Intelligent health risk assessment method and system

Abstract

The invention provides an intelligent health risk assessment method and system, and relates to the technical field of health risk assessment, health big data is constructed, feature selection is performed through the health big data to construct a gradient boosting decision tree model, risk prediction of individual diseases is performed, and time sequence analysis of health indexes is performed according to a first health risk score. Adjusting the first health risk score, carrying out health risk interpretation based on the second health risk score, carrying out health risk hierarchical evaluation according to the personalized health optimization suggestion and the second health risk score, making a hierarchical intervention strategy, carrying out health detection on the target individual, and triggering health early warning according to an abnormal detection result. The technical problems that in the prior art, health risk assessment data sources are single, so that health assessment is not comprehensive and inaccurate, and targeted intervention strategies are lacked are solved. The technical effects of comprehensively and accurately evaluating the individual health risk and providing a targeted intervention strategy are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of health risk assessment, and in particular to a health risk intelligent assessment method and system. Background Art

[0002] In the current medical and health field, personalized health risk assessment is an important part of precision medicine. However, existing technologies often rely on a single data source, such as physical examination reports or questionnaires, ignoring the comprehensive use of multidimensional data on the patient's health status. This single data model is difficult to fully reflect the health status of an individual, resulting in a lack of breadth and depth of data support for risk assessment results. In addition, these methods often ignore the dynamic characteristics of health indicators over time and are unable to adjust risk scores in real time, resulting in deviations between the assessment results and the individual's actual health status.

[0003] The existing technology has the technical problem that the data source of health risk assessment is single, resulting in incomplete and inaccurate health assessment and lack of targeted intervention strategies. Summary of the invention

[0004] The present application provides a health risk intelligent assessment method and system, which is used to solve the technical problems in the prior art that the health risk assessment data source is single, resulting in incomplete and inaccurate health assessment and lack of targeted intervention strategies.

[0005] In view of the above problems, the present application provides a health risk intelligent assessment method and system.

[0006] In a first aspect of the present application, a method for intelligent health risk assessment is provided, the method comprising: collecting multidimensional health data based on natural language to construct health big data; performing feature selection through the health big data to obtain multiple health risk data features, constructing a gradient boosting decision tree model based on the multiple health risk data features, predicting the risk of individual diseases through the gradient boosting decision tree model, and generating a first health risk score; performing time series analysis of health indicators according to the first health risk score to generate time series dynamic change parameters, adjusting the first health risk score according to the time series dynamic change parameters to generate a second health risk score; interpreting health risks based on the second health risk score to generate personalized health optimization suggestions, performing a stratified health risk assessment based on the personalized health optimization suggestions combined with the second health risk score, and formulating a stratified intervention strategy; performing health detection on the target individual according to the stratified intervention strategy, and triggering a health warning based on abnormal detection results.

[0007] According to a second aspect of the present application, a health risk intelligent assessment system is provided, the system comprising: a data acquisition module, the data acquisition module being used to perform multi-dimensional health data acquisition based on natural language and construct health big data; a risk prediction module, the risk prediction module being used to perform feature selection through the health big data to obtain multiple health risk data features, construct a gradient boosting decision tree model based on the multiple health risk data features, perform risk prediction of individual diseases through the gradient boosting decision tree model, and generate a first health risk score; a score adjustment module, the score adjustment module being used to perform time series analysis of health indicators according to the first health risk score, generate time series dynamic change parameters, adjust the first health risk score according to the time series dynamic change parameters, and generate a second health risk score; a risk assessment module, the risk assessment module being used to interpret health risks based on the second health risk score, generate personalized health optimization suggestions, perform health risk stratification assessment based on the personalized health optimization suggestions combined with the second health risk score, and formulate a stratified intervention strategy; a health detection module, the health detection module being used to perform health detection on the target individual according to the stratified intervention strategy, and trigger a health warning based on abnormal detection results.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: The method provided in the embodiment of the present application collects multi-dimensional health data based on natural language, constructs health big data, performs feature selection through health big data, constructs a gradient boosting decision tree model based on multiple health risk data features, predicts the risk of individual diseases, generates a first health risk score, performs time series analysis of health indicators according to the first health risk score, generates time series dynamic change parameters, adjusts the first health risk score, generates a second health risk score, interprets health risks based on the second health risk score, generates personalized health optimization suggestions, performs stratified health risk assessment based on the personalized health optimization suggestions combined with the second health risk score, formulates stratified intervention strategies, performs health detection on target individuals, and triggers health warnings based on abnormal detection results. The technical effect of comprehensively and accurately assessing individual health risks and providing targeted intervention strategies has been achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1A flowchart of a health risk intelligent assessment method provided for this application; Figure 2 A structural diagram of a health risk intelligent assessment system provided in this application.

[0011] Explanation of the accompanying drawings: data collection module 11, risk prediction module 12, score adjustment module 13, risk assessment module 14, health detection module 15. DETAILED DESCRIPTION

[0012] This application provides a health risk intelligent assessment method and system, which is used to solve the technical problems in the prior art that the health risk assessment data source is single, resulting in incomplete and inaccurate health assessment and lack of targeted intervention strategies. It achieves the technical effect of comprehensively and accurately assessing individual health risks and providing targeted intervention strategies.

[0013] Below, the technical solutions in the present invention will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments of the present invention. It should be understood that the present invention is not limited to the example embodiments described herein. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. It should also be noted that, for the convenience of description, only the parts related to the present invention are shown in the accompanying drawings, rather than all of them.

[0014] Embodiment 1, as Figure 1 As shown, the present application provides a health risk intelligent assessment method, the method comprising: Collect multi-dimensional health data based on natural language and build health big data.

[0015] Specifically, natural language processing technology is used to collect multi-dimensional health data of individuals from various medical data sources, such as electronic medical records, social media, health monitoring equipment, etc., to construct health big data, which comprehensively covers health information.

[0016] Furthermore, multi-dimensional health data collection is performed based on natural language to construct health big data, and the method includes: generating multiple character tags by performing optical character recognition on the medical database, classifying and annotating the text data with the multiple character tags, and obtaining multiple medical text identification data; performing voice input according to the collection target to generate multiple voice data, converting the multiple voice data into text to extract multiple text words, and obtaining multiple medical semantic classification data; performing in-clinic dictation according to the doctor-patient conversation record to generate multiple conversation record data, structuring the multiple conversation record data to obtain multiple medical structured health data; performing data cleaning on the multiple medical text identification data, the multiple medical semantic classification data, and the multiple medical structured health data, and performing data standardization and alignment according to the storage architecture based on the data cleaning result to construct the health big data.

[0017] Specifically, first, optical character recognition (OCR) technology is used to comprehensively identify and analyze the medical database, which includes electronic medical records, drug instructions, medical research literature, etc. Optical character recognition technology can automatically identify and convert text in paper documents or images into digital format, thereby generating a series of character tags. The character tags mark the recognized text content, and the character tags are used to classify and characterize the text data. For example, according to medical terms and knowledge points, the text data is accurately classified into diagnostic information, prescription drugs, diagnostic results, etc., and finally multiple medical text identification data are obtained. The medical text identification data is the classification and annotation data of text information, which effectively improves the utilization efficiency and accuracy of medical information. Then, after the target user authorizes and permits, the target user, such as the voice communication between doctors and patients, collects their health voice content, which includes the patient's medical history description or the doctor's health advice. The voice content is converted into text through voice recognition technology, and key text words are extracted from these voice data, and semantic classification is performed, such as symptoms, time nodes, medication habits, etc., thereby generating multiple medical semantic classification data, which are key health information in the classified voice content. Moreover, in the doctor-patient interaction scenario, the doctor-patient conversation content is recorded in real time through the in-clinic dictation technology to generate multiple conversation record data. Using natural language understanding (NLU) technology, multiple unstructured conversation contents such as doctor's advice and patient's answers are converted into structured data, for example, the doctor's medical advice and examination suggestions are mapped into a predefined data format, thereby forming multiple medical structured health data, which is the data for structured storage of doctor-patient interaction content. Finally, the generated medical text identification data, medical semantic classification data and medical structured health data are comprehensively cleaned, and duplicate records are removed using data processing algorithms to ensure the uniqueness and accuracy of the data. At the same time, according to the storage architecture, such as the storage structure of the data table or the field format of data standardization, the cleaned data is aligned and data standardization is performed, and the data from various sources are unified into a consistent structure and integrated to construct health big data, which contains rich text identification information, multi-dimensional semantic classification data and structured health data, providing a data basis for personalized health assessment and risk prediction.

[0018] Feature selection is performed through the health big data to obtain multiple health risk data features, a gradient boosting decision tree model is constructed according to the multiple health risk data features, and the risk of individual diseases is predicted through the gradient boosting decision tree model to generate a first health risk score.

[0019] Specifically, through statistical analysis or machine learning algorithms, feature information related to individual health status or disease risk is identified and extracted from health big data, and multiple health risk data features are obtained, such as age, gender, medical history, living habits, blood pressure, blood sugar level, BMI index and other indicators closely related to health status. After obtaining multiple health risk data features, the multiple features are used to construct a gradient boosting decision tree. The gradient boosting decision tree is an integrated learning method. By gradually training multiple weak learners, each new tree attempts to correct the prediction errors of all previous trees, thereby forming a strong learner that can more accurately predict the target variable, which is the risk of individual disease. Then, the trained gradient boosting decision tree model is used to predict the disease risk of the individual, input multiple health risk data feature values ​​of the individual, and output the disease risk prediction probability or score, that is, the first health risk score. The first health risk score intuitively reflects the possibility of the individual getting sick in the future based on the current health risk data features. The higher the first health risk score, the greater the disease risk faced by the individual, and vice versa. By constructing a gradient boosting decision tree model, it is possible to predict and judge the disease risk of the individual, provide personalized health management suggestions for users, and improve the health management level of users.

[0020] Further, feature selection is performed through the health big data to obtain multiple health risk data features, and a gradient boosting decision tree model is constructed according to the multiple health risk data features. The method includes: using random forest to traverse and train the health big data to obtain multiple initial health risk data features, each of the multiple initial health risk data features has an importance score; performing data dimension reduction on the multiple initial health risk data features according to the multiple importance scores to determine multiple health risk data features; initialization calculation is performed based on the multiple health risk data features to obtain an initial prediction value and set a target value; according to the initial prediction value combined with the target value, the input variables are trained to determine the number of decision trees, and iterate according to the model learning rate based on the number of decision trees to obtain decision tree depth data; according to the decision tree depth data, the input variables are hyperparameter optimized according to grid search to determine multiple parameter combinations, and the multiple parameter combinations are tested and verified. When the multiple parameter combinations tend to converge, the gradient boosting decision tree model is obtained; the risk of individual diseases is predicted by the gradient boosting decision tree model to generate a first health risk score, and the formula is: ; in, Score health risks, is the sample prediction value, and e is a natural constant.

[0021] Specifically, the health big data is processed and the health big data is traversed and trained using the random forest algorithm. Random forest is an ensemble learning algorithm based on multiple decision trees. It randomly extracts multiple data dimensions from the health big data set to form multiple initial health risk data features, such as individual physical examination indicators, health history data, etc. The contribution of multiple initial health risk data features is calculated based on the health big data, and the contribution of each feature in disease risk prediction is evaluated and calculated, thereby generating a set of initial health risk data features with importance scores, where the importance score reflects the intensity of the impact of each feature on disease risk. The contribution calculation formula is as follows: ;in, Features The feature score of is the total number of decision trees, Features In the The contribution value of each tree. Then, the multiple initial health risk data features are arranged in descending order according to the importance score, and the features with higher importance are placed in the front to generate a feature sequence. Then, the multiple initial health risk data features are optimized and screened based on the feature sequence by using data dimensionality reduction technology. Optionally, the multiple initial health risk data features are reduced in dimension using principal component analysis. The formula is: ,in are multiple health risk data features, i.e., principal components, is the feature weight, Score multiple features. By reducing the dimension, retain the features with high contribution and remove the redundant or noise features, so as to obtain the final multiple health risk data features. According to the feature sequence, select the first feature of the feature sequence as the input variable. The input variable refers to the core prediction factor used for model training. The gradient boosting decision tree model is initialized based on multiple health risk data features to obtain the initial prediction value. The initial prediction value is the basic prediction result of the gradient boosting decision tree model when the residual iteration is not performed. The calculation formula is: ;in, is the initial prediction value, is the loss function, is the number of samples, is the target value of sample i. At the same time, the target value is set based on multiple health risk data features. The target value is the standard value used by the gradient boosting decision tree model to guide the prediction results during the training process. It is used to evaluate the deviation of the model prediction value. When setting the target value, the historical data of the health risk data features can be used for statistical analysis and setting. For example, the average value of the historical health risk data features is used as the target value. Then, the residual iterative calculation is performed based on the initial prediction value and the target value to generate multiple residual values. The formula is: ;in, For the The sample in The residual of the tree, is the target variable value of sample i, For the The sample in -1 tree's predicted value, is the loss function, is a positive real number. The residual value represents the deviation between the initial prediction value of the model and the target value. It is used as the target variable for training weak learners to train the initial gradient boosting decision tree model, and multiple weak learners are trained by fitting the residual. The formula is: ;in, For the The output of the weak learner, is the number of samples, is the loss function, For the The sample in The residual of the tree, The output of the weak learner of the i-th sample is used to determine the number of decision trees required in the model. The number of decision trees represents the total number of sub-models used in the model ensemble. According to the set model learning rate, the multiple weak learner sets are updated to build a strong learner. The decision tree model is trained for multiple rounds of iterations to generate decision tree depth data. The learning rate is a positive real number between 0 and 1, which is used to control the step size of each iteration of the model to prevent overfitting. The learning rate is 0.1 or 0.01. The decision tree depth data is used to reflect the complexity of each decision tree split. The update formula for the strong learner is: ;in, is the learning rate, For the The output of the weak learner, For the sample in -1 tree's prediction value. Finally, based on the generated decision tree depth data, the input variables are optimized for hyperparameters using grid search. Grid search is a systematic parameter optimization method that exhaustively tests multiple hyperparameter combinations of the model, such as learning rate, maximum depth, number of split points, etc., to find the best parameter configuration. When the model's performance indicators, such as prediction accuracy or loss function value, tend to converge, the final hyperparameter combination is obtained, and the final strong learner is added to the gradient boosting decision tree model to complete the construction of the gradient boosting decision tree model and obtain the gradient boosting decision tree model.

[0022] Through the constructed gradient boosting decision tree model, individual diseases are analyzed and risks are predicted, and the individual disease prediction value is calculated. , and then use the first health risk score formula to generate a first health risk score S that measures the disease risk based on the predicted value. The first health risk score formula is: ;in, Score health risks, is the sample prediction value, e is a natural constant, and the value range of the health risk score S is 0~1. The closer S is to 1, the higher the disease risk is, and the closer it is to 0, the lower the disease risk is. The sample prediction value Output by the gradient boosting decision tree model. The health score S can intuitively reflect the disease risk. By constructing the gradient boosting decision tree model, the key features of health big data can be effectively utilized to predict the risk of individual diseases, providing a basis for generating health risk scores.

[0023] Perform a time series analysis of the health indicators according to the first health risk score to generate a time series dynamic change parameter, adjust the first health risk score according to the time series dynamic change parameter, and generate a second health risk score.

[0024] Furthermore, a time series analysis of health indicators is performed according to the first health risk score to generate time series dynamic change parameters, and the method includes: constructing health indicators based on the health big data, and retrieving multiple historical health data according to the health indicators combined with the first health risk score; serializing the multiple historical health data according to the collection time to generate a health time series data set; performing a change trend analysis based on the health time series data set to determine a dynamic change parameter set, wherein the dynamic change parameter set includes a data fluctuation amplitude value and a data period repetition pattern; performing a time series change analysis on the health time series data set according to the data fluctuation amplitude value according to the data period repetition pattern, and drawing a dynamic change graph; marking the extreme values ​​of the dynamic change graph according to the data fluctuation amplitude value to determine the time series dynamic change parameters.

[0025] Specifically, health indicators are constructed based on health big data. The health indicators are key variables or calculation results extracted from health data and used to quantify the health status of individuals. The health indicators not only reflect the physiological state of individuals, but are also closely related to the first health risk score. According to the health indicators combined with the first health risk score predicted by the model, historical health data related to individual health are retrieved from the historical database. The retrieved historical health data are serialized according to the acquisition time, and the historical health data are arranged in chronological order to generate a health time series data set. The health time series data set is a data structure constructed with time as the main axis, which can reflect the changes in health status at different time points. For example, a time series data set of a blood sugar index may contain daily measurements over a period of time. Then, according to the generated health time series data set, a change trend analysis is performed, and the data is decomposed into long-term trends, periodic fluctuations and random fluctuations using the time series decomposition method to reveal the overall change trend and periodic characteristics of health indicators. Apply sliding average or exponentially weighted moving average to smooth the data to further eliminate the impact of short-term fluctuations and highlight long-term change trends. And through autocorrelation analysis and Fourier transform, the periodic characteristics and potential patterns of the data are identified, and the change rules such as peaks, troughs, and cycle lengths are extracted. Finally, the change trend analysis results are generated to obtain a dynamic change parameter set, which includes data fluctuation amplitude values ​​and data period repetition patterns. The data fluctuation amplitude values ​​reflect the degree of fluctuation of health indicators in time series, such as the daily range of blood sugar levels. The data period repetition pattern describes whether the health indicators have periodic characteristics, such as some indicators may recur in seasonal changes or specific time periods. According to the periodic repetition pattern and fluctuation amplitude value of the data, the time series change analysis of the health time series data set is performed. Optionally, the periodicity and fluctuation characteristics in the time series data are quantitatively analyzed, such as calculating the cycle length, frequency distribution, and maximum and minimum values ​​of the fluctuation amplitude of the data, and identifying the key change points and stable intervals in the data. Subsequently, data visualization techniques such as line graphs, heat maps, or spectrum maps are used to intuitively present the analysis results and draw dynamic change graphs. By drawing dynamic change graphs, the fluctuations and trends of health indicators can be intuitively displayed, such as showing the rise and fall of blood sugar indicators over time through line graphs. On the basis of the dynamic change graph, extreme values ​​are marked according to the data fluctuation amplitude, marking the peak and trough values ​​in the time series data. Extreme values ​​are usually important reference points for major changes in health status. For example, extremely high blood pressure values ​​may indicate an increase in health risks. By comprehensively analyzing these extreme points and the overall change trend, time series dynamic change parameters are generated. The time series dynamic change parameters not only reveal the dynamic characteristics of health indicators, but also provide data basis for subsequent health risk adjustment and optimization.

[0026] Furthermore, the first health risk score is adjusted according to the time-series dynamic change parameter to generate a second health risk score, and the method includes: weighting the time-series dynamic change parameter based on the dynamic change graph to determine multiple weight coefficients; performing health risk impact analysis according to the multiple weight coefficients, and formulating score adjustment rules according to the impact analysis results; matching the first health risk score based on the score adjustment rule to determine whether the first health risk score is greater than or equal to the expected score threshold in the score adjustment rule; if the first health risk score is less than the expected score threshold in the score adjustment rule, generating an adjustment instruction, performing multi-dimensional adjustment on the first health risk score through the adjustment instruction, and generating the second health risk score.

[0027] Specifically, the weight distribution of the time series dynamic change parameters is performed according to the dynamic change diagram, and the weight distribution is used to measure the degree of influence of each dynamic change parameter on the health risk score. Optionally, the corresponding weight coefficients of each parameter of the time series dynamic change parameter can be assigned according to the magnitude of the fluctuation amplitude and the importance of the periodic pattern. For example, the larger the fluctuation amplitude, the higher the weight; the more significant the periodic repetition pattern, the more prominent the weight, and the weight coefficient reflects the relative degree of influence of each parameter on the health risk. Then, the health risk impact analysis is performed according to the assigned weight coefficients to evaluate the positive or negative impact of each dynamic change parameter on the health risk score, and quantify the specific role of each parameter in the score adjustment. For example, if the fluctuation amplitude of a parameter increases significantly and the periodic pattern becomes shorter, it may indicate an increase in risk and the need to increase the score; and if the fluctuation amplitude decreases, the score may be reduced. Based on the impact analysis results, a score adjustment rule is formulated, and the score adjustment rule includes the direction of the score adjustment, that is, increase or decrease, to ensure that the score result is more in line with the actual health status of the individual. Then, according to the formulated score adjustment rule, the first health risk score is matched and compared with the expected score threshold in the rule. The expected score threshold is a target value preset based on expert experience or historical data, which is used to distinguish the level of individual health risk. If the first health risk score is lower than the expected score threshold, it indicates that the risk has not met expectations, and an adjustment instruction is generated. The generated adjustment instruction corrects the first health risk score through a multi-dimensional adjustment method. Multi-dimensional adjustment refers to the comprehensive consideration of the weights of dynamic change parameters in different dimensions, such as the amplitude of time series changes and periodic characteristics. By comprehensively considering multiple health indicators and time series dynamic change parameters, the first health risk score is adjusted comprehensively and accurately. After the adjustment, the second health risk score is generated. The second health risk score not only considers the static health indicator data, but also fully integrates the influence of time series dynamic change parameters. It can more accurately reflect the comprehensive health risks of individuals under dynamic change conditions, and provide a more comprehensive and accurate data basis for individual health management and risk assessment.

[0028] Further, the time series dynamic change parameters are weighted based on the dynamic change graph to determine multiple weight coefficients, including: traversing the dynamic change graph to identify, determining multiple dynamic change node information and multiple edge information, the multiple edge information is used to connect at least two dynamic change node information; based on the time series dynamic change parameters, the initial eigenvalues ​​of the multiple dynamic change node information are calculated, and the formula is: ;in, is the time series dynamic change parameter of the i-th dynamically changing node information, is the initial eigenvalue of the i-th dynamically changing node information; edge weight calculation is performed based on the multiple edge information to obtain multiple edge weights, and the formula is: ;in, for Dynamically changing node information and Dynamically change the edge weight of node information, for Dynamically changing node information and The temporal distance of the dynamically changing node information; using the graph convolutional network to combine the multiple edge weights with the initial eigenvalues ​​of the multiple dynamically changing node information to perform weight allocation, and generate multiple weight coefficients, the formula is: ;in, After the kth iteration Dynamically change the weight of node information, is the activation function, for Dynamically changing node information and Dynamically change the edge weight of node information, After the k-1th iteration Dynamically change the weight of node information.

[0029] Specifically, the dynamic change graph is a form of graph structure representation, including multiple node information and edge information. The nodes represent specific parameters of dynamic changes, that is, the state or characteristics of parameter data at different time points, such as the time series fluctuation characteristics of health indicators. The edge information reflects the association or change relationship between these parameters. By traversing the dynamic change graph, all dynamic change node information and multiple edge information are identified. The dynamic change node information includes the initial characteristic values ​​of each time series dynamic change parameter, such as fluctuation amplitude, periodic pattern, etc. Multiple edge information is used to connect at least two dynamic change node information to describe the correlation between nodes. Then, using the formula , the initial eigenvalue of the dynamically changing node information is obtained according to the time series dynamic change parameters, where: is the time series dynamic change parameter of the i-th dynamically changing node information, such as the fluctuation amplitude, is the initial eigenvalue of the i-th dynamically changing node information. At the same time, edge weights are calculated based on multiple edge information to obtain multiple edge weights. The edge weight calculation formula is as follows: ,in, for Dynamically changing node information and Dynamically change the edge weight of node information, for Dynamically changing node information and The shorter the temporal distance of dynamically changing node information, the greater the weight, indicating that the temporal correlation between nodes is closer. Combining the initial feature value of the node and the edge weight, the graph convolution network is used to weight the initial features of the dynamically changing node information and generate multiple weight coefficients. The graph convolution network is a neural network model specially designed for processing graph structure data. It can update the feature value of the node in an iterative manner and consider the correlation between nodes. The specific allocation formula is: ;in, After the kth iteration Dynamically change the weight of node information, is the activation function, for Dynamically changing node information and Dynamically change the edge weight of node information, After the k-1th iteration The weight of dynamically changing node information. Through multiple rounds of iterations, each node in the dynamic change graph is eventually assigned a corresponding weight coefficient. The weight coefficient measures the degree of influence of each dynamically changing parameter on the adjustment of the health risk score. Constructing a dynamic change graph, calculating the initial eigenvalues, calculating the edge weights, and using a graph convolutional network for weight distribution make the weight distribution results more accurate and reliable, thereby ensuring that the scoring results are closer to the individual's dynamic health status. By constructing a gradient boosting decision tree model, the key features of health big data can be effectively utilized to accurately and reliably predict and evaluate the risk of individual diseases.

[0030] Furthermore, if the first health risk score is less than the expected score threshold in the score adjustment rule, an adjustment instruction is generated, and a multi-dimensional adjustment is performed on the first health risk score through the adjustment instruction to generate the second health risk score. The method includes: performing a change trend analysis on the time series dynamic change parameter based on the adjustment instruction to generate a first analysis result, and the first analysis result is a deterioration trend analysis result or an improvement trend analysis result; when the first analysis result is the deterioration trend analysis result, the first health risk score is increased and adjusted, and when the first analysis result is the improvement trend analysis result, the first health risk score is adjusted. The first health risk score is adjusted downward to generate a first adjustment result; a volatility analysis is performed on the data fluctuation amplitude value based on the adjustment instruction to generate a second analysis result, wherein the second analysis result is a high fluctuation amplitude analysis result or a low fluctuation amplitude analysis result; when the second analysis result is the high fluctuation amplitude analysis result, an increase adjustment is performed on the first health risk score, and when the second analysis result is the low fluctuation amplitude analysis result, a decrease adjustment is performed on the first health risk score to generate a second adjustment result; the first adjustment result is associated and integrated with the second adjustment result, and the first health risk score is adjusted according to the integration result to generate the second health risk score.

[0031] Specifically, according to the adjustment instruction, the time series dynamic change parameters are analyzed for change trends, and the evolution direction of health risks in the time dimension is evaluated. The change trend analysis generates a first analysis result by interpreting the fluctuation pattern and trend line in the dynamic change graph. If the first analysis result shows that the health risk is deteriorating, for example, the fluctuation amplitude in the dynamic change graph gradually increases or the trend line continues to rise, it indicates that the health condition may deteriorate. If the first analysis result is an improving trend, such as a decrease in the fluctuation amplitude or a decrease in the trend line, it indicates that the health condition has improved, that is, the first analysis result is a deterioration trend analysis result or an improvement trend analysis result. When the first analysis result is the deterioration trend analysis result, it indicates that the health condition may deteriorate, and the first health risk score is adjusted to increase. When the first analysis result is the improvement trend analysis result, it indicates that the health condition has improved, and the first health risk score is adjusted to decrease. Through this process, the first adjustment result is generated.

[0032] Next, based on the adjustment instruction, a volatility analysis is performed on the data fluctuation amplitude value to further evaluate the impact of the fluctuation characteristics of the health indicators on the risk score, and a second analysis result is obtained. The second analysis result is a high fluctuation amplitude analysis result or a low fluctuation amplitude analysis result. Volatility analysis focuses on quantifying the severity of fluctuations in time series data, which is used to reflect the stability of health status. If the second analysis result shows that the fluctuation amplitude is high, for example, some health indicators fluctuate greatly in a short period of time, which may indicate that the health status is unstable, the risk increases, and the first health risk score needs to be adjusted. If the fluctuation amplitude is low, such as the health indicators appear to be stable in the time dimension, indicating that the health status is relatively stable, then the risk is low and the score can be lowered. Through this process, the second adjustment result is generated.

[0033] Finally, the first adjustment result and the second adjustment result are correlated and integrated, and the analysis results of the changing trend and fluctuation of health indicators are comprehensively considered to form a global adjustment strategy for health risks, ensuring that the adjusted second health risk score can fully and accurately reflect the current health status and risk level. For example, if the deterioration trend analysis and high fluctuation analysis results appear at the same time, the score adjustment should be larger, and if the improvement trend and low fluctuation results exist at the same time, the adjustment should be smaller. The integration process is achieved by setting weighted rules or logical relationships. For example, by using logical relationships, priorities can be set according to the directions of the first and second adjustment results. When the first adjustment result is greater than 0 and the second adjustment result is greater than 0, the sum of the two is given priority to maximize the adjustment range; when the first adjustment result is greater than 0 and the second adjustment result is less than 0, the value of trend analysis is used as the main reference, but a certain proportion of fluctuation correction value is subtracted, such as the first adjustment result plus 0.5 multiplied by the second adjustment result. When the first adjustment result and the second adjustment result are in the same direction, they are both positive or negative and are directly added; if the directions are opposite, the smaller adjustment value is suppressed. Through integration, the adjustment results can reflect the global nature of health changes and take into account the immediacy of local fluctuations. According to the results of the associated integration, the first health risk score is subjected to the final multidimensional adjustment to generate the second health risk score. By adjusting the first health risk score to generate the second health risk score, the dual effects of time trend and fluctuation characteristics are fully considered, which can more accurately reflect the comprehensive health risk status of individuals under dynamic conditions.

[0034] Based on the second health risk score, health risk interpretation is performed to generate personalized health optimization suggestions, and a stratified health risk assessment is performed based on the personalized health optimization suggestions and the second health risk score to formulate a stratified intervention strategy.

[0035] Specifically, the health risk is interpreted according to the second health risk score. The health risk interpretation refers to extracting the key reasons for the score generation by analyzing the health risk data characteristics behind the health risk score, and generating personalized health optimization suggestions based on the risk interpretation results. Based on the personalized health optimization suggestions combined with the second health risk score, a stratified health risk assessment is performed. The assessment process divides individuals into the following three risk levels according to the score range of the second health risk score: low risk, medium risk and high risk. And according to the risk level, the individual health status is assessed. Low risk means that the individual is in good health and only requires preventive intervention. For example, it is recommended to continue to maintain a healthy lifestyle and exercise appropriately. Medium risk means that the individual may have potential health risks and further health management is required. For example, it is recommended to have regular physical examinations, focusing on identified risk factors, such as weight management or blood lipid control. High risk means that the individual has significant health problems and needs immediate medical intervention. For example, it is recommended to visit relevant specialists for in-depth examination and treatment of marked high-risk factors, such as cardiovascular disease risks. Then, according to the results of the stratified assessment, a corresponding stratified intervention strategy is formulated. The intervention strategy is tailored according to the individual characteristics and health status requirements of different risk levels. For low-risk individuals, preventive intervention is recommended, such as providing health knowledge education or guiding exercise plans; for medium-risk individuals, regular management is recommended, combined with personalized optimization suggestions for dynamic health monitoring; for high-risk individuals, priority medical intervention is recommended, such as recommending expert diagnosis and treatment or implementing a drug treatment plan. Through the complete closed loop of risk interpretation, stratified assessment and intervention strategy of the second health risk score, not only the accuracy and pertinence of health risk assessment is improved, but also the accuracy and effectiveness of health management can be improved.

[0036] Furthermore, health risk interpretation is performed based on the second health risk score to generate personalized health optimization suggestions, and the method includes: calculating the multiple health risk data features based on the second health risk score to obtain multiple health feature contribution degrees; interpreting health risks based on the multiple health feature contribution degrees combined with the second health risk score to generate a health risk interpretation report, wherein the health risk interpretation report includes positive risk factors and negative risk factors; generating a first optimization suggestion based on the negative risk factor combined with the multiple health risk data features, and generating a second optimization suggestion based on the positive risk factor combined with the multiple health feature contribution degrees; and performing data fusion on the first optimization suggestion and the second optimization suggestion to obtain the personalized health optimization suggestion.

[0037] Specifically, the multiple health risk data features are calculated according to the second health risk score, and the actual contribution value of each feature to the score is quantitatively evaluated by assigning feature weights, using feature importance scores or model interpretation tools such as SHAP value algorithms, to generate health feature contribution, which is a quantitative indicator that reflects the degree of influence of a certain feature in the risk score. Then, based on the health feature contribution combined with the second health risk score, health risk interpretation is performed, positive factors and negative factors that have a greater impact on the health risk score are identified, and a health risk interpretation report is generated. The health risk interpretation report contains positive risk factors and negative risk factors. The positive risk factor refers to a feature that has a positive effect on the health status, such as a normal blood pressure value or a good exercise habit, and the negative risk factor refers to a feature that has a negative impact on the health status, such as hyperlipidemia or a history of smoking. The health risk interpretation report not only outlines the main risk points of the current health status, but also analyzes which features can be further optimized. Based on the negative risk factors in the health risk interpretation report, combined with the multiple health risk data features, a first optimization suggestion is generated. The first optimization suggestion focuses on improving health status, such as suggesting to adjust the diet structure to lower cholesterol, or suggesting to increase the frequency of exercise to improve cardiovascular health. At the same time, the second optimization suggestion is generated based on the positive risk factor combined with the contribution of health characteristics. The second optimization suggestion is used to consolidate and maintain the health status, such as encouraging patients to continue to maintain good eating habits or stick to the existing exercise plan. Finally, the first optimization suggestion is fused with the second optimization suggestion by weighing priorities and constraints to form personalized health optimization suggestions. The data fusion process balances the contradictions in health management by integrating the improvement direction of positive and negative factors. For example, the first optimization suggestion for a patient is to increase the frequency of daily exercise to reduce BMI, and the second optimization suggestion is to maintain the existing knee joint protection plan to avoid joint injury. When fusion of data, by analyzing the possible contradictions between the two suggestions, such as increasing exercise may increase the pressure on the knee joint, setting weights or constraints to balance the impact of the two, the final fusion solution is to increase the frequency of exercise, but choose low-impact exercise, such as swimming or cycling, to reduce the burden on the knee joint, while maintaining the knee joint protection plan, combined with moderate exercise to improve overall health indicators. Through data fusion, personalized health optimization suggestions integrate weight loss goals and joint protection needs, achieving a dynamic balance of health risk factors and ensuring the feasibility and scientificity of optimization suggestions. By combining risk interpretation, feature analysis and optimization suggestions, conflicts between health management measures can be effectively avoided, providing patients with comprehensive and practical health improvement plans, and improving the accuracy and feasibility of individual health management.

[0038] The target individuals are subjected to health checks according to the tiered intervention strategy, and health warnings are triggered according to abnormal detection results.

[0039] Specifically, according to the stratified intervention strategy, the target individuals are tested for health. For low-risk individuals, the test content is mainly based on basic health indicators, such as weight, blood pressure and heart rate, which are mainly used to regularly monitor health status. For medium-risk individuals, targeted items are added, such as blood sugar levels, cholesterol and organ function tests. For high-risk individuals, the test content is more comprehensive, including imaging examinations, special indicator analysis, such as tumor markers, etc. The collected test data is tested for abnormalities. When there are abnormal data, a health warning is triggered according to the abnormal detection results. The health warning refers to the abnormal data detected by the abnormality, which quickly identifies potential health risks and sends warning information to the target individuals. The warning information includes detailed abnormal detection results, possible health risks and recommended intervention measures. For example, for high-risk individuals, if the electrocardiogram shows acute abnormalities, it means that the target individual's heart function may have serious problems. The warning information will clearly indicate the abnormal type of electrocardiogram, such as arrhythmia, myocardial ischemia, etc., and point out possible health risks, and recommend emergency medical treatment. Health testing and warning through stratified intervention strategies can reveal health risks in a timely manner, providing accurate and efficient support for individualized health management.

[0040] Embodiment 2, based on the same inventive concept as a health risk intelligent assessment method in the above embodiment, Figure 2 As shown, the present application provides a health risk intelligent assessment system, wherein the system comprises: A data acquisition module 11, the data acquisition module 11 is used to collect multi-dimensional health data based on natural language and construct health big data; a risk prediction module 12, the risk prediction module 12 is used to perform feature selection through the health big data to obtain multiple health risk data features, and construct a gradient boosting decision tree model based on the multiple health risk data features, and predict the risk of individual diseases through the gradient boosting decision tree model to generate a first health risk score; a score adjustment module 13, the score adjustment module 13 is used to perform time series analysis of health indicators according to the first health risk score, generate time series dynamic change parameters, adjust the first health risk score according to the time series dynamic change parameters, and generate a second health risk score; a risk assessment module 14, the risk assessment module 14 is used to interpret health risks based on the second health risk score, generate personalized health optimization suggestions, perform health risk stratification assessment based on the personalized health optimization suggestions combined with the second health risk score, and formulate a stratified intervention strategy; a health detection module 15, the health detection module 15 is used to perform health detection on the target individual according to the stratified intervention strategy, and trigger a health warning based on abnormal detection results.

[0041] Furthermore, the data acquisition module 11 is used to perform the following steps: generate multiple character tags by performing optical character recognition on the medical database, classify and annotate the text data with the multiple character tags, and obtain multiple medical text identification data; perform voice input according to the acquisition target to generate multiple voice data, convert the multiple voice data into text to extract multiple text words, and obtain multiple medical semantic classification data; perform in-clinic dictation according to the doctor-patient conversation record to generate multiple conversation record data, structure the multiple conversation record data to obtain multiple medical structured health data; clean the multiple medical text identification data, the multiple medical semantic classification data, and the multiple medical structured health data, and perform data standardization and alignment according to the storage architecture based on the data cleaning results to construct the health big data.

[0042] Further, the risk prediction module 12 is used to perform the following steps: using random forest to traverse the health big data to obtain multiple initial health risk data features, each of the multiple initial health risk data features has an importance score; arranging the multiple initial health risk data features in descending order according to the multiple importance scores to generate a feature sequence, performing data dimension reduction on the multiple initial health risk data features based on the feature sequence, and determining multiple health risk data features; extracting the first-order health risk data feature as an input variable based on the feature sequence, performing initialization calculation based on the multiple health risk data features, obtaining an initial prediction value, and setting a target value; training to determine the number of decision trees, iterating according to the model learning rate based on the number of decision trees to obtain decision tree depth data; according to the decision tree depth data, performing hyperparameter optimization on the input variables according to grid search to determine multiple parameter combinations, testing and verifying the multiple parameter combinations, and obtaining the gradient boosting decision tree model when the multiple parameter combinations tend to converge; using the gradient boosting decision tree model to predict the risk of individual diseases and generate a first health risk score, the formula is: ;in, Score health risks, is the sample prediction value, and e is a natural constant.

[0043] Furthermore, the score adjustment module 13 is used to perform the following steps: construct a health index based on the health big data, and retrieve multiple historical health data according to the health index combined with the first health risk score; serialize the multiple historical health data according to the collection time to generate a health time series data set; perform a change trend analysis based on the health time series data set to determine a dynamic change parameter set, and the dynamic change parameter set includes a data fluctuation amplitude value and a data period repetition pattern; perform a time series change analysis on the health time series data set according to the data fluctuation amplitude value according to the data period repetition pattern, and draw a dynamic change graph; mark the extreme values ​​of the dynamic change graph according to the data fluctuation amplitude value to determine the time series dynamic change parameters.

[0044] Further, the score adjustment module 13 is also used to perform the following steps: weighting the time series dynamic change parameters based on the dynamic change graph to determine multiple weight coefficients; performing health risk impact analysis according to the multiple weight coefficients, and formulating score adjustment rules according to the impact analysis results; matching the first health risk score based on the score adjustment rule to determine whether the first health risk score is greater than or equal to the expected score threshold in the score adjustment rule; if the first health risk score is less than the expected score threshold in the score adjustment rule, generating an adjustment instruction, performing multi-dimensional adjustment on the first health risk score through the adjustment instruction, and generating the second health risk score; Furthermore, the score adjustment module 13 is further used to perform the following steps: traverse the dynamic change graph to identify, determine multiple dynamically changing node information and multiple edge information, the multiple edge information is used to connect at least two dynamically changing node information; calculate the initial characteristic values ​​of the multiple dynamically changing node information based on the time series dynamic change parameter, the formula is: ;in, is the time series dynamic change parameter of the i-th dynamically changing node information, is the initial eigenvalue of the i-th dynamically changing node information; edge weight calculation is performed based on the multiple edge information to obtain multiple edge weights, and the formula is: ;in, for Dynamically changing node information and Dynamically change the edge weight of node information, for Dynamically changing node information and The temporal distance of the dynamically changing node information; using the graph convolutional network to combine the multiple edge weights with the initial eigenvalues ​​of the multiple dynamically changing node information to perform weight allocation, and generate multiple weight coefficients, the formula is: ;in, After the kth iteration Dynamically change the weight of node information, is the activation function, for Dynamically changing node information and Dynamically change the edge weight of node information, After the k-1th iteration Dynamically change the weight of node information.

[0045] Furthermore, the score adjustment module 13 is also used to perform the following steps: based on the adjustment instruction, performing a change trend analysis on the time series dynamic change parameter to generate a first analysis result, the first analysis result being a deterioration trend analysis result or an improvement trend analysis result; when the first analysis result is the deterioration trend analysis result, the first health risk score is adjusted to increase, and when the first analysis result is the improvement trend analysis result, the first health risk score is adjusted to decrease, and a first adjustment result is generated; based on the adjustment instruction, performing a volatility analysis on the data fluctuation amplitude value to generate a second analysis result, the second analysis result being a high fluctuation amplitude analysis result or a low fluctuation amplitude analysis result; when the second analysis result is the high fluctuation amplitude analysis result, the first health risk score is adjusted to increase, and when the second analysis result is the low fluctuation amplitude analysis result, the first health risk score is adjusted to decrease, and a second adjustment result is generated; the first adjustment result is associated and integrated with the second adjustment result, and the first health risk score is adjusted according to the integration result to generate the second health risk score.

[0046] Furthermore, the risk assessment module 14 is used to perform the following steps: calculate the multiple health risk data features based on the second health risk score to obtain multiple health feature contribution degrees; interpret the health risks based on the multiple health feature contribution degrees combined with the second health risk score, and generate a health risk interpretation report, wherein the health risk interpretation report includes positive risk factors and negative risk factors; generate a first optimization suggestion based on the negative risk factor combined with the multiple health risk data features, and generate a second optimization suggestion based on the positive risk factor combined with the multiple health feature contribution degrees; perform data fusion on the first optimization suggestion and the second optimization suggestion to obtain the personalized health optimization suggestion.

[0047] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0048] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A health risk intelligent assessment method, characterized in that: The method comprises: Collect multi-dimensional health data based on natural language and build health big data; Performing feature selection through the health big data to obtain multiple health risk data features, constructing a gradient boosting decision tree model based on the multiple health risk data features, predicting the risk of individual diseases through the gradient boosting decision tree model, and generating a first health risk score; Performing a time series analysis of health indicators according to the first health risk score to generate a time series dynamic change parameter, and adjusting the first health risk score according to the time series dynamic change parameter to generate a second health risk score; Perform health risk interpretation based on the second health risk score, generate personalized health optimization suggestions, perform health risk stratification assessment based on the personalized health optimization suggestions combined with the second health risk score, and formulate a stratified intervention strategy; Perform health checks on target individuals according to the tiered intervention strategy, and trigger health warnings based on abnormal test results; The step of adjusting the first health risk score according to the time series dynamic change parameter to generate a second health risk score includes: Based on the dynamic change graph, weights are assigned to the time series dynamic change parameters to determine a plurality of weight coefficients; Conduct health risk impact analysis according to the multiple weight coefficients, and formulate score adjustment rules based on the impact analysis results; Matching the first health risk score based on the score adjustment rule, and determining whether the first health risk score is not less than an expected score threshold in the score adjustment rule; If the first health risk score is less than the expected score threshold in the score adjustment rule, an adjustment instruction is generated, and a multi-dimensional adjustment is performed on the first health risk score through the adjustment instruction to generate the second health risk score; The step of performing feature selection through the health big data to obtain a plurality of health risk data features, and constructing a gradient boosting decision tree model according to the plurality of health risk data features includes: Using random forest to traverse and train the health big data to obtain a plurality of initial health risk data features, each of the plurality of initial health risk data features having an importance score; Arrange the multiple initial health risk data features in descending order according to multiple importance scores to generate a feature sequence, and perform data dimension reduction on the multiple initial health risk data features based on the feature sequence to determine multiple health risk data features; Extracting the first-order health risk data feature as an input variable based on the feature sequence, performing initialization calculation according to the multiple health risk data features, obtaining an initial prediction value, and setting a target value; Performing training according to the initial prediction value combined with the target value to determine the number of decision trees, iterating according to the model learning rate based on the number of decision trees to obtain decision tree depth data; According to the decision tree depth data, the input variables are optimized for hyperparameters according to grid search, multiple parameter combinations are determined, and the multiple parameter combinations are tested and verified. When the multiple parameter combinations tend to converge, the gradient boosting decision tree model is obtained; The step of predicting the risk of an individual disease by using the gradient boosting decision tree model to generate a first health risk score includes: ; in, Score health risks, is the sample prediction value, and e is a natural constant.

2. A health risk intelligent assessment method as claimed in claim 1, characterized in that: The multi-dimensional health data collection based on natural language and the construction of health big data include: By performing optical character recognition on a medical database, a plurality of character labels are generated, and the text data are classified and annotated using the plurality of character labels to obtain a plurality of medical text identification data; Perform voice input according to the acquisition target to generate multiple voice data, convert the multiple voice data into text to extract multiple text words, and obtain multiple medical semantic classification data; Perform in-clinic dictation based on the doctor-patient conversation record to generate multiple conversation record data, and perform structured processing on the multiple conversation record data to obtain multiple medical structured health data; The multiple medical text identification data, the multiple medical semantic classification data, and the multiple medical structured health data are cleaned, and data standardization and alignment are performed according to the storage architecture based on the data cleaning results to construct the health big data.

3. A health risk intelligent assessment method as claimed in claim 1, characterized in that: The step of performing a time series analysis of health indicators according to the first health risk score to generate a time series dynamic change parameter includes: constructing a health index based on the health big data, and retrieving a plurality of historical health data according to the health index combined with the first health risk score; Serializing the plurality of historical health data according to the collection time to generate a health time series data set; Performing a change trend analysis based on the health time series data set to determine a dynamic change parameter set, wherein the dynamic change parameter set includes a data fluctuation amplitude value and a data period repetition mode; Performing time series change analysis on the health time series data set according to the data period repetition mode and the data fluctuation amplitude value, and drawing a dynamic change graph; The dynamic change graph is marked with extreme values ​​according to the data fluctuation amplitude value to determine the time series dynamic change parameters.

4. A health risk intelligent assessment method as claimed in claim 3, characterized in that: The performing multi-dimensional adjustment on the first health risk score by using the adjustment instruction to generate the second health risk score includes: Performing a change trend analysis on the timing dynamic change parameter based on the adjustment instruction to generate a first analysis result, wherein the first analysis result is a deterioration trend analysis result or an improvement trend analysis result; When the first analysis result is the deterioration trend analysis result, the first health risk score is adjusted upward, and when the first analysis result is the improvement trend analysis result, the first health risk score is adjusted downward to generate a first adjustment result; Performing a volatility analysis on the data fluctuation amplitude value based on the adjustment instruction to generate a second analysis result, where the second analysis result is a high fluctuation amplitude analysis result or a low fluctuation amplitude analysis result; When the second analysis result is the high fluctuation range analysis result, the first health risk score is increased and adjusted; when the second analysis result is the low fluctuation range analysis result, the first health risk score is decreased and adjusted to generate a second adjustment result; The first adjustment result and the second adjustment result are associated and integrated, and the first health risk score is adjusted according to the integration result to generate the second health risk score.

5. A health risk intelligent assessment method as claimed in claim 1, characterized in that: The performing health risk interpretation based on the second health risk score and generating personalized health optimization suggestions includes: Calculating the plurality of health risk data features based on the second health risk score to obtain a plurality of health feature contribution degrees; Performing health risk interpretation based on the contribution of the multiple health characteristics combined with the second health risk score, and generating a health risk interpretation report, wherein the health risk interpretation report includes positive risk factors and negative risk factors; Generate a first optimization suggestion based on the negative risk factor combined with the multiple health risk data features, and generate a second optimization suggestion based on the positive risk factor combined with the multiple health feature contribution degrees; The first optimization suggestion is data-fused with the second optimization suggestion to obtain the personalized health optimization suggestion.

6. A health risk intelligent assessment method as claimed in claim 1, characterized in that: Based on the dynamic change graph, weights are assigned to the time series dynamic change parameters to determine a plurality of weight coefficients, including: Traversing the dynamic change graph to identify and determine a plurality of dynamically changing node information and a plurality of edge information, wherein the plurality of edge information is used to connect at least two dynamically changing node information; The initial characteristic values ​​of the plurality of dynamically changing node information are calculated based on the time series dynamically changing parameters, and the formula is: ; in, is the time series dynamic change parameter of the i-th dynamically changing node information, is the initial eigenvalue of the i-th dynamically changing node information; The edge weights are calculated based on the multiple edge information to obtain multiple edge weights, and the formula is: ; in, for Dynamically changing node information and Dynamically change the edge weight of node information, for Dynamically changing node information and The temporal distance of dynamically changing node information; The graph convolutional network is used to combine the multiple edge weights with the initial eigenvalues ​​of the multiple dynamically changing node information to perform weight allocation, and generate multiple weight coefficients. The formula is: ; in, After the kth iteration Dynamically change the weight of node information, is the activation function, for Dynamically changing node information and Dynamically change the edge weight of node information, After the k-1th iteration Dynamically change the weight of node information.

7. A health risk intelligent assessment system, characterized in that: The system is used to implement the steps of a health risk intelligent assessment method according to any one of claims 1 to 6, and the system comprises: A data collection module, which is used to collect multi-dimensional health data based on natural language and construct health big data; A risk prediction module, wherein the risk prediction module is used to perform feature selection through the health big data to obtain multiple health risk data features, construct a gradient boosting decision tree model according to the multiple health risk data features, perform risk prediction of individual diseases through the gradient boosting decision tree model, and generate a first health risk score; A score adjustment module, the score adjustment module is used to perform a time series analysis of health indicators according to the first health risk score, generate a time series dynamic change parameter, and adjust the first health risk score according to the time series dynamic change parameter to generate a second health risk score; a risk assessment module, the risk assessment module being used to interpret health risks based on the second health risk score, generate personalized health optimization suggestions, perform stratified health risk assessment based on the personalized health optimization suggestions combined with the second health risk score, and formulate stratified intervention strategies; A health detection module is used to perform health detection on the target individual according to the tiered intervention strategy and trigger health warning according to abnormal detection results.

Images