User profiling system based on patient pro-inflammatory dietary behavior analysis

Abstract

The present application belongs to the technical field of artificial intelligence and medical health, and specifically relates to a user portrait generation system based on patient pro-inflammatory dietary behavior analysis. The system collects dietary and physiological data, calculates time series pro-inflammatory load index, identifies dietary behavior patterns, and integrates multi-dimensional characteristics such as clinical, genetic, microbial and lifestyle, uses an interpretable machine learning model to generate a dynamic user portrait containing pro-inflammatory risk level, behavior vulnerability dimension and intervention suggestions, and supports accurate nutrition intervention and continuous tracking. The system outputs the dietary behavior vulnerability dimension through the interpretable machine learning model, so that doctors or nutritionists can quickly understand the core problems of patients and develop targeted strategies, greatly improving the efficiency of doctor-patient communication and intervention compliance.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of artificial intelligence and medical health, and particularly relates to a user portrait generation system based on patient pro-inflammatory dietary behavior analysis. BACKGROUND

[0002] In the field of digital health and precision nutrition management, constructing user portraits based on individual behavior data has become an important technical path to realize personalized intervention. This field combines nutritional epidemiology, behavior science and artificial intelligence technology, aiming to analyze users' eating habits, lifestyle and health indicators, identify potential health risks and provide customized recommendations. Among them, pro-inflammatory dietary behavior is a key inducement of chronic inflammation and related metabolic diseases, and its quantitative evaluation and dynamic tracking are of great significance for disease prevention.

[0003] Among them, the user portrait generation system based on patient pro-inflammatory dietary behavior analysis focuses on extracting pro-inflammatory features from multi-source dietary records (such as food frequency questionnaires, mobile application logs or image recognition data), and constructing structured user portraits reflecting individual inflammation risk levels. The core of this technical direction is to convert complex dietary patterns into computable inflammation potential indicators, and to associate and model with other health dimensions of users to support clinical decision-making or health management services.

[0004] Existing technologies face multiple challenges in realizing such user portraits: dietary data collection relies on user active input, with problems of recording bias and insufficient completeness; pro-inflammatory scoring models are mostly based on static food composition databases, making it difficult to dynamically adapt to individual metabolic differences and the influence of food processing methods; existing portrait systems lack the ability to integrate behavior time series and situational factors (such as meal times and emotional states), resulting in coarse-grained portraits and limited prediction validity; in addition, different data sources have strong semantic heterogeneity, and there is no unified pro-inflammatory behavior representation framework, which restricts the general application of portraits in cross-platform health services. Therefore, there is an urgent need for a systematic solution that can integrate multi-dimensional dynamic dietary behavior data, accurately quantify individual pro-inflammatory risk and generate high-explainability user portraits. SUMMARY

[0005] The purpose of the present application is to provide a user portrait generation system based on patient pro-inflammatory dietary behavior analysis, to solve the problem of lack of dynamic, quantitative and structured modeling of individual dietary pro-inflammatory potential in current chronic inflammation-related disease management, leading to difficulties in developing personalized nutrition intervention strategies, and difficulties in predicting and tracking intervention effects.

[0006] The technical scheme of the present application is a user portrait generation system based on patient pro-inflammatory dietary behavior analysis, which comprises a dietary data acquisition module, a pro-inflammatory index calculation module, a behavior pattern recognition module, a multi-dimensional feature fusion module, and a dynamic portrait generation module; the dietary data acquisition module is used to continuously acquire the dietary intake records of the user through a mobile terminal or a wearable device, and the dietary intake records include food types, intake amounts, intake times, intake frequencies, and accompanying physiological state data; the pro-inflammatory index calculation module is used to call a pre-constructed dietary pro-inflammatory potential database based on the dietary intake records, map the pro-inflammatory scores of each food item, and perform weighted calculation combined with the intake amounts to generate a time-sequenced individual pro-inflammatory load index; the behavior pattern recognition module is used to perform time-series pattern mining on the time-sequenced individual pro-inflammatory load index to identify statistically significant dietary behavior patterns, including a periodic burst pattern of high pro-inflammatory load, a stable pattern of sustained low pro-inflammatory load, and a correlation pattern of pro-inflammatory load fluctuation with circadian rhythm or emotional state; the multi-dimensional feature fusion module is used to perform cross-domain alignment and feature embedding of the dietary behavior patterns with clinical indicator data, genetic polymorphism information, gut microbiome characteristics, and lifestyle parameters to construct a high-dimensional heterogeneous feature vector; and the dynamic portrait generation module is used to generate a structured user portrait based on the high-dimensional heterogeneous feature vector through an interpretable machine learning model, and the user portrait includes a pro-inflammatory risk level, a dietary behavior vulnerability dimension, a nutrition intervention sensitive window period, and personalized dietary adjustment recommendations.

[0007] Further, the dietary pro-inflammatory potential database is constructed based on large-scale epidemiological research and in vitro cytokine release experiment data, and each food item is associated with a standardized pro-inflammatory score, which comprehensively reflects the potential influence intensity of the food on serum C-reactive protein, interleukin 6, and tumor necrosis factor alpha levels after intake, with a score range of -10 to +10, where negative values represent anti-inflammatory effects and positive values represent pro-inflammatory effects.

[0008] Further, the pro-inflammatory index calculation module uses the logarithmic transformed value of the intake amount as a weight factor when performing weighted calculation, and introduces a food processing method correction coefficient, which quantitatively adjusts the amplification or inhibition effect of different processing methods such as frying, grilling, pickling, or raw food on pro-inflammatory potential, with a correction coefficient value range of 0.5 to 2.0.

[0009] Further, the behavior pattern recognition module uses a sliding window combined with a hidden Markov model to divide the individual pro-inflammatory load index sequence into low, medium, and high pro-inflammatory states, and decodes the most likely state transition path through the Viterbi algorithm to extract state residence time, state transition frequency, and state trigger conditions as core parameters of the behavior pattern.

[0010] Further, the multi-dimensional feature fusion module adopts a graph neural network architecture to realize cross-domain feature alignment, wherein nodes represent feature entities of different data sources, edges represent biological correlations or statistical correlations between features, neighbor information is aggregated through a multi-layer message passing mechanism to generate a unified embedding vector for each user, and the embedding vector has a dimension of 128.

[0011] Further, the interpretable machine learning model adopted by the dynamic portrait generation module is a gradient boosting tree ensemble model based on an attention mechanism, which outputs contribution weights of each input feature to the prediction result while predicting the pro-inflammatory risk level, and the contribution weights are used to generate explanatory labels of the dietary behavior vulnerability dimension, including “high sugar driven type”, “red meat dependent type”, “processed food sensitive type” or “dietary fiber deficiency type”.

[0012] Further, the dynamic portrait generation module is also configured with a portrait update triggering mechanism, which automatically triggers incremental update of the user portrait and records a portrait version change log when newly collected dietary data causes a change in the 7-day moving average of the individual pro-inflammatory load index of more than 15%, or when the behavior pattern recognition module detects a new significant behavior pattern.

[0013] Further, the system further comprises an intervention feedback closed loop module for receiving user feedback data on the execution of personalized dietary adjustment suggestions, the execution feedback data including the suggestion adoption rate, the actual intake deviation and the subjective feeling score, and feeding back the execution feedback data to the multi-dimensional feature fusion module for dynamically calibrating the intervention sensitive window period parameters in the user portrait.

[0014] Further, the clinical indicator data includes fasting blood glucose, glycosylated hemoglobin, blood lipid profile, liver function indicators and systemic inflammation markers; the genetic polymorphism information includes single nucleotide polymorphism sites related to inflammation pathways, covering TNF, IL6, CRP and PPARG genes; the intestinal microbiome features include the relative abundance ratio of pro-inflammatory bacteria and anti-inflammatory bacteria; and the lifestyle parameters include sleep duration, physical activity intensity, stress level and smoking and drinking status.

[0015] Further, the system runs in a cloud-edge collaborative computing architecture, the dietary data collection module is deployed on the user terminal device, the pro-inflammatory index calculation module and the behavior pattern recognition module are deployed on the edge computing node to ensure data privacy and real-time performance, and the multi-dimensional feature fusion module and the dynamic portrait generation module are deployed on the cloud server to utilize high-performance computing resources for complex model inference.

[0016] Compared with the prior art, the advantages and positive effects of the present application are that:

[0017] The present application first constructs a complete technical chain from original dietary behavior to structured user portrait, breaks through the limitation of traditional nutrition assessment which only relies on static dietary frequency questionnaire or single nutrient analysis by quantifying the pro-inflammatory load of individual diet and identifying its dynamic behavior pattern; the system introduces multi-source heterogeneous health data for cross-domain fusion, so that the generated user portrait not only reflects the diet behavior itself, but also embeds the individual biological background and lifestyle context, significantly improving the clinical relevance and intervention guidance value of the portrait; the dynamic updating mechanism and intervention feedback closed loop design used make the user portrait have the ability of continuous evolution, which can truly reflect the dynamic process of user behavior change and physiological response, and provide a quantifiable, traceable and interpretable decision support tool for precision nutrition intervention of chronic inflammation related diseases; in addition, the system outputs the dietary behavior vulnerability dimension through an interpretable machine learning model, so that doctors or nutritionists can quickly understand the core problem of patients and develop targeted strategies, greatly improving the efficiency of doctor-patient communication and intervention compliance. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is the overall technical scheme architecture schematic diagram of the present application;

[0019] Figure 2 is the core principle framework schematic diagram of multi-dimensional feature fusion and dynamic portrait generation in the present application. DETAILED DESCRIPTION

[0020] Embodiment 1

[0021] In order to make the purpose, technical scheme and advantages of the present application more clear and explicit, the present application is described in detail below in combination with specific embodiments. The present embodiment provides a user portrait generation system based on patient pro-inflammatory dietary behavior analysis, and the core goal is to construct a dynamic, quantitative and structured technical means to evaluate the pro-inflammatory potential of individual diet, thereby solving the technical difficulties of current personalized nutrition intervention strategy development, intervention effect prediction and tracking in chronic inflammation related disease management. Through fine data collection, multi-level data processing, intelligent pattern recognition and interpretable machine learning model, the present system realizes the overall transformation from original dietary behavior data to structured user portrait with clinical guidance significance.

[0022] The workflow of the present system starts from the continuous acquisition of dietary data, then quantitatively evaluates the pro-inflammatory potential of these data, and further identifies the unique dietary behavior patterns of users. Then, the system deeply fuses these behavior patterns with multi-source heterogeneous biomedical and lifestyle data, and finally generates a structured user portrait through an interpretable machine learning model. The entire system architecture is designed as a cloud-edge collaborative computing mode, ensuring the balance between data privacy, real-time processing and high-performance complex model inference.

[0023] Dietary data collection module: The dietary data collection module is the front end of the system. Its function is to continuously obtain the user's dietary intake records through mobile terminals or wearable devices. These records include food types, intake amounts, intake times, intake frequencies, and accompanying physiological state data. The design of this module fully considers user friendliness, data collection continuity, accuracy, and privacy security.

[0024] Data collection mechanism: This module uses smartphones, smartwatches, and other mobile terminals or wearable devices carried by users daily as data input interfaces. A specially designed application is deployed on the mobile terminal, which supports users to conveniently record their dietary intake through various methods such as text input, voice recognition, image recognition, and barcode scanning. For food types, the application provides a pre-set food ontology library for intelligent matching, while also allowing users to customize new food items and upload pictures for auxiliary identification. Intake amounts are quantified by standardized units (such as grams, milliliters, servings) and can be obtained through user input, image estimation, or smart scale linkage. Intake time and frequency are automatically recorded by the application.

[0025] Acquisition of accompanying physiological state data: In addition to dietary information, this module also collects users' physiological state data through wearable devices such as smartwatches and smartbands. These data include, but are not limited to, heart rate, steps, sleep duration, body temperature, and fasting blood glucose and postprandial blood glucose data obtained through external interface connections. All physiological data are synchronized to the mobile terminal application in real time or near real time, and are time-stamped aligned with the corresponding dietary intake records to form a complete dietary-physiological correlation data set.

[0026] Data preprocessing and storage: Before data is transmitted to the backend for processing, the dietary data collection module performs preliminary data verification and cleaning on the terminal device. This includes checking the format of the data for standardization, removing duplicate records, and handling null and abnormal values (e.g., through interpolation, mean filling, or warning users to correct). After encryption locally, the data is uploaded to edge computing nodes or cloud storage through secure transmission protocols (e.g., Transport Layer Security). To ensure data traceability and integrity, each data record contains a unique identifier, timestamp, geographic location information (optional), and data source device identifier. Data is stored in a distributed time series database in the cloud to facilitate efficient querying and analysis by subsequent modules.

[0027] Proinflammatory index calculation module: The proinflammatory index calculation module is responsible for converting the original dietary intake records obtained by the dietary data collection module into time series individual proinflammatory load indexes. This conversion process is the core of the system to quantify the proinflammatory potential of diet.

[0028] Data input matching with food items: This module receives structured dietary intake records from the dietary data collection module. For each food category in a record, the module first performs an exact match with the pre-built database of dietary pro-inflammatory potential. The matching process employs a text similarity-based algorithm (e.g., cosine similarity, Jaccard similarity) combined with hierarchical relationships in the food ontology for multi-level matching to ensure accuracy of identification. If a new food item not present in the database is encountered, the system flags it as a pending item for review and can trigger manual review or supplementary learning from external nutrition knowledge graphs.

[0029] Dietary pro-inflammatory potential database: This database is the foundation for the pro-inflammatory score mapping performed by this module. The database is constructed based on large-scale epidemiological study data and in vitro cytokine release experiment data. Epidemiological studies provide evidence of long-term associations between food intake and population inflammatory markers (e.g., high-sensitivity C-reactive protein, interleukin 6, tumor necrosis factor alpha), while in vitro experiments provide direct effects of food extracts or metabolites on immune cell inflammation factor secretion. Each food item is associated with a standardized pro-inflammatory score in this database, ranging from -10 to +10. Negative values indicate anti-inflammatory effects, which can suppress inflammatory responses or promote inflammation resolution, while positive values indicate pro-inflammatory effects, which can induce or exacerbate inflammatory responses. For example, deep-sea fish rich in omega-3 fatty acids may have a negative score, while processed foods high in saturated fat and sugar may have a positive score. The database is maintained using a version control mechanism and is regularly updated and verified through new scientific research findings.

[0030] Pro-inflammatory score weighted calculation: After successfully matching the food item and obtaining its base pro-inflammatory score, this module performs a weighted calculation combined with the intake amount. The weight factor here uses the log-transformed value of the intake amount to reflect the biological characteristic that the pro-inflammatory burden is not linearly related to the intake amount, but the marginal effect gradually decreases with the increase of intake amount. Specifically, if the intake amount is , the weight factor is or , where the base is selected according to the data distribution characteristics.

[0031] Food processing mode correction: To more accurately assess the pro-inflammatory potential of food, this module further introduces a food processing mode correction factor. This correction factor quantitatively adjusts the amplification or inhibition effect of different processing methods such as frying, grilling, pickling, raw food, boiling, etc. on the pro-inflammatory potential of food. The correction factor ranges from 0.5 to 2.0. For example, for a certain original pro-inflammatory score of 2, if it is fried, its correction factor may be 1.5, and the pro-inflammatory score will be amplified to 2 x 1.5 = 3; if it is raw or simply boiled, the correction factor may be 0.8, and the pro-inflammatory score will be reduced to 2 x 0.8 = 1.6. These correction factors are based on research data on the impact of food science, nutrition and cooking on nutritional components, and can be refined according to different food categories. Finally, the individual pro-inflammatory load index of each meal or each day is the cumulative or average value of the corrected pro-inflammatory scores of all the food intake, forming a continuous time series data.

[0032] Behavior pattern recognition module: The behavior pattern recognition module aims to conduct in-depth analysis on the time series of individual pro-inflammatory load index generated by the pro-inflammatory index calculation module, in order to identify statistically significant dietary behavior patterns. These patterns can reveal the internal rules and potential risks of users' dietary behavior.

[0033] Time series pattern mining basis: This module receives continuous individual pro-inflammatory load index time series. In order to effectively identify patterns, the system first applies sliding window technology to segment the time series. The width of the sliding window can be dynamically adjusted according to the time granularity of the analysis (e.g. 24 hours, 7 days) and the fluctuation characteristics of the data, and the step length of the window determines the density of pattern recognition. Through sliding window, the system converts the continuous pro-inflammatory load index sequence into a series of local, time-contextual sub-sequences.

[0034] Hidden Markov Model State Division: Within each sliding window, or for the entire time series, the module uses a Hidden Markov Model to divide the individual pro-inflammatory load index sequence into states. Hidden Markov Model is a statistical model used to describe a random process with hidden variables. In this system, the observable variable is the pro-inflammatory load index, and the hidden variable is the user's dietary pro-inflammatory state. The system presets or learns through clustering algorithm to divide the pro-inflammatory state into low, medium and high three categories.

[0035] Low pro-inflammatory state: Corresponding to the pro-inflammatory load index continuously below a certain preset threshold, indicating that the dietary structure is dominated by anti-inflammatory food and the intake of processed food is extremely low.

[0036] Medium pro-inflammatory state: Corresponding to the pro-inflammatory load index fluctuating in the middle interval, indicating that the dietary structure is relatively balanced, or occasionally consuming pro-inflammatory food, but the frequency and amount are within a controllable range.

[0037] High pro-inflammatory state: Corresponds to a pro-inflammatory load index consistently above a certain high threshold, indicating a high proportion of pro-inflammatory foods in the dietary structure, or a high intake of processed foods.

[0038] The training data for the Hidden Markov Model comes from a large number of users' historical pro-inflammatory load index sequences. The model's parameters, including the initial state probability, state transition probability matrix, and observation probability distribution, are estimated using the Expectation-Maximization algorithm.

[0039] Viterbi algorithm decoding and pattern parameter extraction: After the Hidden Markov Model is trained, this module uses the Viterbi algorithm to decode the individual pro-inflammatory load index sequence to find the most likely state transition path. The Viterbi algorithm can efficiently calculate the most likely hidden state sequence corresponding to a given observation sequence. By analyzing this optimal path, the system can extract core behavioral pattern parameters:

[0040] State residence time: The length of time a user stays in a certain pro-inflammatory state (e.g., high pro-inflammatory state). A longer high pro-inflammatory state residence time indicates a sustained dietary pro-inflammatory risk.

[0041] State transition frequency: The frequency of a user's transitions between different pro-inflammatory states. Frequent transitions from low pro-inflammatory state to high pro-inflammatory state and then quickly recovering may indicate periodic "indulgence" behavior.

[0042] State trigger conditions: The module identifies external events or internal physiological changes that cause state transitions through correlation analysis and machine learning classifiers (e.g., decision trees or logistic regression). These trigger conditions may include specific times (e.g., weekends), specific emotional states (e.g., stress, anxiety), or associations with specific physiological indicators (e.g., blood glucose fluctuations). For example, the system may identify that "high pro-inflammatory load periodic burst patterns" usually occur during weekend dinners, or that "pro-inflammatory load fluctuates with emotional state" significantly increases when the user feels stressed.

[0043] Multi-dimensional feature fusion module: The core task of the multi-dimensional feature fusion module is to align and embed the dietary behavior patterns output by the behavior pattern recognition module with the user's clinical indicator data, genetic polymorphism information, gut microbiome characteristics, and lifestyle parameters into a high-dimensional heterogeneous feature vector. This process aims to integrate the user's multi-aspect health information into a comprehensive and unified digital representation.

[0044] Data sources and feature extraction: This module receives input from multiple independent data sources:

[0045] Dietary behavior pattern data: The behavior pattern parameters generated by the behavior pattern recognition module, such as the residence time, transition frequency, and corresponding trigger conditions of high, medium, and low pro-inflammatory states.

[0046] Clinical biomarker data: includes fasting glucose, glycosylated hemoglobin, lipid profile (e.g., total cholesterol, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, triglycerides), liver function indicators (e.g., alanine transaminase, aspartate transaminase, total bilirubin), and systemic inflammation markers (e.g., high-sensitivity C-reactive protein, erythrocyte sedimentation rate). These data are standardized and cleaned through the electronic health record system of the medical institution or test reports uploaded by the user.

[0047] Genetic polymorphism information: covers single nucleotide polymorphism sites (SNPs) related to inflammation pathways, such as specific sites of the tumor necrosis factor gene (TNF), interleukin 6 gene (IL6), C-reactive protein gene (CRP), and peroxisome proliferator-activated receptor gamma gene (PPARG). These data are obtained through genetic sequencing service institutions and are encoded for genotypes (e.g., 0, 1, 2 represent different allele combinations).

[0048] Intestinal microbiome characteristics: mainly include the relative abundance ratio of pro-inflammatory and anti-inflammatory bacterial genera. These data are obtained through 16S ribosomal ribonucleic acid sequencing or metagenomic sequencing analysis, and key microbial taxonomic unit abundance data are extracted through bioinformatics processing.

[0049] Lifestyle parameters: include sleep duration, physical activity intensity (e.g., daily steps, exercise minutes), stress level (estimated by heart rate variability through questionnaires or wearable devices), and smoking and drinking status. Some of these data are supplemented by physiological data from the dietary data collection module, and some are synchronized through user questionnaires or external smart devices.

[0050] Graph neural network architecture: to achieve effective alignment and fusion of cross-domain features, this module adopts a graph neural network architecture. In this architecture:

[0051] Node representation: feature entities from different data sources are abstracted as nodes in the graph. For example, a user's dietary pro-inflammatory pattern can be a node, their fasting glucose value is another node, a specific genotype is a node, and the abundance of a certain intestinal bacterial genus is also a node. To capture heterogeneity, nodes can be assigned different types (e.g., "dietary pattern node", "clinical indicator node", "gene node").

[0052] Edge representation: edges between nodes represent biological associations or statistical correlations between features. For example, there is a known biological pathway association between genes and disease markers, a causal relationship between dietary patterns and intestinal flora, and statistical correlations between different clinical indicators. This association information can be obtained from existing biomedical knowledge graphs, literature, or through correlation analysis on large-scale anonymous data. The weight of the edge can represent the strength or confidence of the association.

[0053] Message Passing Mechanism: The graph neural network aggregates neighbor information through a multi-layer message passing mechanism. At each layer, each node collects information from its neighbor nodes and updates its own representation by fusing its own features with neighbor features through an aggregation function (e.g., sum, average, max pooling, or more complex attention mechanisms). This iterative message passing enables efficient propagation and integration of information throughout the heterogeneous graph, capturing complex high-order relationships between multi-modal data.

[0054] Feature Embedding: After multiple layers of message passing, the graph neural network generates a unified embedding vector for each user. This embedding vector is a low-dimensional, dense representation of all fused features, with a dimension of 128. This 128-dimensional vector captures the user's comprehensive features in multiple dimensions such as dietary behavior, clinical, genetic, microbiome, and lifestyle, and provides highly refined input for the subsequent dynamic profile generation module.

[0055] Dynamic Profile Generation Module: The dynamic profile generation module is the decision-making core of the system, which functions to generate a structured user profile based on the high-dimensional heterogeneous feature vector output by the multi-dimensional feature fusion module through an interpretable machine learning model. This profile not only provides pro-inflammatory risk assessment, but also reveals the underlying dietary behavior vulnerability in depth and provides specific intervention recommendations.

[0056] Interpretable Machine Learning Model: This module uses a gradient boosting tree ensemble model based on attention mechanism. Gradient boosting trees (such as XGBoost or LightGBM) are powerful ensemble learning models that are good at handling heterogeneous data and have good predictive performance. By introducing attention mechanisms, such as SHAP (Shapley Additive explanations) or LIME (Local Interpretable Model-agnostic Explanations) methods, the model can output the contribution weights of each input feature (i.e., each dimension in the 128-dimensional embedding vector, or back to the original features) to the prediction result while predicting the pro-inflammatory risk level. These contribution weights are the key to the model's "explanation" of its decision-making process.

[0057] User Profile Content Composition: The generated structured user profile contains multiple key components:

[0058] Pro-inflammatory risk level: The system divides the user's pro-inflammatory risk into, for example, low, medium, and high levels according to the prediction results. This provides a visual risk assessment for clinicians and users.

[0059] Dietary behavior vulnerability dimensions: Based on the feature contribution weights from the model output, the module can identify the dietary-related features that have the most significant impact on the user's pro-inflammatory risk. By performing semantic mapping on these high-contribution features, the system generates explanatory labels, such as "high-sugar driven" (indicating that high sugar intake is a major driver of pro-inflammatory risk), "red meat-dependent" (indicating that high red meat intake is a key factor), "processed food-sensitive" (indicating a particularly strong reaction to processed foods), or "dietary fiber-deficient" (indicating insufficient dietary fiber intake leading to elevated pro-inflammatory risk). These dimensions directly point out the weak links in the user's dietary behavior that need the most improvement.

[0060] Nutrition intervention sensitive windows: Combining the state transition trigger conditions of the behavior pattern recognition module and the fluctuation periods of clinical indicators, the system predicts the time periods when the user is most likely to accept or most in need of nutrition intervention. For example, if the system identifies that the user tends to have high pro-inflammatory dietary behavior when under stress, it will prompt that this is a sensitive intervention window when the user's stress level rises.

[0061] Personalized dietary adjustment recommendations: Based on the pro-inflammatory risk level, dietary behavior vulnerability dimensions, and intervention sensitive windows, the system generates specific and executable dietary adjustment recommendations. These recommendations can be macroscopic (e.g., "increase vegetable and fruit intake") or microscopic (e.g., "replace the main course at dinner with brown rice and reduce red meat intake"), or even precise to specific food recommendations and avoidance. The recommendation content will be filtered and optimized according to the user's historical preferences, allergies, and clinical contraindications.

[0062] Image update trigger mechanism: To ensure the real-time and accuracy of the user's image, the dynamic image generation module is configured with an automatic update trigger mechanism. This mechanism will automatically trigger incremental updates of the user's image when one of the following two conditions occurs:

[0063] Newly collected dietary data causes the 7-day moving average of the individual's pro-inflammatory load index to change by more than 15%. This means that the user's overall dietary habits or pro-inflammatory level have changed significantly, requiring re-evaluation.

[0064] The behavior pattern recognition module detects new significant behavior patterns. For example, the user changes from a persistent low pro-inflammatory load pattern to a high pro-inflammatory load periodic burst pattern.

[0065] When the trigger condition is met, the system will automatically start a new image generation cycle and record the image version change log to track the evolution history of the user's image and the intervention effect.

[0066] Intervention feedback loop module: The intervention feedback loop module is the key to the system's continuous optimization and personalized learning. It is responsible for receiving user feedback data on the implementation of personalized dietary adjustment recommendations and feeding these data back to the multi-dimensional feature fusion module for dynamic calibration of the intervention-sensitive window period parameters in the user profile.

[0067] Feedback data collection: This module actively or passively collects user feedback data through mobile terminal applications or associated smart health devices. These data include:

[0068] Adoption rate of recommendations: The proportion of system recommendations that users actually adopt, for example, 3 out of 5 recommended adjustments are implemented.

[0069] Actual intake deviation: The difference between the user's actual dietary intake and the system's recommendations, for example, the user actually consumes 80 grams of protein instead of the system's recommended 100 grams.

[0070] Subjective feeling score: User subjective score on their physical feelings (e.g., mental state, gastrointestinal comfort, improvement of inflammation symptoms) after adopting recommendations, usually in the form of Likert scale.

[0071] Feedback data feedback and parameter calibration: After cleaning and standardization, the received feedback data is fed back to the multi-dimensional feature fusion module. In the multi-dimensional feature fusion module, these feedback data are considered as a special "context feature" and participate in the feature fusion process together with the original clinical, genetic, and microbiome data. In this way, the system can learn the user's response patterns to different intervention recommendations and the actual effects of different intervention strategies.

[0072] Dynamic calibration of intervention-sensitive window period parameters: Feedback data are particularly important for calibrating intervention-sensitive window period parameters in the user profile. For example, if the system makes recommendations during a "high pressure" period, but the user feedback shows low adoption rate and poor subjective feelings, the system will learn that the user's intervention sensitivity is low in this specific situation. Conversely, if a recommendation has a high adoption rate and good results within a specific time period, the system will enhance the recognition weight of that time period as a sensitive window period. This dynamic calibration process can be achieved through reinforcement learning algorithms or adaptive Bayesian updates, enabling the system to more accurately predict when intervention can achieve the best results based on the user's real behavior and physiological responses, thereby improving the effectiveness and compliance of intervention.

[0073] Cloud-edge collaborative computing architecture: The system adopts a cloud-edge collaborative computing architecture to balance the needs of data privacy, real-time processing capability, and high-performance computing resources.

[0074] The dietary data collection module is deployed on user terminal devices: mobile terminals and wearable devices directly deploy the client of the dietary data collection module, which is responsible for the generation, preliminary verification, local encryption, and secure upload of raw data. This maximizes the privacy of users' personal sensitive data, as the raw data is encrypted before leaving the user's device, and users have more control over their data.

[0075] The pro-inflammatory index calculation module and the behavior pattern recognition module are deployed on edge computing nodes: edge computing nodes can be home smart gateways, community health station servers, or small regional data centers. These two modules are deployed on edge nodes, which have the following advantages:

[0076] Data privacy protection: the calculation of pro-inflammatory index and behavior pattern involves users' detailed dietary information, which can reduce the transmission of raw sensitive data on the wide area network and reduce the risk of leakage.

[0077] Real-time: edge computing can significantly reduce the delay of data transmission to the cloud, supporting quasi-real-time pro-inflammatory load evaluation and behavior pattern analysis, thereby achieving more timely health feedback and early warning.

[0078] Network bandwidth optimization: after a large amount of raw dietary data is calculated on the edge, only the high-level pro-inflammatory index and behavior pattern summary needs to be transmitted to the cloud, effectively saving network bandwidth resources.

[0079] The multi-dimensional feature fusion module and the dynamic portrait generation module are deployed on the cloud server: the cloud server has strong computing resources, storage capacity, and elasticity, which can support:

[0080] Complex model inference: the graph neural network in the multi-dimensional feature fusion module and the attention mechanism gradient boosting tree model in the dynamic portrait generation module require a large amount of computing resources for model training, inference, and continuous optimization, which can be efficiently completed on the cloud high-performance computing cluster.

[0081] Cross-domain data integration: clinical, genetic, and microbiome heterogeneous data are usually stored in different centralized databases, and the cloud can easily integrate and manage these data.

[0082] Model updating and management: centralized in the cloud can easily manage model versions, global optimization, and secure deployment, ensuring that all users can enjoy the latest and most accurate portrait services.

[0083] The communication between the edge node and the cloud server uses encrypted application programming interface and message queue service to ensure the security and reliability of data transmission. This hierarchical architecture effectively balances data security, computing efficiency, and system scalability.

[0084] The embodiment builds a closed-loop intelligent user portrait generation system through the cooperation of the above modules. It no longer relies on traditional static questionnaires or single nutrient analysis, but breaks through the limitations of traditional evaluation by quantifying the pro-inflammatory load of individual diets and identifying their dynamic behavior patterns. The system innovatively introduces multi-source heterogeneous health data for cross-domain fusion, so that the generated user portrait not only reflects the diet behavior itself, but also embeds the individual biological background and lifestyle context, significantly improving the clinical relevance and intervention guidance value of the portrait. The dynamic updating mechanism and intervention feedback closed loop design make the user portrait have the ability to continuously evolve, and can truly reflect the dynamic process of user behavior change and physiological response, providing a quantifiable, traceable and interpretable decision support tool for precise nutrition intervention of chronic inflammation-related diseases. In addition, the system outputs the diet behavior vulnerability dimension through an interpretable machine learning model, so that doctors or nutritionists can quickly understand the core problems of patients and develop targeted strategies, greatly improving the efficiency of doctor-patient communication and intervention compliance.

[0085] Compared with the prior art, the core progress of the present application in technology is to convert the individual dietary pro-inflammatory potential from an abstract concept into a dynamic and quantitative indicator, and reveal its behavior pattern through deep time series analysis. Secondly, through the graph neural network, the seamless fusion of multi-source heterogeneous health data is realized, solving the problem that traditional methods are difficult to integrate biological, clinical and behavior data. Thirdly, the interpretable machine learning model is introduced, so that the user portrait not only has prediction ability, but also provides semantic explanation of behavior vulnerability, greatly enhancing the clinical application value. Finally, through the dynamic updating and feedback closed loop mechanism, the whole system can continuously learn and self-optimize, adapt to the dynamic changes of individual health status, and form a precise health management platform that can continuously evolve.

[0086] Embodiment 2

[0087] On the basis of the foregoing embodiments, the embodiment further elaborates the construction process, maintenance mechanism of the dietary pro-inflammatory potential database and the interaction details with the pro-inflammatory index calculation module, and specifically describes the determination method of the food processing method correction coefficient.

[0088] The construction of the dietary pro-inflammatory potential database is the basis for the system to accurately quantify the pro-inflammatory load of individuals. The construction of the database is a multi-stage, interdisciplinary complex process, aiming to ensure that the pro-inflammatory score associated with each food item has a high degree of scientific basis and credibility.

[0089] Data source integration and preprocessing: the construction of the database starts from large-scale data collection and integration. The main data sources include:

[0090] Large-scale epidemiological study data: Collect cohort, case-control, and cross-sectional study data globally on dietary patterns, specific food intake, and their association with chronic inflammation markers (e.g., serum C-reactive protein, interleukin-6, tumor necrosis factor-alpha, etc.). These data are usually in the form of articles, reports, and need systematic literature review, data extraction, and standardization.

[0091] In vitro cytokine release experiment data: Integrate laboratory results from cell culture models testing the effects of different food extracts, nutrients, or their metabolites on immune cell (e.g., macrophages, lymphocytes) inflammatory factor (e.g., interleukin-1 beta, tumor necrosis factor-alpha) release. These data provide biological evidence of the direct impact of food on inflammation pathways.

[0092] Clinical intervention trial data: Collect data from randomized controlled trials on the effects of specific dietary interventions (e.g., Mediterranean diet, high-fat diet) on human inflammation marker levels. These data directly reflect the real effects of food in the human body.

[0093] All collected data undergo strict quality control and preprocessing before integration, including removing duplicate data, handling missing values, standardizing measurement units, unifying disease diagnosis standards, and performing effect size calculation and meta-analysis on different study results.

[0094] Pro-inflammatory score quantification model: Based on data integration, a multi-factor comprehensive evaluation model is used to quantify the pro-inflammatory score of each food. This model combines the strength of epidemiological associations, in vitro experimental evidence, and clinical trial results with weighted fusion.

[0095] Epidemiological weight: Based on statistical methods (e.g., regression analysis, risk ratio calculation), assess the strength and statistical significance of the correlation between food intake and inflammation marker levels. Foods with stronger and more consistent correlations are given higher weights.

[0096] In vitro experimental evidence strength: Score the direct pro-inflammatory or anti-inflammatory effects of food based on the observed changes in inflammatory factor release and their dose-effect relationships in in vitro experiments.

[0097] Clinical evidence level: Assign corresponding weights based on the evidence level of clinical intervention trials (e.g., randomized controlled trial evidence is better than observational studies) and effect size.

[0098] Finally, the standardized pro-inflammatory score of each food Calculated by the following formula:

[0099]

[0100] wherein, a weight coefficient representing each piece of evidence, whose sum is 1; a standardized effect value representing each piece of evidence. Effect values are uniformly mapped to a score interval of -10 to +10. -10 represents the strongest anti-inflammatory effect, and +10 represents the strongest pro-inflammatory effect.

[0101] Database Structure and Maintenance: The dietary pro-inflammatory potential database adopts a relational database structure, with core tables including:

[0102] Food Ontology Table: Contains food unique identifier, food name (Chinese and English), classification, common cooking methods, main nutritional components, etc.

[0103] Pro-inflammatory Score Table: Stores food unique identifier, standardized pro-inflammatory score, literature citation for the basis of the score, update date, etc.

[0104] Correction Coefficient Table: Stores food classification, processing method, corresponding correction coefficient value, source, etc.

[0105] Database maintenance is an ongoing process. New scientific research results and data will be regularly imported into the database and verified through an expert review mechanism. Once new research changes the pro-inflammatory score or correction coefficient of a specific food, the database will be incrementally updated, and version change logs will be recorded to ensure that all calculations are based on the latest scientific evidence.

[0106] Determination of Food Processing Method Correction Coefficient

[0107] The determination of food processing method correction coefficient is an important part of the precision of the pro-inflammatory index calculation module, and its purpose is to reflect the influence of different processing methods on the intrinsic pro-inflammatory or anti-inflammatory characteristics of food. The value range of the correction coefficient is 0.5 to 2.0.

[0108] Experimental data acquisition: Correction coefficients are mainly determined by the following methods:

[0109] Nutrition experiments: After the same batch of food is processed in different ways (such as raw, boiled, steamed, baked, fried, pickled), analyze the changes in its nutritional components (such as fat oxidation products, final glycation products, trans fatty acids, antioxidants). These changes are closely related to the intensity of inflammatory response.

[0110] Bioactivity detection: Extract food extracts processed in different ways and conduct in vitro cytokine release experiments again to directly measure their impact on inflammatory response.

[0111] Expert evaluation and consensus: For processing methods that lack direct experimental data, a committee composed of nutrition, food science, and inflammation immunology experts evaluates and determines preliminary correction coefficient values based on existing knowledge and experience.

[0112] Iterative optimization and validation: The preliminary set of correction factors will be validated through retrospective analysis of historical user data. For example, observe whether the actual change in inflammatory marker levels in users after intake of a particular processing method is consistent with the prediction of the corrected pro-inflammatory score. If there is a significant deviation, the correction factors will be iteratively adjusted and optimized.

[0113] For example, frying and grilling usually generate more advanced glycation end products (AGEs) and oxidized fats, which have strong pro-inflammatory effects, so their correction factors are usually greater than 1. Boiling and steaming usually retain the nutritional components of food and reduce the generation of harmful substances, so their correction factors may be close to 1 or slightly less than 1. Cured food may have pro-inflammatory potential due to high salt content and specific additives, and the correction factor may be slightly higher than 1.

[0114] The pro-inflammatory index calculation module can generate highly detailed and personalized individual pro-inflammatory load index time series by accurately matching dietary data, calling the latest version of the pro-inflammatory potential database, and combining intake weight and processing correction. This sequence not only reflects the pro-inflammatory effects of the user's dietary behavior, but also captures its dynamic changes and potential risks, providing a solid data foundation for subsequent behavior pattern recognition and user portrait generation.

[0115] Embodiment 3

[0116] This embodiment is based on the previous embodiments and further elaborates on the working principle of the interpretable machine learning model, the generation logic of the dietary behavior vulnerability dimensions, and the specific implementation details of the portrait update triggering mechanism in the dynamic portrait generation module. It also provides detailed instructions on the generation and optimization process of personalized dietary adjustment recommendations.

[0117] Working principle of interpretable machine learning model: The dynamic portrait generation module uses a gradient boosting tree ensemble model based on attention mechanism. Its core advantage is that it can clearly reveal the basis for decision-making while providing high-precision predictions.

[0118] Gradient Boosting Tree Ensemble: The model is composed of multiple weak learners (usually decision trees) in series. Each tree learns from the residual of the previous tree, gradually optimizing the prediction performance. This model can handle high-dimensional heterogeneous feature data and automatically capture non-linear relationships and interactions between features.

[0119] Attention mechanism and feature contribution weight: In this model, the attention mechanism is not a traditional self-attention layer in deep learning, but refers to the evaluation of the contribution of each input feature to the final prediction result (e.g., pro-inflammatory risk level) by methods such as SHAP (Shapley Additive explanations) or LIME (Local Interpretable Model-agnostic Explanations).

[0120] SHAP value calculation: SHAP values are based on cooperative game theory, which fairly distributes the contribution of each feature to each prediction. For each user, SHAP values can quantify how each input feature (e.g., "high sugar intake frequency", "specific genotype of gene locus XX", "relative abundance of pro-inflammatory bacteria") affects the model's output, and these impact values are added up, equal to the difference between the predicted value and the baseline value (the predicted value when all features take the average value). A positive SHAP value indicates that the feature pushes the predicted value up, and a negative value indicates that it pushes it down. By calculating the SHAP values of all features, we can get the exact contribution weight of each feature in the user profile to the pro-inflammatory risk.

[0121] Generation logic of dietary behavior vulnerability dimensions: Based on the feature contribution weight of the model output, the system can further generate explanatory dietary behavior vulnerability dimensions.

[0122] Feature contribution ranking: For each user, the system will sort the features in descending order according to the absolute value of their SHAP values, and find the top N features (e.g., N is set to 5 or 10) that have the greatest impact on the user's pro-inflammatory risk.

[0123] Feature semantic mapping: Map these high-contribution features to predefined dietary behavior vulnerability dimensions. This mapping process is based on nutritional knowledge and expert experience. For example:

[0124] If the SHAP value of the "high sugar intake frequency" feature is the highest and positive, the user may be labeled as "high sugar driven".

[0125] If the SHAP value of the "red meat intake" feature is significantly positive, it may be labeled as "red meat dependent".

[0126] If the SHAP value of the "processed food intake proportion" feature is high and positive, it may be labeled as "processed food sensitive".

[0127] If the SHAP value of the "dietary fiber intake" feature is significantly negative (indicating that a lack of dietary fiber increases the risk), it may be labeled as "dietary fiber deficient".

[0128] If the SHAP value of the "vegetable and fruit intake frequency" feature is significantly negative, it can be labeled as "insufficient vegetable and fruit intake type".

[0129] These dimensions provide highly semantic explanations, helping doctors, nutritionists, and users quickly understand the root causes of pro-inflammatory risks. A user can have multiple vulnerability dimensions simultaneously, reflecting the complexity of their dietary behaviors.

[0130] Implementation details of image update trigger mechanism

[0131] The image update trigger mechanism ensures that the user image can timely reflect the user's latest health status and behavior changes.

[0132] 7-day moving average change detection of individual pro-inflammatory load index: The system continuously calculates the 7-day moving average of the individual pro-inflammatory load index for each user. Every day, the system compares the current 7-day moving average with the previous day's 7-day moving average. If the change amplitude (absolute value) exceeds 15%, i.e. , the image update is triggered. This moving average-based detection method can smooth out short-term fluctuations and focus on long-term trends in dietary pro-inflammatory potential.

[0133] Behavior pattern recognition module detects new significant behavior patterns: When the behavior pattern recognition module detects that a user switches from one main pro-inflammatory state (e.g., persistent low pro-inflammatory load) to another main pro-inflammatory state (e.g., periodic bursts of high pro-inflammatory load) through the hidden Markov model, or detects a new, previously unobserved stable state pattern, the image update is triggered. This is usually done by comparing the newly identified pattern with the set of patterns recorded in the user's historical image, and if there is a significant difference and passes the statistical significance test, the update is triggered.

[0134] Logging and notification: Each time the image is updated, the system automatically records a detailed image version change log, including update time, trigger reason, key differences between old and new image versions. At the same time, the system sends notifications to relevant medical staff and users, reminding them that the user image has been updated and suggesting viewing the latest health report and recommendations.

[0135] Generation and optimization of personalized dietary adjustment recommendations: The generation of personalized dietary adjustment recommendations is an intelligent recommendation process based on the multi-dimensional information of the user image, and is continuously optimized through the intervention feedback loop.

[0136] Suggestion library and rule engine: The system maintains a large library of dietary adjustment suggestions, including various food substitutions, cooking method changes, mealtime adjustments, and other strategies. Each suggestion is associated with specific pro-inflammatory risk factors and dietary behavior vulnerability dimensions. A rule-based expert system filters a preliminary set of suggestions from the library based on the user's pro-inflammatory risk level, primary vulnerability dimensions, and intervention-sensitive windows.

[0137] Constraint and preference filtering: Based on the preliminary suggestion set, the system further filters and optimizes suggestions considering user's individualized information:

[0138] Clinical contraindications: For example, high-protein diet suggestions are filtered out for patients with renal dysfunction.

[0139] Allergy history: Foods known to be allergenic to the user are filtered out.

[0140] Dietary preferences: By analyzing user historical dietary data or questionnaires, the system understands user's taste preferences and dietary habits, and prioritizes food and cooking methods that are more likely to be accepted by the user.

[0141] Feasibility assessment: Considering user's geographical location, seasonal factors, and purchasing power, the system recommends foods that are available and affordable.

[0142] Multi-objective optimization and reinforcement learning: To generate optimal personalized suggestions, the system uses multi-objective optimization algorithms to balance the goals of reducing pro-inflammatory load, improving user adherence, and ensuring nutritional balance. User adoption rate and actual effect data provided by the intervention feedback closed-loop module are used to train the reinforcement learning model. Through continuous trial and error and learning, the reinforcement learning model optimizes the suggestion generation strategy, enabling the system to dynamically adjust its recommendation algorithm based on user's individualized response, thereby maximizing the long-term effectiveness of the intervention. For example, if a user has high adherence and significant effect on the suggestion to "reduce sugar intake", the system will continue to reinforce this type of suggestion in the future; if a suggestion is theoretically effective but user adherence is consistently low, the system will learn to provide alternative, more easily adopted solutions for that user.

[0143] This embodiment demonstrates how the system uses interpretable machine learning techniques to create unique and insightful health portraits for each user in a complex feature space through a detailed description of the dynamic portrait generation module. These portraits not only provide intuitive risk assessments, but also further reveal deep behavioral drivers and can be continuously optimized based on user feedback, providing unprecedented intelligent support for precision nutrition intervention for chronic inflammation-related diseases, and achieving seamless conversion from data to executable and optimized personalized health management solutions.

Claims

1. A user profiling system based on analysis of patient pro-inflammatory dietary behaviors, characterized in that, The method comprises a dietary data collection module, a pro-inflammatory index calculation module, a behavior pattern recognition module, a multi-dimensional feature fusion module, and a dynamic portrait generation module. The dietary data collection module is used to continuously acquire dietary intake records of a user through a mobile terminal or a wearable device, and the dietary intake records include food categories, intake amounts, intake times, intake frequencies, and accompanying physiological state data. The pro-inflammatory index calculation module is used to call a pre-constructed dietary pro-inflammatory potential database based on the dietary intake records, perform pro-inflammatory score mapping on each food item, and perform weighted calculation in combination with intake amounts to generate a time-sequenced individual pro-inflammatory load index. The behavior pattern recognition module is used to perform time-series pattern mining on the time-sequenced individual pro-inflammatory load index to identify dietary behavior patterns with statistical significance. The multi-dimensional feature fusion module is used to perform cross-domain alignment and feature embedding of the dietary behavior patterns and clinical indicator data, genetic polymorphism information, intestinal microbiome characteristics, and lifestyle parameters of a user to construct a high-dimensional heterogeneous feature vector. The dynamic portrait generation module is used to generate a structured user portrait based on the high-dimensional heterogeneous feature vector through an interpretable machine learning model, and the user portrait includes a pro-inflammatory risk level, a dietary behavior vulnerability dimension, a nutritional intervention sensitive window period, and individualized dietary adjustment suggestions. The behavior pattern recognition module uses a sliding window combined with a hidden Markov model to divide the individual pro-inflammatory load index sequence into states, identifies low, medium, and high pro-inflammatory states, and decodes the most likely state transition path through a Viterbi algorithm to extract state residence time, state transition frequency, and state trigger conditions as core parameters of the behavior pattern.

2. The patient pro-inflammatory dietary behavior analysis based user profiling system of claim 1, wherein, The dietary pro-inflammatory potential database is constructed based on large-scale epidemiological research and in vitro cytokine release experiment data, and each food item is associated with a standardized pro-inflammatory score that comprehensively reflects the potential influence of the food on serum C-reactive protein, interleukin 6, and tumor necrosis factor alpha levels after intake, with a score range of -10 to 10, where negative values represent anti-inflammatory effects and positive values represent pro-inflammatory effects.

3. The patient pro-inflammatory dietary behavior analysis based user profiling system of claim 1, wherein, The pro-inflammatory index calculation module uses the logarithmic transformation value of the intake amount as a weight factor and introduces a food processing method correction coefficient when performing weighted calculation, which quantitatively adjusts the amplification or inhibition effect of pro-inflammatory potential according to different processing methods such as frying, grilling, pickling, or raw food, with a correction coefficient value range of 0.5 to 2.

0.

4. The patient pro-inflammatory dietary behavior analysis based user profiling system of claim 1, wherein, The multi-dimensional feature fusion module uses a graph neural network architecture to realize cross-domain feature alignment, where nodes represent feature entities from different data sources, and edges represent biological associations or statistical correlations between features. Through a multi-layer message passing mechanism, neighbor information is aggregated to generate a unified embedding vector for each user, with a dimension of 128.

5. The patient pro-inflammatory dietary behavior analysis based user profiling system of claim 1, wherein, The interpretable machine learning model adopted by the dynamic portrait generation module is a gradient boosting tree ensemble model based on attention mechanism, which outputs the contribution weight of each input feature to the prediction result while predicting the pro-inflammatory risk level, and the contribution weight is used to generate the explanatory label of the dietary behavior vulnerability dimension.

6. The patient pro-inflammatory dietary behavior analysis based user profiling system of claim 1, wherein, The dynamic portrait generation module is also configured with a portrait update triggering mechanism, which automatically triggers incremental update of the user portrait and records the portrait version change log when the newly collected dietary data causes the change amplitude of the 7-day moving average of the individual pro-inflammatory load index to exceed 15%, or the behavior pattern recognition module detects a new significant behavior pattern.

7. The patient pro-inflammatory dietary behavior analysis based user profiling system of claim 1, wherein, An intervention feedback closed loop module is also included, which is used to receive user execution feedback data on personalized dietary adjustment suggestions, including suggestion adoption rate, actual intake deviation and subjective feeling score, and inject the execution feedback data back to the multi-dimensional feature fusion module for dynamic calibration of the intervention sensitive window period parameters in the user portrait.

8. The patient pro-inflammatory dietary behavior analysis based user profiling system of claim 1, wherein, The clinical indicator data includes fasting blood glucose, glycosylated hemoglobin, blood lipid profile, liver function indicators and systemic inflammation markers; the genetic polymorphism information includes single nucleotide polymorphism sites related to inflammation pathways, covering TNF, IL6, CRP and PPARG genes; the intestinal microbiome features include the relative abundance ratio of pro-inflammatory and anti-inflammatory bacterial genera; and the lifestyle parameters include sleep duration, physical activity intensity, stress level and smoking and drinking status.

9. The patient pro-inflammatory dietary behavior analysis based user profiling system of claim 1, wherein, The system runs in a cloud-edge collaborative computing architecture, the dietary data collection module is deployed on the user terminal device, the pro-inflammatory index calculation module and the behavior pattern recognition module are deployed on the edge computing node, and the multi-dimensional feature fusion module and the dynamic portrait generation module are deployed on the cloud server.

Images