CN-122025021-A - Intelligent food material evaluation method, system and storage medium for chronic disease population

CN122025021ACN 122025021 ACN122025021 ACN 122025021ACN-122025021-A

Abstract

The invention discloses an intelligent food material evaluation method, system and storage medium for chronic disease population. The method responds to a food material evaluation request of a user, acquires static physiological data, historical time sequence monitoring data and unstructured historical diet log text of the user, and constructs a multi-source heterogeneous health historical data set through processing. An intelligent evaluation model consisting of a data fusion layer, a portrait construction layer and a causal inference evaluation layer is constructed, and the model can fuse multi-mode time sequence data into feature vectors and construct and incrementally update the dynamic health portrait of the user based on the feature vectors. After the model is trained by combining pre-training and fine-tuning, the system can generate an evaluation result comprising a health score, a personalized suggestion and a predicted physiological influence curve through causal inference prediction according to the current health portrait of the user and food material information to be evaluated. The invention realizes dynamic individuation accurate evaluation, and continuously improves the suitability of suggestions and the long-term efficacy of healthy intervention through a closed-loop optimization mechanism.

Inventors

YANG WENJUN
WEN JIANQUAN
YU MIN
LONG HAI
HU BOFAN
XIAO GANG
LI JINNUO
DING DEWAN

Assignees

湖南创星人工智能研究院有限公司

Dates

Publication Date: 20260512
Application Date: 20260205

Claims (10)

1. An intelligent food material evaluation method for chronic disease groups is characterized by comprising the following steps: responding to a food material evaluation request of a user, and acquiring static physiological data, historical time sequence monitoring data and unstructured historical diet log text of the user; Processing the historical time sequence monitoring data and the unstructured historical diet log text respectively, and combining the static physiological data to construct a multi-source heterogeneous health historical data set; An intelligent evaluation model consisting of a data fusion layer, a portrayal construction layer and a causal inference evaluation layer is constructed, wherein the data fusion layer is used for uniformly mapping multi-mode time sequence data in the multi-source heterogeneous health historical data set and generating a historical fusion feature vector, the portrayal construction layer is used for constructing and continuously updating a dynamic health portrayal of a user by adopting an incremental learning mechanism based on the historical fusion feature vector, the causal inference evaluation layer is an individualized diet reaction prediction model based on a multi-mode large language model and is used for generating an output token sequence containing at least one of food scoring, health suggestion and predicted physiological influence curve according to the dynamic health portrayal and input food information to be evaluated; training the intelligent evaluation model by adopting a mode of combining pre-training and fine tuning based on the multi-source heterogeneous health historical data set until the model converges to obtain an optimized intelligent evaluation model; Inputting food material information to be evaluated and current health data of a user into the optimized intelligent evaluation model, generating an output token sequence through the causal inference evaluation layer, and analyzing the output token sequence to obtain at least one evaluation result of health score, personalized suggestion and predicted physiological influence curve for the food material.
2. The intelligent food material assessment method according to claim 1, wherein said constructing a multi-source heterogeneous health history data set comprises the steps of: Carrying out named entity identification on the unstructured historical diet log text, extracting food entities, cooking modes and subjective feeling entities, and converting fuzzy component description in the text into estimated eating weight values based on a pre-trained component mapping model to obtain structured historical dining information; Performing outlier cleaning, smooth filtering and resampling according to a preset frequency on the historical time sequence monitoring data to obtain regular historical time sequence monitoring data; Acquiring basic nutrition component data corresponding to the food material entity through a standard nutrition knowledge base interface; And integrating the static physiological data, the structured historical meal information, the regular historical time sequence monitoring data and the basic nutrition component data to generate the multi-source heterogeneous health historical data set.
3. The intelligent food material assessment method according to claim 2, wherein the data fusion layer generates a history fusion feature vector, comprising the steps of: Establishing a front time window and a rear time window of preset duration by taking the time point of each dining event as a reference; Time aligning and correlating the structured historical meal information, the regular historical time series monitoring data, and the basal nutritional ingredient data within the time window; mapping the aligned multi-mode historical data into feature vectors respectively, and carrying out normalization processing on the feature vectors; And splicing the normalized feature vectors to generate the historical fusion feature vector.
4. The intelligent food material assessment method according to claim 3, wherein said representation construction layer constructs and updates said dynamic health representation, comprising the steps of: Constructing the dynamic health portrait comprising a static layer and a dynamic layer based on the history fusion feature vector, wherein the static layer is used for storing long-term stability information in the static physiological data; The dynamic layer extracts historical metabolic trend characteristics of key physiological indexes from the regular historical time sequence monitoring data based on a time sequence analysis model; generating a historical diet preference vector by performing embedded learning on the structured historical dining information; fitting personalized influence coefficients of the basic nutrition component data on the key physiological indexes based on a regression model; And updating the historical metabolic trend characteristics, the historical diet preference vector and the personalized influence coefficient in the dynamic layer after receiving new user health data each time by adopting an online learning or incremental learning mode.
5. The intelligent food material assessment method according to claim 1, wherein the causal inference assessment layer comprises: the feature coding sub-layer is used for coding the features of the dynamic health portrait and the features of the food information to be evaluated into a unified joint vector representation; The causal effect prediction sub-layer is used for carrying out inverse fact reasoning on the joint vector representation based on the multi-modal large language model and outputting individualized physiological influence prediction; An output generation sublayer for converting the physiological impact predictions into a structured token sequence comprising readable scores, specific suggestions and prediction curves.
6. The intelligent food material assessment method according to claim 1, wherein said employing a combination of pre-training and fine-tuning comprises: Pre-training the intelligent assessment model on a large-scale public health dataset to learn a general health and nutrition model representation; based on the multi-source heterogeneous health history data set, performing supervised fine tuning on the pre-trained model to optimize evaluation performance for individual users; and adopting a reinforcement learning strategy to further optimize the fine-tuned model by combining feedback data of the historical evaluation result by the user.
7. The intelligent food material assessment method according to claim 1, further comprising a post-processing step prior to parsing said output token sequence: based on a preset user disease tabu rule base, carrying out security check on an evaluation result corresponding to the output token sequence; combining the accumulated nutrient intake data of the user on the same day, and carrying out nutrition rationality verification on the assessment result; if any check fails, triggering a multi-objective optimization algorithm to regenerate an evaluation result.
8. An intelligent food material evaluation system for chronic disease population, comprising: The historical data processing module is used for responding to the input information of the user, receiving the static physiological data of the user and respectively processing the historical time sequence monitoring data and the unstructured historical diet log text so as to construct a multi-source heterogeneous health historical data set; The model construction module is used for constructing an intelligent evaluation model consisting of a data fusion layer, a portrait construction layer and a causal inference evaluation layer, wherein the data fusion layer is used for uniformly mapping multi-mode time sequence data in the multi-source heterogeneous health historical data set and generating a historical fusion feature vector, the portrait construction layer is used for constructing and continuously updating a dynamic health portrait of a user by adopting an incremental learning mechanism based on the historical fusion feature vector, and the causal inference evaluation layer is an individualized diet reaction prediction model based on a multi-mode large language model and is used for generating an output token sequence containing at least one of food scoring, health suggestion and predicted physiological influence curve according to the dynamic health portrait and the input food information; The model optimization module is used for training the intelligent evaluation model in a mode of combining pre-training and fine tuning based on the multi-source heterogeneous health historical data set until the model converges to obtain an optimized intelligent evaluation model; The evaluation result output module is used for inputting the user input information to be evaluated into the optimized intelligent evaluation model, generating an output token sequence through the causal inference evaluation layer, and analyzing the output token sequence to obtain at least one evaluation result of food health score, personalized suggestion and predicted physiological influence curve.
9. The system as recited in claim 8, further comprising: the user interaction and feedback module is used for presenting the evaluation result to a user and collecting feedback data of the user on the evaluation result; And the closed-loop optimization module is used for updating the dynamic health portrait and optimizing the intelligent evaluation model according to the feedback data of the user on the evaluation result.
10. A computer-readable storage medium, on which a computer program is stored, characterized in that the intelligent food material evaluation method according to any one of claims 1 to 7 is implemented, when the program is executed by a processor.

Description

Intelligent food material evaluation method, system and storage medium for chronic disease population Technical Field The application relates to the technical field of health management and artificial intelligence, in particular to an intelligent food material evaluation method, system and storage medium for chronic disease population. Background Diet management of chronic diseases (e.g., diabetes, hypertension, obesity, etc.) is a key element in controlling the condition and delaying complications. Traditional diet management methods rely mainly on general dietary guidelines or doctor experience advice, with the following technical drawbacks: First, the data dimension is single, and the individuation degree is low. The existing system provides a 'one-cut' suggestion based on limited static information (such as age, BMI and disease type) of a user, and the suggestion lacks real individual adaptability due to the fact that the existing system cannot effectively integrate multi-source heterogeneous data such as dynamic physiological indexes (such as continuous blood sugar monitoring data) of the user, unstructured diet logs and subjective somatosensory feedback. Second, the proposed mechanism is stiff and lacks dynamic optimization capabilities. The traditional method is mainly based on an open-loop system with fixed rules, the generated diet advice is not changed once given, and the diet advice cannot be dynamically adjusted and optimized according to the execution feedback and health condition evolution of a user, so that the long-term service effect is difficult to maintain. Third, the evaluation granularity is coarse, and the accurate intervention is insufficient. The prior art stays at the macro nutrient control or food material type filtering level, cannot deeply quantify the micro dynamic influence of specific food materials and components thereof on individual key physiological indexes, and cannot meet the accurate management requirements of high-risk chronic patients on diet safety and effectiveness. Therefore, providing an intelligent food material evaluation method, system and storage medium for chronic disease group to solve the above-mentioned problems is a urgent problem for those skilled in the art. Disclosure of Invention The invention aims to overcome the defects of the prior art, and provides an intelligent food material evaluation method, system, storage medium and electronic equipment for chronic disease groups, so as to solve the technical problems of low individuation level and poor long-term management efficiency caused by insufficient data fusion, static stiffness evaluation and coarse intervention granularity in the prior art. Based on the above objects, in a first aspect, the present invention provides an intelligent food material evaluation method for chronic disease population, comprising the following steps: responding to a food material evaluation request of a user, and acquiring static physiological data, historical time sequence monitoring data and unstructured historical diet log text of the user; Processing the historical time sequence monitoring data and the unstructured historical diet log text respectively, and combining the static physiological data to construct a multi-source heterogeneous health historical data set; An intelligent evaluation model consisting of a data fusion layer, a portrayal construction layer and a causal inference evaluation layer is constructed, wherein the data fusion layer is used for uniformly mapping multi-mode time sequence data in the multi-source heterogeneous health historical data set and generating a historical fusion feature vector, the portrayal construction layer is used for constructing and continuously updating a dynamic health portrayal of a user by adopting an incremental learning mechanism based on the historical fusion feature vector, the causal inference evaluation layer is an individualized diet reaction prediction model based on a multi-mode large language model and is used for generating an output token sequence containing at least one of food scoring, health suggestion and predicted physiological influence curve according to the dynamic health portrayal and input food information to be evaluated; training the intelligent evaluation model by adopting a mode of combining pre-training and fine tuning based on the multi-source heterogeneous health historical data set until the model converges to obtain an optimized intelligent evaluation model; Inputting food material information to be evaluated and current health data of a user into the optimized intelligent evaluation model, generating an output token sequence through the causal inference evaluation layer, and analyzing the output token sequence to obtain at least one evaluation result of health score, personalized suggestion and predicted physiological influence curve for the food material. Preferably, the constructing the multi-source heterogeneous health history data set in