary habits and lifestyle data such as taste preferences and exercise frequency. This raw data undergoes preliminary cleaning and integration within the data collection module to ensure data quality.

[0085] 2. The data preprocessing and feature engineering module receives these raw data, performs deep cleaning and normalization on them, and uses advanced technologies such as temporal feature extraction, entity embedding encoding, and semantic feature encoding to transform heterogeneous multimodal health data into unified, high-dimensional user multimodal health feature vectors. The feature vectors comprehensively and concisely represent the user's current health status and personalized needs. In addition, the knowledge base management module continuously acquires and integrates the latest nutritional research results, the correlation between diseases and nutrition, and dietary guidelines from authoritative sources, and builds and maintains a dynamically updated knowledge graph and multicultural food database.

[0086] 3. The personalized nutrition needs modeling module receives the user's multimodal health feature vector and relevant knowledge entity embeddings provided by the knowledge base management module. Internally, it employs a multi-layered deep learning model architecture, deeply fusing user features with knowledge graph embeddings. Based on the user's chronic disease type, physiological indicators, dietary preferences, and lifestyle, it constructs a nutrition needs model, ultimately outputting a precise personalized nutrient requirement vector. This vector not only includes specific recommended amounts or limits for various macronutrients and micronutrients but also reflects the avoidance requirements for specific food components.

[0087] 4. The recipe generation and optimization module uses the nutrient requirement vector output by the personalized nutrition requirement modeling module, combined with a multilingual and multicultural food database and cooking knowledge base, to intelligently select ingredients, combine dishes, and build a multi-meal meal plan for the day, generating a series of candidate recipes that meet the user's nutritional and health constraints.

[0088] 5. The recipe generation and optimization module further conducts a comprehensive adaptability assessment of these candidate recipes. Assessment dimensions include nutritional balance and compliance with chronic disease indicators, user compliance and taste preferences, ingredient availability and cost-effectiveness, ease of cooking and preparation time, and cultural and regional adaptability. Through these refined assessments, the system calculates a comprehensive score for each candidate recipe and selects the recipe with the highest score as the final recommendation.

[0089] 6. The user interface module presents optimized nutritional recipes to users in a visual manner, providing detailed nutritional analysis reports, explanations of the correlation between the recipes and chronic disease intervention goals, and practical meal preparation tips and precautions. The system also offers recipe adjustment and saving functions to enhance the user experience. During this process, the user interface module continuously collects explicit user feedback on recommended recipes, such as ratings and comments, as well as implicit feedback from changes in user health indicators and adherence monitored by the system backend. This feedback data is transmitted to the model training and update module for continuous retraining and iterative updates of the personalized nutritional needs model. It also feeds back into the knowledge base management module and recommendation strategies, thereby achieving adaptive learning and continuous optimization of the entire system. Through this closed-loop feedback mechanism, the system can provide users with increasingly accurate and effective personalized nutritional recipe recommendations over time.

[0090] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A system for the use of an artificial intelligence-based nutritional recipe recommendation system in the management of chronic disease interventions, characterized by, The system comprises a data collection module, a data preprocessing and feature engineering module, a knowledge base management module, a personalized nutritional demand modeling module, a recipe generation and optimization module, a user interface module, and a model training and updating module, and the modules are connected through network communication; The data collection module is used to collect health-related data of the user, including a biological sensor interface unit, a medical record integration unit, and a user input unit; The data preprocessing and feature engineering module is used to perform deep cleaning, normalization, feature extraction, and feature conversion on the raw data to generate a standardized feature vector; The knowledge base management module is used to collect, update, and maintain professional knowledge related to nutrition and chronic disease intervention; The personalized nutritional demand modeling module is used to combine user-specific features and professional knowledge base to construct a user-specific nutritional demand model; The recipe generation and optimization module is used to generate candidate recipes based on the nutritional demand model and perform multi-dimensional optimization; The user interface module is used for interaction between the system and the user; The model training and updating module is used to continuously train and iteratively update the artificial intelligence model in the personalized nutritional demand modeling module.

2. The system for the intervention management of chronic diseases based on the artificial intelligence-based nutritional recipe recommendation system according to claim 1, characterized in that: The biological sensor interface unit is used to connect wearable devices or medical-grade sensors to obtain real-time physiological indicators of the user, including blood glucose value, blood pressure data, body weight, body fat percentage, and heart rate variability; the medical record integration unit is used to import the user's chronic disease diagnosis information, medical history, medication, and medical examination reports from the hospital information system or personal health records; and the user input unit is used to obtain the user's dietary habit preferences, food allergy information, dietary taboos, daily routines, exercise intensity, personalized dietary needs, and feedback data on recommended recipes.

3. The system for the intervention management of chronic diseases based on the artificial intelligence-based nutritional recipe recommendation system according to claim 1, characterized in that: The data preprocessing and feature engineering module includes a data cleaning unit, a normalization unit, a feature extraction unit, and a feature conversion unit; the data cleaning unit is used to handle missing values, outliers, and noise; the normalization unit is used to unify features with different dimensions and value ranges to a specific interval; the feature extraction unit is used to mine high-level and meaningful features from raw data; and the feature conversion unit is used to encode features of different modalities.

4. The system for the intervention management of chronic diseases based on the artificial intelligence-based nutritional recipe recommendation system according to claim 1, characterized in that: The knowledge base management module includes a knowledge acquisition unit, a knowledge storage unit, and a knowledge updating unit; the knowledge acquisition unit is used to acquire nutrition knowledge, food nutrient data, disease and nutrition association rules, and dietary intervention recommendations from authoritative nutrition journals, medical research reports, national dietary guidelines, food composition databases, and disease clinical pathways; the knowledge storage unit uses a graph database or a relational database to structure the knowledge, including constructing a knowledge graph, in which nodes represent food, nutrients, diseases, symptoms, and user attribute entities, and edges represent relationships between entities; and the knowledge updating unit is used to periodically or trigger-based update the knowledge base.

5. The system for the intervention management of chronic diseases based on the artificial intelligence-based nutritional recipe recommendation system according to claim 1, characterized in that: The personalized nutritional requirement modeling module comprises a multi-layer neural network structure, which includes a deep neural network, a long short-term memory network or a transformer network; the input of the multi-layer neural network structure is a user standardized feature vector output by the data preprocessing and feature engineering module and a related knowledge entity embedding vector provided by the knowledge base management module, which is used to learn the complex nonlinear relationship between the user features and the nutritional requirements and dynamically infer the specific requirement amount and limitation of macro and micro nutrients.

6. The system for the intervention management of chronic diseases based on the artificial intelligence-based nutritional recipe recommendation system according to claim 1, characterized in that: The recipe generation and optimization module comprises a recipe generation unit and a recipe optimization unit; the recipe generation unit is used to screen food material combinations from a food database according to nutritional component constraints and food allergen avoidance, and combine specific dishes and recipes in combination with a cooking knowledge base; the recipe optimization unit is used to perform multi-objective optimization on candidate recipes, and the optimization objectives include nutritional balance, compliance, food material availability, meal preparation convenience and cost-effectiveness, and the optimization algorithm adopts a multi-objective genetic algorithm or a reinforcement learning method.

7. The system for the intervention management of chronic diseases based on the artificial intelligence-based nutritional recipe recommendation system according to claim 1, characterized in that: The user interface module comprises a recipe display unit, a feedback collection unit and a health report unit; the recipe display unit is used to present the recommended nutritional recipes in an interface, including dish name, picture, method, food material, nutritional component analysis and correlation description with chronic disease management goals; The feedback collection unit is used to obtain the scores, comments, preference marks and actual diet and health index changes of the user on the recommended recipes; and the health report unit is used to generate a personalized health report regularly, which summarizes the user's diet performance, chronic disease index changes, nutritional intake analysis and health suggestions.

8. The application method of the artificial intelligence-based nutritional recipe recommendation system according to any one of claims 1-7 in the management of chronic disease intervention, characterized in that, The method comprises the following steps: S1, obtaining user personalized health data, mainly comprehensively and accurately collecting various health-related information reflecting individual differences of the user; S2, multi-modal feature coding of user health data, converting the original and heterogeneous user health data obtained in S1 into feature vector representation through artificial intelligence technology; S3, obtaining the latest nutrition and chronic disease intervention knowledge, mainly ensuring the scientificity and timeliness of the recommended recipes and solving the problem of slow knowledge updating of traditional systems; S4, constructing a personalized nutritional requirement model, mainly tailoring the precise nutritional requirements of each user according to their unique characteristics; S5, generating candidate recipes meeting the user's requirements, mainly intelligently screening, combining and generating a series of candidate recipes meeting the user's nutritional, health and preference constraints according to the personalized nutritional requirement model constructed in S1, a large food database and a cooking knowledge base; S6, evaluating the comprehensive adaptability of the candidate recipes, mainly performing multi-dimensional and refined evaluation on the candidate recipes generated in S5 to ensure that the recommended recipes are not only reasonable in nutrition; S7, recommending the optimized nutritional recipes to the user, mainly presenting the nutritional recipes evaluated and optimized in S6 to the user in a clear, intuitive and easy-to-understand manner; S8, collecting user feedback and updating the model, mainly actively collecting the actual experience and health improvement effect of the user on the recommended recipes, and integrating these feedback data into the model training.

Images