Artificial intelligence-based auxiliary enterprise employee serious disease risk prediction method, apparatus and device, and medium

Abstract

The invention relates to an artificial intelligence-based auxiliary enterprise employee serious disease risk prediction method and device, equipment and a medium. The method comprises the following steps: acquiring a multi-source heterogeneous data set of enterprise employees; the multi-source heterogeneous data set comprises static health record data, dynamic physiological monitoring data, post exclusive work behavior data and organization environment data; generating a dynamic risk feature vector based on the multi-source heterogeneous data set; inputting the dynamic risk feature vector into a trained serious disease risk prediction model to obtain a personalized serious disease risk portrait; the personalized serious disease risk portrait comprises the occurrence probability of the serious disease of the enterprise employee, the type of the high-risk serious disease and the risk contribution proportion. By adopting the method, dynamic and accurate prediction of the enterprise employee serious disease risk can be realized, and the pertinence of enterprise health management is effectively improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of health risk prediction, and particularly relates to an artificial intelligence-based enterprise employee serious illness risk prediction method, device, equipment and medium. BACKGROUND

[0002] With the deep integration of artificial intelligence technology in the medical and health field, disease risk prediction technology based on artificial intelligence has emerged.

[0003] This technology can mine potential laws from massive health data, build prediction models, and realize the assessment of individual future disease risk, thereby providing the possibility for early intervention and precise health management.

[0004] However, the current technology has obvious deficiencies. First, the scene adaptability is poor, and a general serious illness prediction model is difficult to match the job health risk characteristics of enterprise employees, and the prediction accuracy is limited; second, the data integration is insufficient, and scattered health data cannot fully support the in-depth analysis of serious illness risk; third, the management mode is extensive, and most of them are static post-treatment, lacking dynamic monitoring and accurate early warning of employee serious illness risk; fourth, the individualization is insufficient, and it is difficult to provide targeted risk assessment results according to the professional characteristics of employees, and it cannot meet the actual needs of enterprise fine health management and employee serious illness prevention. SUMMARY

[0005] Therefore, it is necessary to provide an artificial intelligence-based enterprise employee serious illness risk prediction method in view of the above technical problems.

[0006] In a first aspect, the present application provides an artificial intelligence-based enterprise employee serious illness risk prediction method, comprising:

[0007] Obtaining a multi-source heterogeneous data set of enterprise employees; the multi-source heterogeneous data set includes static health record data, dynamic physiological monitoring data, post-specific work behavior data and organizational environment data;

[0008] Generating a dynamic risk feature vector based on the multi-source heterogeneous data set;

[0009] Inputting the dynamic risk feature vector into a trained serious illness risk prediction model to obtain an individualized serious illness risk portrait; the individualized serious illness risk portrait includes the probability of occurrence of serious illness of the enterprise employee, the type of high-risk serious illness and the risk contribution proportion.

[0010] In one of the embodiments, the method further comprises:

[0011] Based on the individualized serious illness risk portrait, matching risk driving factors from a preset intervention strategy library to generate an associated intervention strategy cluster;

[0012] generate a health intervention plan according to the associated intervention strategy cluster; the health intervention plan comprises health suggestions, physical examination item recommendations, and offline service guidance;

[0013] collect execution compliance data of the enterprise employees on the health intervention plan, and obtain physiological index trajectory data of the enterprise employees after a preset period;

[0014] generate a feedback sample data set based on the execution compliance data and the physiological index trajectory data;

[0015] adjust weight parameters of the critical illness risk prediction model using the feedback sample data set to obtain an updated critical illness risk prediction model.

[0016] In one of the embodiments, the health intervention plan is generated according to the associated intervention strategy cluster, comprising:

[0017] generate a candidate intervention plan based on the associated intervention strategy cluster;

[0018] score the individualized adaptation degree of the candidate intervention plan according to the post characteristics and historical compliance preferences of the enterprise employees;

[0019] select an optimal candidate intervention plan sequence based on the individualized adaptation degree score, and fill in the health suggestion content, the physical examination item parameters, and the offline service resource information into the optimal candidate intervention plan sequence to generate the health intervention plan.

[0020] In one of the embodiments, the individualized adaptation degree score of the candidate intervention plan is scored according to the post characteristics and historical compliance preferences of the enterprise employees, comprising:

[0021] The individualized adaptation degree score is calculated using the following formula:

[0022]

[0023] wherein, is the individualized adaptation degree score, is the total number of candidate intervention plans, is the base weight of the th intervention plan, is the matching degree of the th intervention plan and the post characteristics, is the preference coefficient based on the historical compliance, is the time decay factor, is the execution time period of the th intervention plan.

[0024] In one of the embodiments, the weight parameters of the critical illness risk prediction model are adjusted using the feedback sample data set to obtain an updated critical illness risk prediction model, comprising:

[0025] Based on the feedback sample data set, key feedback features are extracted by an incremental learning method;

[0026] The key feedback features are fused with a dynamic risk feature vector to obtain an enhanced feature representation;

[0027] Using the enhanced feature representation, the weight parameters of the critical illness risk prediction model are optimized and adjusted by a back propagation algorithm to obtain an updated critical illness risk prediction model.

[0028] In one of the embodiments, based on the personalized critical illness risk portrait, a risk driving factor is matched from a preset intervention strategy library to generate an associated intervention strategy cluster, including:

[0029] Based on the risk contribution proportion, the high-risk critical illness type and the corresponding risk driving factor are prioritized to obtain a core intervention target;

[0030] Taking the core intervention target as a search key, direct intervention strategies and indirect intervention strategies are searched in parallel from the preset intervention strategy library to generate an initial strategy set;

[0031] Based on the preset synergistic effect rule, the initial strategy set is optimized to generate an associated intervention strategy cluster.

[0032] In one of the embodiments, based on a multi-source heterogeneous data set, a dynamic risk feature vector is generated, including:

[0033] Based on the multi-source heterogeneous data set, a unified spatiotemporal benchmark fusion data set is obtained through spatiotemporal consistency reconstruction;

[0034] The fusion data set is subjected to time series data regularization to obtain a standardized time series matrix;

[0035] Based on the standardized time series matrix, a multi-scale time semantic tensor is obtained;

[0036] Based on the multi-scale time semantic tensor, static health record data and organizational environment data are mapped into static embedding vectors through static feature embedding;

[0037] Based on the multi-scale time semantic tensor and the static embedding vector, a heterogeneous fusion feature is obtained;

[0038] The heterogeneous fusion feature is subjected to feature integration to obtain a dynamic risk feature vector.

[0039] In a second aspect, the present application also provides an artificial intelligence assisted enterprise employee critical illness risk prediction device, including:

[0040] The data acquisition module is used to acquire multi-source heterogeneous datasets of enterprise employees; the multi-source heterogeneous datasets include static health record data, dynamic physiological monitoring data, job-specific work behavior data, and organizational environment data;

[0041] The risk feature vector generation module is used to generate dynamic risk feature vectors based on multi-source heterogeneous datasets.

[0042] The risk profile generation module is used to input dynamic risk feature vectors into a trained critical illness risk prediction model to obtain a personalized critical illness risk profile. The personalized critical illness risk profile includes the probability of critical illness occurrence, high-risk critical illness types, and risk contribution ratio of enterprise employees.

[0043] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0044] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0045] The aforementioned method, device, equipment, and medium for AI-assisted prediction of critical illness risk among enterprise employees acquire multi-source heterogeneous datasets of enterprise employees. These datasets include static health record data, dynamic physiological monitoring data, job-specific work behavior data, and organizational environment data. Based on these datasets, dynamic risk feature vectors are generated. These vectors are then input into a trained critical illness risk prediction model to obtain a personalized critical illness risk profile. This profile includes the probability of critical illness occurrence, high-risk critical illness types, and their risk contribution percentage for each employee. This method enables dynamic and accurate prediction of critical illness risk among enterprise employees and effectively improves the targeted nature of enterprise health management. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 A flowchart illustrating an artificial intelligence-assisted method for predicting the risk of serious illnesses among enterprise employees, provided as an exemplary embodiment of this application;

[0048] Figure 2This is a schematic diagram of the structure of an artificial intelligence-assisted enterprise employee critical illness risk prediction device provided as an exemplary embodiment of this application. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0050] The application scenarios of the embodiments of this application will be described in conjunction with the implementation environment.

[0051] The implementation environment of the embodiments of this application will be described. Indicatively, the implementation environment includes a smart terminal, which can possess strong computing and data processing capabilities to meet data processing needs.

[0052] The AI-assisted method for predicting the risk of serious illness among employees provided in this application can be used in enterprise health management, prevention of serious illness among employees, development of personalized intervention plans, actuarial assessment, and human resource planning.

[0053] In one embodiment, such as Figure 1 As shown, an artificial intelligence-based method for predicting the risk of serious illnesses among enterprise employees is provided. This embodiment illustrates the application of this method to a smart terminal. It is understood that this method can also be applied to a server, or to a system including both a terminal and a server, and is implemented through the interaction between the smart terminal and the server. In this embodiment, the method includes the following steps:

[0054] Step S101: Obtain multi-source heterogeneous datasets of enterprise employees; the multi-source heterogeneous datasets include static health record data, dynamic physiological monitoring data, job-specific work behavior data, and organizational environment data.

[0055] Among them, multi-source heterogeneous datasets can be comprehensive data collections from different data sources with differences in data type, format and structure, used for predicting the risk of serious illnesses of enterprise employees. These include static health record data, dynamic physiological monitoring data, job-specific work behavior data and organizational environment data. Static health record data can be health-related information from medical systems and institutional or corporate HR archives that does not change rapidly over time and has persistence and stability (such as basic physiological information, past health records, and fixed health indicator baselines); dynamic physiological monitoring data can be dynamic data reflecting real-time or periodic changes in the physiological state of employees (such as physiological indicators that can be collected in real time, such as heart rate, blood pressure, blood oxygen saturation, blood glucose fluctuations, electrocardiogram, body fat percentage, sleep quality, and exercise intensity); job-specific work behavior data can be dynamic data that is strongly correlated with the characteristics of employees' jobs and reflects work habits and behavior patterns in occupational scenarios (such as work hours, overtime frequency, task intensity, sedentary / standing time, number of high-altitude operations, night shift records, frequency of repetitive operations, cross-departmental collaboration time, and outdoor work time); organizational environment data can be a combination of static and dynamic data that is directly related to the corporate environment and affects the health status of employees (such as office area temperature and humidity, air quality, noise intensity, radiation dose, lighting conditions, ventilation efficiency, job environment risk level, departmental work pressure atmosphere, and the completeness of workplace health and safety systems).

[0056] Specifically, the smart terminal acquires dynamic physiological monitoring data, job-specific work behavior data, and organizational environment data of enterprise employees, while retrieving pre-stored static health record data of employees, and integrates them to obtain a complete multi-source heterogeneous dataset.

[0057] Step S102: Generate dynamic risk feature vectors based on multi-source heterogeneous datasets.

[0058] Among them, the dynamic risk feature vector can be a high-dimensional numerical vector that dynamically reflects the employee's critical illness risk status.

[0059] Specifically, smart terminals can perform spatiotemporal consistency reconstruction processing on multi-source heterogeneous datasets to obtain a fused dataset with a unified spatiotemporal benchmark. Subsequently, the fused dataset is subjected to time-series data normalization, and data of different magnitudes and dimensions are transformed into a standardized time-series matrix with a unified standard. At the same time, non-quantitative information in static health record data and organizational environment data is transformed into computable static embedding vectors through static feature embedding algorithms. Furthermore, multi-scale temporal semantic tensors and static embedding vectors are fused heterogeneously to finally generate dynamic risk feature vectors.

[0060] Step S103: Input the dynamic risk feature vector into the trained critical illness risk prediction model to obtain a personalized critical illness risk profile; the personalized critical illness risk profile includes the probability of critical illness occurrence of enterprise employees, high-risk critical illness types, and risk contribution ratio.

[0061] The critical illness risk prediction model can employ a deep neural network based on an attention mechanism, comprising a feature embedding layer, multi-scale temporal convolutional layers, bidirectional LSTM layers, and an interpretable output layer. This model can dynamically assess and attribute the critical illness risk of enterprise employees, supporting personalized health intervention decisions. For example, the input layer of the critical illness risk prediction model can receive dynamic risk feature vectors, and the output layer can generate a personalized critical illness risk profile.

[0062] Personalized critical illness risk profiles are digital risk analysis reports generated based on the output of critical illness risk prediction models for specific employees. These reports include the probability of critical illness occurrence for employees, types of high-risk critical illnesses, and the percentage of risk contribution.

[0063] Specifically, the smart terminal inputs the dynamic risk feature vector into the trained critical illness risk prediction model. Then, the critical illness risk prediction model, through collaborative processing at various levels, deeply mines key information related to critical illness risk from the dynamic risk feature vector, and finally outputs a personalized critical illness risk profile. This profile is a digital risk analysis report for a specific employee, which mainly covers the probability of critical illness occurrence, high-risk critical illness types, and risk contribution ratio of the employee.

[0064] In the aforementioned AI-assisted method for predicting the risk of serious illnesses among enterprise employees, a smart terminal acquires a multi-source heterogeneous dataset of enterprise employees. This dataset includes static health record data, dynamic physiological monitoring data, job-specific work behavior data, and organizational environment data. Based on this dataset, a dynamic risk feature vector is generated. This dynamic risk feature vector is then input into a trained serious illness risk prediction model to obtain a personalized serious illness risk profile. The personalized profile includes the probability of serious illness occurrence, high-risk serious illness types, and risk contribution percentage for each employee. This method enables dynamic and accurate prediction of the risk of serious illnesses among enterprise employees and effectively improves the targeting of enterprise health management.

[0065] In one embodiment, the method may further include the following steps:

[0066] Step S201: Based on the personalized critical illness risk profile, risk driving factors are matched from the preset intervention strategy library to generate a cluster of related intervention strategies.

[0067] The preset intervention strategy library can be a pre-built resource collection that stores various critical illness risk intervention programs, covering direct and indirect intervention strategies for different high-risk critical illness types and risk driving factors.

[0068] Risk drivers can be key factors that are related to the risk of serious illness among employees and influence the probability of serious illness.

[0069] The associated intervention strategy cluster can be a set of strategies specifically adapted to the employee's critical illness risk, generated by sorting high-risk critical illness types and corresponding risk driving factors to obtain core intervention targets, retrieving direct and indirect intervention strategies in parallel from a preset intervention strategy library, and then optimizing them through synergistic effect rules.

[0070] Specifically, the smart terminal can prioritize high-risk critical illness types and their corresponding risk drivers (key factors affecting the probability of critical illness occurrence) based on the risk contribution ratio in the personalized critical illness risk profile, and determine the core intervention targets. Then, using the core intervention targets as search keys, it can retrieve suitable intervention strategies in parallel from a preset intervention strategy library (pre-built and stored various critical illness risk intervention plans, covering direct and indirect intervention strategies for different high-risk critical illness types and risk drivers) to generate an initial strategy set. Finally, the initial strategy set is optimized and integrated according to preset synergy effect rules to generate a cluster of related intervention strategies that are specifically adapted to the critical illness risk of employees.

[0071] Step S202: Generate a health intervention plan based on the associated intervention strategy cluster; the health intervention plan includes health advice, recommendations for physical examination items, and guidance on offline services.

[0072] The health advice can be tailored to an employee's individualized critical illness risk profile (e.g., high-risk critical illness types, risk contribution percentage) and job characteristics, providing targeted guidance on diet, exercise, rest, stress management, and other aspects.

[0073] Recommended medical examination items can be based on employees' personalized critical illness risk profiles (such as high-risk critical illness types and risk contribution ratios) and job characteristics, and can be targeted with quantitative indicators related to medical examinations, including the type of medical examination items, examination frequency, and reference range of indicators.

[0074] Offline service resource information can be key supporting information to fill the optimal candidate intervention plan sequence. Based on the individualized critical illness risk profile of employees and job characteristics, it can be integrated details of offline health service resources that can be implemented, including the addresses, appointment methods, and exclusive service items of cooperative medical institutions, health management centers, and physical examination institutions.

[0075] A health intervention plan can be a personalized solution that generates candidate intervention plans based on a cluster of related intervention strategies, selects the optimal candidate intervention plan sequence through a personalized fit score, and then fills in health advice content, physical examination item parameters, and offline service resource information.

[0076] Specifically, smart terminals can conduct personalized fit assessments based on clusters of related intervention strategies, combined with employee job characteristics and historical compliance preferences, to select the intervention plan with the best fit, fill in targeted health advice, physical examination item parameters and offline service resource information, and finally generate a personalized health intervention plan that includes health advice, physical examination item recommendations and offline service guidance.

[0077] Step S203: Collect data on the compliance of enterprise employees with the health intervention plan and obtain physiological indicator trajectory data of enterprise employees after a preset period.

[0078] Among them, compliance data can be data collected on the actual implementation of health intervention plans by enterprise employees, including information such as whether intervention measures were implemented, the frequency of implementation, and the degree of implementation.

[0079] Physiological indicator trajectory data can be a series of quantitative data related to the physiological functions of enterprise employees that are continuously collected within a preset period, covering changes in indicators such as blood pressure, blood sugar, and heart rate that are related to the risk of serious illness.

[0080] Specifically, the smart terminal collects data on the actual implementation of the health intervention plan by enterprise employees, including whether various intervention measures are implemented, the frequency of implementation, and the degree of implementation, as well as other information related to compliance. At the same time, after the preset period ends, it continuously acquires quantitative data on physiological indicators related to the risk of serious illness, such as blood pressure, blood sugar, and heart rate, and generates complete physiological indicator trajectory data.

[0081] Step S204: Based on the execution compliance data and physiological indicator trajectory data, generate a feedback sample dataset.

[0082] The feedback sample dataset can be a dataset that integrates data on employee compliance with health intervention plans with physiological indicator trajectory data after a preset period.

[0083] Specifically, the smart terminal cleans and standardizes the data collected on the compliance of the health intervention plan for enterprise employees (including whether it was implemented, frequency, degree, etc.) and the physiological indicator trajectory data after a preset period (including changes in serious disease-related indicators such as blood pressure, blood sugar, and heart rate), removing invalid and redundant information. Then, through data association mapping, a correspondence between the intervention implementation and the changes in physiological indicators is established. The compliance data and physiological indicator trajectory data are integrated and packaged according to a preset structure, and finally a feedback sample dataset containing complete correlation information can be directly used for model optimization.

[0084] Step S205: Adjust the weight parameters of the critical illness risk prediction model using the feedback sample dataset to obtain the updated critical illness risk prediction model.

[0085] Specifically, the smart terminal extracts features from the feedback sample dataset, obtains key feedback information, and fuses it with the original dynamic risk feature vector. Based on the fused feature information, it optimizes and adjusts the weight parameters of the critical illness risk prediction model, corrects the model prediction bias, and finally obtains the updated critical illness risk prediction model.

[0086] In this embodiment, the intelligent terminal generates a cluster of related intervention strategies and a personalized health intervention plan, collects data on adherence and physiological indicator trajectories to form feedback samples, optimizes the critical illness risk prediction model, realizes a dynamic closed loop of risk prediction and intervention, and improves the accuracy of enterprise health management.

[0087] In one embodiment, generating a health intervention plan based on a cluster of associated intervention strategies may include the following steps:

[0088] Step S301: Generate candidate intervention plans based on the associated intervention strategy clusters.

[0089] Among them, the candidate intervention plan can be a set of alternative plans with intervention potential that are initially generated based on the cluster of related intervention strategies and combined with the risk characteristics of serious illnesses of enterprise employees.

[0090] Specifically, the smart terminal combines the individual critical illness risk characteristics of enterprise employees to decompose, combine and initially adapt the various intervention strategies in the strategy cluster, transforming the abstract strategies into a specific solution framework with executable capabilities, and then generating a set of candidate intervention plans that include multiple potential intervention paths and cover different risk response scenarios.

[0091] Step S302: Based on the job characteristics and historical compliance preferences of the company's employees, the candidate intervention plans are individually assessed for their suitability.

[0092] Among them, the personalized fit score can be a quantitative indicator of the fit of candidate intervention plans calculated by combining the job characteristics and historical compliance preferences of enterprise employees through a specific formula.

[0093] Specifically, for the generated candidate intervention plans, the smart terminal uses a specific formula to calculate a personalized fit score, taking into account the job characteristics and historical compliance preferences of the company's employees.

[0094] Step S303: Based on the personalized fit score, select the optimal candidate intervention plan sequence, and fill in the health advice content, physical examination item parameters and offline service resource information into the optimal candidate intervention plan sequence to generate a health intervention plan.

[0095] Among them, the optimal candidate intervention plan sequence can be the combination of the best fit selected from the candidate intervention plans based on personalized fit scores and ranked by risk intervention priority.

[0096] Specifically, the smart terminal can select the most suitable combination of solutions from the candidate intervention plans based on the personalized suitability score, and form the optimal candidate intervention plan sequence according to the risk intervention priority. Then, based on the employee's personalized critical illness risk profile and job characteristics, the optimal candidate intervention plan sequence is filled with core information - targeted health advice on diet, exercise, work and rest and stress management, physical examination parameters that match the high-risk critical illness type, and offline service resource information including the address, appointment method and exclusive service of cooperative medical institutions and health management centers. Through information integration and plan improvement, a personalized health intervention plan that is both targeted and executable is finally generated.

[0097] In this embodiment, the smart terminal generates candidate plans by disassembling and combining related intervention strategies, selects the optimal sequence through personalized adaptability scoring, fills in core information to generate a health intervention plan, and realizes targeted and executable intervention for employees' serious illness risks.

[0098] In one embodiment, candidate intervention programs are individually scored for suitability based on the job characteristics and historical compliance preferences of employees, including:

[0099] Calculate the personalized fit score using the following formula:

[0100]

[0101] in, It is a personalized fit score. It is the total number of candidate intervention programs. It is the first The basic weights of each intervention plan It is the first The degree of match between the intervention plan and the job characteristics It is a preference coefficient based on historical compliance. It is the time decay factor. It is the first The implementation timeframe of an intervention plan.

[0102] Specifically, It can be used to quantify the fit between candidate intervention plans and target employees. The higher the score, the more the intervention plan fits the employee's job characteristics and execution habits. It can dynamically change according to the types of high-risk critical illnesses and the number of risk-driving factors among the company's employees; It can be the priority assignment of various intervention programs in the preset intervention strategy library (e.g., the basic weight of intervention plans that directly target core risk drivers is set to 0.8-1.0, and the weight of auxiliary intervention plans is set to 0.3-0.7). It can be the result of a matching degree calculation between the job characteristic database (including information such as job working hours, work environment, labor intensity, and operation procedures) and the requirements for the implementation of the intervention plan; It can reflect employees' past willingness and persistence in implementing similar intervention programs; It can be industry health management practice data and statistical results of the company's historical intervention effects. These are fixed empirical values, and the longer the intervention program is, the more obvious the attenuation effect is.

[0103] In this embodiment, the smart terminal uses the above formula to comprehensively consider factors such as the basic weight of the candidate intervention plan, job matching degree, historical compliance preference coefficient and execution time period, and quantitatively calculates the personalized suitability score, providing an objective basis for selecting the optimal intervention plan.

[0104] In one embodiment, adjusting the weight parameters of the critical illness risk prediction model using the feedback sample dataset to obtain an updated critical illness risk prediction model may include the following steps:

[0105] Step S401: Based on the feedback sample dataset, extract key feedback features using an incremental learning method.

[0106] Among them, key feedback features can be core features extracted from the feedback sample dataset that reflect the correlation between the effectiveness of health interventions and the risk of serious illnesses.

[0107] Specifically, the smart terminal uses an incremental learning method to conduct in-depth analysis of the feedback sample dataset, focusing on the correlation between the implementation of health interventions and changes in physiological indicators. It extracts core information that reflects the correlation between intervention effects and the risk of serious illnesses, including key feedback features such as the quality of intervention implementation and the fluctuation trend of physiological indicators, and removes irrelevant and redundant data.

[0108] Step S402: The key feedback features and the dynamic risk feature vector are fused to obtain an enhanced feature representation.

[0109] Among them, the enhanced feature representation can be the enhanced feature form obtained by fusing the key feedback features with the original dynamic risk feature vector.

[0110] Specifically, the smart terminal can extract key feedback features from the feedback sample dataset and perform feature fusion processing with the original dynamic risk feature vector generated based on multi-source heterogeneous datasets. By unifying the data format and aligning the feature dimensions, it integrates historical risk basic information and health intervention feedback correlation information, eliminates feature redundancy, strengthens core correlations, and finally generates enhanced feature representations.

[0111] Step S403: Using enhanced feature representation, the weight parameters of the critical illness risk prediction model are optimized and adjusted through the backpropagation algorithm to obtain the updated critical illness risk prediction model.

[0112] Specifically, the smart terminal can input the enhanced feature representation, which integrates historical risk information and intervention feedback information, into the critical illness risk prediction model. Through the backpropagation algorithm, it can trace the source of the deviation between the model prediction results and the actual health status, and calculate the adjustment magnitude and direction of the weight parameters at each level. Then, based on the adjustment magnitude and direction of the weight parameters at each level, iteratively optimize the model weights, correct the model prediction deviation, and finally obtain the updated critical illness risk prediction model.

[0113] In this embodiment, the intelligent terminal incremental learning extracts key feedback features from the feedback sample dataset, fuses them with the dynamic risk feature vector to generate enhanced feature representations, and then uses the backpropagation algorithm to optimize the weight parameters of the critical illness risk prediction model, continuously correcting prediction biases and significantly improving the model's accuracy and scenario adaptability in predicting the critical illness risk of enterprise employees.

[0114] In one embodiment, based on a personalized critical illness risk profile, risk-driving factors are matched from a pre-defined intervention strategy library to generate a cluster of related intervention strategies, which may include the following steps:

[0115] Step S501: Based on the risk contribution ratio, prioritize the high-risk critical illness types and their corresponding risk driving factors to obtain the core intervention targets.

[0116] Among them, the core intervention targets can be those that need to be focused on intervention, which are determined by prioritizing high-risk critical illness types and their corresponding risk drivers based on the risk contribution ratio in the personalized critical illness risk profile.

[0117] Specifically, smart terminals can use the risk contribution ratio in personalized critical illness risk profiles as the core basis, focus on high-risk critical illness types and their corresponding risk driving factors, construct scientific priority ranking rules, comprehensively consider the occurrence probability, potential impact and correlation strength of various critical illnesses and driving factors, classify and rank high-risk critical illness types and their corresponding risk driving factors, screen out core targets that play a key role in the critical illness risk of employees and require priority intervention, and finally identify core intervention targets.

[0118] Step S502: Using the core intervention target as the search key, retrieve direct intervention strategies and indirect intervention strategies in parallel from the preset intervention strategy library to generate an initial strategy set.

[0119] Among them, direct intervention strategies can be intervention programs designed for core intervention targets (specific high-risk critical illness types and corresponding risk drivers) that can directly act on the root causes of risk.

[0120] Indirect intervention strategies can be auxiliary intervention programs designed for core intervention targets. They do not directly target the root causes of serious disease risks, but rather provide support for direct intervention strategies by indirectly optimizing the living environment, adjusting behavioral habits, and strengthening health awareness, thereby reducing the probability of risk induction. Together with direct intervention strategies, they constitute the initial strategy set.

[0121] Specifically, the smart terminal can use the identified core intervention targets as key information for retrieval, access the preset intervention strategy library, start a parallel retrieval mechanism, and simultaneously match intervention plans for the core intervention targets: direct intervention strategies that can directly address the root causes of serious disease risks or indirect intervention strategies that provide support through optimizing the living environment and adjusting behavioral habits; after the retrieval is completed, effective strategies are integrated, duplicate or invalid content is eliminated, and an initial strategy set covering core interventions and auxiliary support is generated.

[0122] Step S503: Based on the preset synergy effect rules, optimize the initial strategy set to generate a cluster of related intervention strategies.

[0123] The preset synergy rules can be core rules pre-defined to optimize the initial strategy set. For example, the preset synergy rules can evaluate and screen the strategies in the initial strategy set from dimensions such as the correlation of strategy coordination (e.g., whether direct and indirect interventions are functionally complementary and logically unconflicted), the synergy of effects (e.g., whether the combination of two types of strategies can amplify the effectiveness of risk intervention, rather than simply adding them together), the feasibility of execution (e.g., whether the execution cost and time cost of the strategy combination are within the acceptable range for employees), and the risk adaptability (e.g., whether the strategy combination accurately matches the risk level and characteristics of the core intervention target). Conflicting, inefficient, or infeasible strategies are eliminated, and the strategy combination with the best synergy effect is retained and integrated.

[0124] Specifically, the smart terminal can comprehensively evaluate the direct and indirect intervention strategies in the initial strategy set based on preset synergy rules. Then, by verifying whether the strategies are functionally complementary and logically free from conflict, it determines whether the combination can amplify the intervention effect, checks whether the execution and time costs are controllable, confirms whether it accurately matches the characteristics of the core intervention target, eliminates conflicting, inefficient, and infeasible strategies, retains and integrates the strategy combination with the best synergy effect, and finally generates a cluster of related intervention strategies.

[0125] In this embodiment, the smart terminal determines the core intervention target based on the risk contribution ratio, retrieves direct and indirect intervention strategies in parallel from the preset intervention strategy library to generate an initial set, and then optimizes it through synergy effect rules to generate a cluster of targeted and highly efficient related intervention strategies, providing data support for the subsequent formulation of personalized health intervention plans.

[0126] In one embodiment, generating a dynamic risk feature vector based on a multi-source heterogeneous dataset may include the following steps:

[0127] Step S601: Based on the multi-source heterogeneous dataset, a fused dataset with a unified spatiotemporal benchmark is obtained through spatiotemporal consistency reconstruction.

[0128] Among them, the fused dataset can be a unified dataset obtained by reconstructing multi-source heterogeneous datasets in a spatiotemporal consistency manner.

[0129] Specifically, smart terminals can use spatiotemporal consistency reconstruction technology to address multi-source heterogeneous datasets (including static health records, dynamic physiological monitoring data, etc.) that have different sources, formats, and spatiotemporal misalignments. This will unify the time dimension benchmark and spatial correlation standard of the data, eliminate format conflicts, temporal deviations, and logical misalignments of data from different sources, integrate scattered health-related data resources, and generate a fusion dataset with a unified structure, consistent benchmarks, and complete information.

[0130] Step S602: Perform time-series data normalization on the fused dataset to obtain a standardized time-series matrix.

[0131] Among them, time series data normalization can be a standardization process performed on the fused dataset, which unifies the time granularity of the data, fills in missing values, corrects abnormal data, eliminates inconsistencies in the time series dimension, and transforms scattered time series data into a dataset with a normalized structure and uniform format.

[0132] A standardized time series matrix can be a structured data form obtained by normalizing the time series data of a fused dataset.

[0133] Specifically, smart terminals can perform time-series data normalization processing on the fused dataset obtained through spatiotemporal consistency reconstruction. By unifying the time granularity of the data (e.g., standardizing sampling intervals by day or hour), reasonable algorithms are used to fill in missing data values, identify and correct abnormal deviations, and completely eliminate format differences and data inconsistencies in the time-series dimension. This transforms the originally scattered and irregular time-series data into structured data with regular structure, unified format, and reliable values, ultimately generating a standardized time-series matrix.

[0134] Step S603: Based on the normalized temporal matrix, obtain the multi-scale temporal semantic tensor.

[0135] Among them, the multi-scale temporal semantic tensor can be a high-dimensional tensor structure generated by extracting temporal correlation features at different time granularities (such as hour, day, week) based on a standardized temporal matrix.

[0136] Specifically, smart terminals can use a standardized time-series matrix as a basis, setting multi-dimensional time granularities such as hours, days, and weeks. Time-series feature extraction algorithms can then be used to mine data correlation patterns and trends at different granularities, including short-term physiological indicator fluctuations, medium-term behavioral pattern correlations, and long-term risk accumulation characteristics. These multi-scale time-series correlation features are then dimensionally fused and reconstructed into tensors, integrating them to generate a high-dimensional tensor structure containing multi-dimensional, multi-time-scale semantic information.

[0137] Step S604: Based on the multi-scale temporal semantic tensor, static health record data and organizational environment data are mapped into static embedding vectors through static feature embedding.

[0138] Among them, the static embedding vector can be a low-dimensional dense vector generated by mapping non-quantitative information (such as past medical history, job type, office environment conditions, etc.) in static health record data with non-quantitative information in organizational environment data through a static feature embedding algorithm based on multi-scale temporal semantic tensor.

[0139] Specifically, smart terminals can quantify and normalize employees' static health record data (such as medical history and genetic characteristics) and organizational environment data (such as job category and work environment rating), and then project them into a latent space compatible with dynamic temporal features through an embedding layer to generate low-dimensional dense static embedding vectors.

[0140] Step S605: Based on the multi-scale temporal semantic tensor and the static embedding vector, heterogeneous fusion features are obtained.

[0141] Among them, heterogeneous fusion features can be comprehensive features obtained by cross-modal fusion processing of multi-scale temporal semantic tensors and static embedding vectors.

[0142] Specifically, intelligent terminals can establish a mapping between temporal dynamic information and static attribute information by unifying the data dimensions and numerical scales of multi-scale temporal semantic tensors and static embedding vectors, thereby eliminating the fusion barriers caused by feature heterogeneity, deeply integrating the dynamic risk change patterns and static risk basic attributes, extracting complementary correlation information between multi-scale temporal semantic tensors and static embedding vectors, and finally generating heterogeneous fusion features that are both comprehensive and correlated.

[0143] Step S606: Integrate the heterogeneous fusion features to obtain a dynamic risk feature vector.

[0144] Specifically, smart terminals can filter out key dimensions that are significantly related to the risk of serious illnesses and eliminate redundant information; then, a multilayer perceptron is used for nonlinear transformation, and finally, the feature scale is unified to a standard distribution through normalization processing to generate a stable and usable dynamic risk feature vector.

[0145] In this embodiment, the intelligent terminal generates a comprehensive and dynamic risk feature vector by sequentially performing spatiotemporal consistency reconstruction, time-series data normalization, multi-scale feature extraction, static feature embedding, heterogeneous feature fusion, and feature integration on multi-source heterogeneous datasets. This provides high-quality and standardized core feature support for subsequent accurate prediction of critical illness risks.

[0146] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0147] Based on the same inventive concept, this application also provides a device for implementing the aforementioned AI-assisted enterprise employee critical illness risk prediction device. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more AI-assisted enterprise employee critical illness risk prediction device embodiments provided below can be found in the limitations of the AI-assisted enterprise employee critical illness risk prediction method described above, and will not be repeated here.

[0148] In one exemplary embodiment, such as Figure 2 As shown, an artificial intelligence-assisted critical illness risk prediction device 700 for enterprise employees is provided, comprising:

[0149] The data acquisition module 701 is used to acquire multi-source heterogeneous datasets of enterprise employees; the multi-source heterogeneous datasets include static health record data, dynamic physiological monitoring data, job-specific work behavior data, and organizational environment data.

[0150] The risk feature vector generation module 702 is used to generate dynamic risk feature vectors based on multi-source heterogeneous datasets.

[0151] The risk profile generation module 703 is used to input dynamic risk feature vectors into the trained critical illness risk prediction model to obtain a personalized critical illness risk profile. The personalized critical illness risk profile includes the probability of critical illness occurrence, high-risk critical illness types, and risk contribution ratio of enterprise employees.

[0152] In one embodiment, the device further includes:

[0153] The intervention strategy cluster generation module is used to match risk-driving factors from a pre-set intervention strategy library based on a personalized critical illness risk profile, and generate related intervention strategy clusters.

[0154] The intervention plan generation module is used to generate health intervention plans based on related intervention strategy clusters; the health intervention plans include health advice, physical examination item recommendations, and offline service guidance;

[0155] The feedback data collection module is used to collect data on the compliance of employees with the health intervention plan and to obtain the physiological indicator trajectory data of employees after a preset period.

[0156] The feedback sample dataset generation module is used to generate feedback sample datasets based on execution compliance data and physiological indicator trajectory data.

[0157] The optimization and update module is used to adjust the weight parameters of the critical illness risk prediction model using the feedback sample dataset, so as to obtain the updated critical illness risk prediction model.

[0158] In one embodiment, the intervention plan generation module generates a health intervention plan based on a cluster of associated intervention strategies, including:

[0159] The intervention plan generation unit is used to generate candidate intervention plans based on related intervention strategy clusters;

[0160] The fit scoring unit is used to give personalized fit scores to candidate intervention plans based on the job characteristics and historical compliance preferences of employees.

[0161] The intervention plan integration and generation unit is used to select the optimal candidate intervention plan sequence based on the personalized fit score, and fill in the optimal candidate intervention plan sequence with health advice content, physical examination item parameters and offline service resource information to generate a health intervention plan.

[0162] In one embodiment, the fit scoring unit performs personalized fit scoring on candidate intervention plans based on the job characteristics and historical compliance preferences of the company's employees, including:

[0163] Calculate the personalized fit score using the following formula:

[0164]

[0165] in, It is a personalized fit score. It is the total number of candidate intervention programs. It is the first The basic weights of each intervention plan It is the first The degree of match between the intervention plan and the job characteristics It is a preference coefficient based on historical compliance. It is the time decay factor. It is the first The implementation timeframe of an intervention plan.

[0166] In one embodiment, the optimization and update module adjusts the weight parameters of the critical illness risk prediction model using the feedback sample dataset to obtain an updated critical illness risk prediction model, including:

[0167] The feedback feature extraction unit is used to extract key feedback features based on the feedback sample dataset using an incremental learning method.

[0168] The feature representation generation unit is used to fuse key feedback features with dynamic risk feature vectors to obtain enhanced feature representations;

[0169] The weight optimization unit is used to optimize and adjust the weight parameters of the critical illness risk prediction model by using enhanced feature representation and backpropagation algorithm to obtain an updated critical illness risk prediction model.

[0170] In one embodiment, the intervention strategy cluster generation module, based on a personalized critical illness risk profile, matches risk-driving factors from a pre-set intervention strategy library to generate associated intervention strategy clusters, including:

[0171] The pre-target determination unit is used to prioritize high-risk critical illness types and their corresponding risk drivers based on the risk contribution ratio, thereby obtaining core intervention targets.

[0172] The set generation unit is used to retrieve direct and indirect intervention strategies in parallel from a preset intervention strategy library using the core intervention target as the retrieval key, and generate an initial strategy set.

[0173] The strategy cluster optimization unit is used to optimize the initial strategy set based on preset synergy rules to generate a cluster of related intervention strategies.

[0174] In one embodiment, the risk feature vector generation module generates dynamic risk feature vectors based on multi-source heterogeneous datasets, including:

[0175] The fusion dataset generation unit is used to obtain a fusion dataset with a unified spatiotemporal benchmark based on multi-source heterogeneous datasets through spatiotemporal consistency reconstruction.

[0176] The matrix generation unit is used to perform time-series data normalization on the fused dataset to obtain a standardized time-series matrix.

[0177] Semantic tensor generation unit, used to obtain multi-scale temporal semantic tensors based on normalized temporal matrices;

[0178] The vector mapping unit is used to map static health record data and organizational environment data into static embedding vectors based on multi-scale temporal semantic tensors through static feature embedding.

[0179] The heterogeneous fusion feature generation unit is used to obtain heterogeneous fusion features based on multi-scale temporal semantic tensors and static embedding vectors;

[0180] The feature vector integration unit is used to integrate heterogeneous fused features to obtain dynamic risk feature vectors.

[0181] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the artificial intelligence-assisted method for predicting the risk of serious illnesses of enterprise employees as described above.

[0182] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0183] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0184] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for predicting the risk of serious illness among enterprise employees based on artificial intelligence, characterized in that, The method includes: Acquire multi-source heterogeneous datasets of enterprise employees; the multi-source heterogeneous datasets include static health record data, dynamic physiological monitoring data, job-specific work behavior data, and organizational environment data; Based on the aforementioned multi-source heterogeneous dataset, a dynamic risk feature vector is generated; The dynamic risk feature vector is input into the trained critical illness risk prediction model to obtain a personalized critical illness risk profile; the personalized critical illness risk profile includes the probability of critical illness occurrence, high-risk critical illness types, and risk contribution ratio of enterprise employees.

2. The method according to claim 1, characterized in that, The method further includes: Based on the personalized critical illness risk profile, risk driving factors are matched from the preset intervention strategy library to generate a cluster of related intervention strategies; Based on the aforementioned cluster of related intervention strategies, a health intervention plan is generated; the health intervention plan includes health advice, recommendations for physical examination items, and guidance on offline services. Collect data on employee compliance with the health intervention plan and obtain physiological indicator trajectory data of employees after a preset period. Based on the execution compliance data and the physiological indicator trajectory data, a feedback sample dataset is generated; The weight parameters of the critical illness risk prediction model are adjusted using the feedback sample dataset to obtain the updated critical illness risk prediction model.

3. The method according to claim 2, characterized in that, The step of generating a health intervention plan based on the associated intervention strategy cluster includes: Based on the aforementioned cluster of related intervention strategies, candidate intervention plans are generated; Based on the job characteristics and historical compliance preferences of the company's employees, the candidate intervention plans are individually assessed for suitability. Based on the personalized fit score, the optimal candidate intervention plan sequence is selected, and health advice content, physical examination item parameters and offline service resource information are filled into the optimal candidate intervention plan sequence to generate the health intervention plan.

4. The method according to claim 3, characterized in that, The process of personalizing the suitability score of the candidate intervention plan based on the job characteristics and historical compliance preferences of the company's employees includes: Calculate the personalized fit score using the following formula: ； in, It is a personalized fit score. It is the total number of candidate intervention programs. It is the first The basic weights of each intervention plan It is the first The degree of matching between the intervention plan and the characteristics of the job. It is a preference coefficient based on historical compliance. It is the time decay factor. It is the first The implementation period of an intervention plan.

5. The method according to claim 2, characterized in that, The step of adjusting the weight parameters of the critical illness risk prediction model using the feedback sample dataset to obtain the updated critical illness risk prediction model includes: Based on the aforementioned feedback sample dataset, key feedback features are extracted using an incremental learning method. The key feedback features are fused with the dynamic risk feature vector to obtain an enhanced feature representation. Using the enhanced feature representation, the weight parameters of the critical illness risk prediction model are optimized and adjusted through the backpropagation algorithm to obtain the updated critical illness risk prediction model.

6. The method according to claim 2, characterized in that, Based on the personalized critical illness risk profile, risk driving factors are matched from a pre-set intervention strategy library to generate a cluster of related intervention strategies, including: Based on the risk contribution ratio, the high-risk critical illness types and their corresponding risk driving factors are prioritized to obtain the core intervention targets. Using the core intervention target as the search key, direct intervention strategies and indirect intervention strategies are retrieved in parallel from the preset intervention strategy library to generate an initial strategy set; Based on preset synergy effect rules, the initial strategy set is optimized to generate the associated intervention strategy cluster.

7. The method according to any one of claims 1 to 6, characterized in that, The generation of dynamic risk feature vectors based on the multi-source heterogeneous dataset includes: Based on the aforementioned multi-source heterogeneous dataset, a fusion dataset with a unified spatiotemporal benchmark is obtained through spatiotemporal consistency reconstruction. The fused dataset is subjected to time-series data normalization to obtain a standardized time-series matrix; Based on the standardized temporal matrix, a multi-scale temporal semantic tensor is obtained; Based on the multi-scale temporal semantic tensor, the static health record data and the organizational environment data are mapped into static embedding vectors through static feature embedding. Based on the multi-scale temporal semantic tensor and the static embedding vector, heterogeneous fusion features are obtained; The heterogeneous fusion features are integrated to obtain the dynamic risk feature vector.

8. A device for predicting the risk of serious illness for enterprise employees based on artificial intelligence, characterized in that, The device includes: The data acquisition module is used to acquire multi-source heterogeneous datasets of enterprise employees; the multi-source heterogeneous datasets include static health record data, dynamic physiological monitoring data, job-specific work behavior data, and organizational environment data; The risk feature vector generation module is used to generate dynamic risk feature vectors based on the multi-source heterogeneous dataset. The risk profile generation module is used to input the dynamic risk feature vector into the trained critical illness risk prediction model to obtain a personalized critical illness risk profile; the personalized critical illness risk profile includes the probability of critical illness occurrence, high-risk critical illness types, and risk contribution ratio of enterprise employees.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images