Health service recommendation scheme generation method and device, equipment and medium

By acquiring and analyzing multi-dimensional claims health datasets, and using graph neural network technology to generate personalized health service recommendations, this solves the problem of low matching degree caused by traditional reliance on paper-based assessments. It achieves the integration of insurance claims and health services, providing higher matching degree and dual value.

CN120913740APending Publication Date: 2025-11-07CHINA PING AN LIFE INSURANCE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511022330.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-07-23
Publication Date
2025-11-07

AI Technical Summary

Technical Problem

In existing technologies, health service recommendations mainly rely on paper-based personal data submitted by customers, which lacks objective and comprehensive health indicators, resulting in a low degree of matching between the recommended solutions and customers.

Method used

By acquiring the health dataset of target customers' claims, including basic profile data, claims records, and health monitoring data, we can conduct claims fraud detection and health risk assessment, use graph neural network technology to generate personalized health service recommendations, and integrate multi-dimensional data for precise analysis and matching.

Benefits of technology

It integrates insurance claims with health services, providing customers with the dual value of "risk protection + health management," and significantly improving the matching degree between recommended health service plans and customers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120913740A_ABST
    Figure CN120913740A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of data processing, and discloses a health service recommendation scheme generation method and device, equipment and a medium, and a complete customer health portrait data set is constructed by integrating medical records, health monitoring data, claim settlement records and other multi-dimensional data of target customers. And based on the data set, accurate claim settlement fraud detection is carried out, and the customer health risk level is quantified. And then, basic portrait data of the customer and a double-score result (claim settlement fraud score and health risk score) are jointly input into a graph neural network model to form a complete closed loop of data acquisition-intelligent analysis-risk control-service optimization, and intelligent generation of a personalized health service scheme is realized. Through fusion of insurance claim settlement and health service, double values of'risk guarantee + health management 'are provided for the customer, so that the integrating degree of the finally output health service scheme and the actual demand of the customer is remarkably improved, and the matching degree of the health service recommendation scheme and the customer is effectively improved.
Need to check novelty before this filing date? Find Prior Art

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a health service recommendation scheme generation method and device, equipment and medium. BACKGROUND

[0002] With the development of health technology, intelligent health management has become the focus of attention in various industries. By tailoring health service recommendation schemes for customers, more convenient and accurate health management services can be provided, thereby effectively maintaining customer health status and reducing the risk of disease. In related technologies, health service recommendations mainly rely on paper personal information submitted by customers as the basis for evaluation. However, this single-dimensional data evaluation method has obvious limitations, lacks objective and comprehensive health indicators, and results in low matching degree of recommended schemes and customers. SUMMARY

[0003] The present application provides a health service recommendation scheme generation method, device, computer equipment and medium to solve the technical problem that only paper personal information submitted by customers is used as the basis for health service recommendation evaluation, lacking objective and comprehensive health indicators, resulting in low matching degree of recommended schemes and customers.

[0004] In a first aspect, a health service recommendation scheme generation method is provided, comprising:

[0005] Obtaining a claim health data set of a target customer, wherein the claim health data set includes basic portrait data, claim records, medical records and health monitoring data;

[0006] Based on the claim health data set, performing claim fraud detection and health risk assessment on the target customer to obtain a claim fraud score and a health risk score of the target customer;

[0007] Based on the claim fraud score, the health risk score and the claim health data set, a health service recommendation scheme for the target customer is generated through a graph neural network technology.

[0008] In a second aspect, a health service recommendation scheme generation device is provided, comprising:

[0009] An obtaining module is configured to obtain a claim health data set of a target customer, wherein the claim health data set includes basic portrait data, claim records, medical records and health monitoring data;

[0010] A first generation module is configured to perform claim fraud detection and health risk assessment on the target customer based on the claim health data set to obtain a claim fraud score and a health risk score of the target customer;

[0011] The second generation module is configured for generating a health service recommendation scheme for the target customer based on the claim fraud score, the health risk score and the claim health dataset by using a graph neural network technology.

[0012] In a third aspect, a computer device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the method for generating the health service recommendation scheme when executing the computer program.

[0013] In a fourth aspect, a computer readable storage medium is provided, which stores a computer program, and the computer program implements the steps of the method for generating the health service recommendation scheme when executed by a processor.

[0014] In the method, device, computer device and storage medium for generating the health service recommendation scheme, the multi-dimensional data such as the medical record, the health monitoring data and the claim record of the target customer are integrated to construct a complete customer health portrait dataset. Based on the dataset, the claim fraud is accurately detected, and the health risk level of the customer is quantified. Then, the basic portrait data of the customer and the double-score results (the claim fraud score and the health risk score) are input into a graph neural network model to form a complete closed loop of "data collection-intelligent analysis-risk control-service optimization", and the intelligent generation of the personalized health service scheme is realized. Compared with the traditional mode in which the recommendation is only based on the paper materials submitted by the customer and the recommendation is manually evaluated, the fusion of the insurance claim and the health service is realized, and the double value of "risk protection + health management" is provided for the customer. Through the cross analysis and intelligent matching of the multi-dimensional data, the matching degree of the health service scheme and the actual demand of the customer is significantly improved, and the matching degree of the health service recommendation scheme and the customer is effectively improved. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the description of the embodiments of the present application. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative labor.

[0016] Figure 1 is a schematic diagram of an application environment of the method for generating the health service recommendation scheme in an embodiment of the present application;

[0017] Figure 2 is a flowchart of the method for generating the health service recommendation scheme in an embodiment of the present application;

[0018] Figure 3 is Figure 2is a specific embodiment flowchart of step S10 in the method;

[0019] Figure 4 is Figure 2 is a specific embodiment flowchart of step S20 in the method;

[0020] Figure 5 is Figure 2 is a specific embodiment flowchart of step S30 in the method;

[0021] Figure 6 is a structural diagram of a health service recommendation scheme generation device in an embodiment of the present application;

[0022] Figure 7 is a structural diagram of a computer device in an embodiment of the present application;

[0023] Figure 8 is another structural diagram of a computer device in an embodiment of the present application. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0025] The health service recommendation scheme generation method provided by the embodiments of the present application can be applied in, for example, Figure 1In an application environment of the application, a client communicates with a server through a network. The server obtains a claim health data set of a target customer; based on the claim health data set, the target customer is subjected to claim fraud detection and health risk assessment to obtain a claim fraud score and a health risk score of the target customer; and based on the claim fraud score, the health risk score and the claim health data set, a health service recommendation scheme of the target customer is generated through a graph neural network technology. By integrating the medical records, health monitoring data, claim records and other multi-dimensional data of the target customer, a complete customer health portrait data set is constructed. Based on the data set, accurate claim fraud detection and quantification of the customer health risk level are performed. Subsequently, the customer basic portrait data and the double score results (claim fraud score and health risk score) are jointly input into the graph neural network model to form a complete closed loop of "data collection-intelligent analysis-risk control-service optimization", and intelligent generation of a personalized health service scheme is realized. Compared with the traditional mode of relying on the paper materials submitted by the customer for manual assessment and recommendation, the application realizes the integration of insurance claim and health service, and provides the customer with double values of "risk protection + health management". Through cross analysis and intelligent matching of multi-dimensional data, the matching degree of the finally output health service scheme and the actual needs of the customer is significantly improved, and the matching degree of the health service recommendation scheme and the customer is effectively improved. The client can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers and portable wearable devices. The server can be realized by an independent server or a server cluster composed of multiple servers. The application will be described in detail through specific embodiments.

[0026] Please refer to Figure 2 as shown, Figure 2 A flowchart of a health service recommendation scheme generation method provided by an embodiment of the application is shown in the figure, which includes the following steps:

[0027] S10: Obtain a claim health data set of a target customer, wherein the claim health data set includes basic portrait data, claim records, medical records and health monitoring data.

[0028] In this step, the claim health data set of the target customer is systematically collected, which comprehensively records the detailed information of the health status, insurance claim history and medical records of the target customer. The multi-dimensional key data specifically includes: target customer basic portrait data, covering age, gender, occupation, lifestyle and other data that can outline the basic profile of the target customer; complete claim records, including claim frequency, claim amount, accident detail description and accident image data; medical records of the target customer, covering basic diagnosis and treatment information and treatment process data; dynamic health monitoring data of the target customer, such as blood pressure, blood glucose, heart rate, body fat rate and other clinical parameters reflecting the real-time physiological state of the target customer, and physical examination data of the target customer.

[0029] In actual application scenarios, current health care service recommendation generally adopts a static templating scheme, the recommendation logic is fixed, and cannot integrate customer historical claim characteristics (such as high-frequency claim diseases), is separated from dynamic health data (such as wearable device monitoring indicators, medication compliance), lacks risk prediction ability, and cannot adjust service strategies in real time according to changes in customer health status. This recommendation mode leads to a significant deviation between services and actual customer needs. The present application constructs a multi-modal data fusion system, integrates medical records, device monitoring, claim history and other multi-dimensional data sources to form a complete evidence chain for evaluating customer health risks and claim fraud, realizes the deep integration of claim and health care scenarios, and thus realizes personalized health care service recommendation.

[0030] In an embodiment of the present application, as shown in Figure 3 The specific data acquisition scheme is provided, S10, that is, the claim health data set of the target customer is acquired, specifically including the following steps S11-S12:

[0031] S11: Acquire multi-modal data of the target customer, wherein the multi-modal data includes structured data and unstructured data.

[0032] In this step, a variety of data collection means are used to widely collect multi-modal data of the target customer. These multi-modal data cover structured data and unstructured data. Structured data such as health records, etc.; unstructured data such as medical accident images, etc.

[0033] In actual application scenarios, in the medical insurance field, the deep integration of life insurance individual claim business and health care and old-age care scenarios has become the mainstream trend of industry development. The current collaborative mode faces significant data integration challenges: core business data (including medical claim records, accident proof files, etc.) and dynamic health data (such as wearable device monitoring indicators, chronic disease management archives, etc.) are scattered in independent business systems, forming a serious data barrier. This fragmented data management mode not only causes customers to need to repeatedly submit paper proof materials, but also causes low business processing efficiency and potential underwriting risks. Therefore, the present application constructs a multi-source data acquisition system, realizes the real-time collection of structured data (diagnosis code, test indicators, etc.) and unstructured data (medical images, rehabilitation records, etc.) through the interface of medical HIS system, customer mobile application, Internet of Things sensor equipment, intelligent wearable device and old-age care institution monitoring platform, etc. Channels, realize multi-source data fusion, effectively break the traditional data barrier.

[0034] S12: Integrate multi-modal data based on a message queue to generate a claim health data set.

[0035] In this step, real-time and efficient integration of multi-source data is achieved through a distributed message queue, generating a high-quality claim settlement health data set.

[0036] In practical application scenarios, a multi-source heterogeneous data collection system is built based on the Java technology stack, and a Kafka distributed message queue is used to realize real-time access of medical records (such as electronic medical records, diagnosis reports), accident proofs (such as image data, third-party appraisal documents), and wearable device data (such as dynamic heart rate, motion trajectory, etc.) and other heterogeneous data sources. The system realizes high-performance network communication through JavaNIO, encapsulates the standardized interface of Kafka Producer and Consumer based on the Spring framework, realizes asynchronous batch processing combined with thread pool and CompletableFuture, and distributes the data to the downstream processing module through partition strategy after serialization to JSON format by Jackson. For unstructured data (DICOM images, rehabilitation video records), a Cassandra distributed database is used, and multiple copy writing and consistent hash storage are realized through Java Driver; for structured data (such as electronic claim settlement forms, customer health records), multi-level indexing is built through Java REST client based on Elasticsearch, supporting real-time retrieval and aggregation analysis, ensuring data traceability and fast query capability. Further, at the Java technology implementation level, the Leader Epoch mechanism and fine-grained offset control strategy of Kafka are used to ensure the strong consistency of cross-system data synchronization. The producer end verifies the target partition Leader Epoch state in real time through Java API, ensuring the accuracy of message routing; the consumer end realizes double offset management based on KafkaConsumer interface, maintaining real-time consumption position in memory and persisting the confirmation offset to the consumer_offsets storage topic through asynchronous threads at regular intervals. For medical data storage scenarios, the lightweight transaction (LWTPaxos protocol) of Cassandra and the version conflict detection (version_type = external) of Elasticsearch are used to realize the consistency of double-database writing, ensuring the association verification of medical images and structured claim settlement records during storage.

[0037] In the above manner, a real-time processing engine based on multi-modal data fusion is constructed, and through a distributed stream processing architecture, medical records, device monitoring, and claim settlement records are synchronized at the millisecond level. The full-link integrity from data collection, transmission to storage is guaranteed, providing a high-trust multi-source evidence chain for health service recommendation.

[0038] S20: performing claim fraud detection and health risk assessment on the target customer based on the claim health data set to obtain a claim fraud score and a health risk score of the target customer.

[0039] In this step, the target customer is subjected to intelligent analysis in dual dimensions of claim fraud detection and health risk assessment based on the claim health data set. In the claim fraud detection link, the consistency between the accident description and the accident image of the claim record and the authenticity of the accident image are deeply mined to systematically assess the possibility of fraud behavior in each claim event and give a quantitative fraud probability score. In the health risk assessment dimension, a machine learning model is used to integrate the static attributes (such as age, occupation) of the target customer, disease history, claim frequency and dynamic physiological indicators (such as blood pressure, blood sugar, etc.) to obtain an accurate health risk coefficient.

[0040] Through the above-mentioned manner, the claim fraud score and the health risk level score of the target customer are accurately calculated, and the two quantitative core evaluation indexes will be used as important decision-making basis for subsequent development of personalized health management service recommendation scheme customization.

[0041] In an embodiment of the present application, as shown in Figure 4 a specific claim fraud detection and health risk assessment scheme is provided, and in S20, the target customer is subjected to claim fraud detection and health risk assessment based on the claim health data set to obtain a claim fraud score and a health risk score of the target customer, which specifically includes the following steps S21-S25:

[0042] S21: obtaining first accident features in the accident description text through natural language processing technology.

[0043] S22: obtaining second accident features in the accident image through image processing method.

[0044] S23: generating a claim fraud score of the target customer based on the first accident features, the second accident features and the claim record.

[0045] For steps S21-S23, through multi-modal claim record analysis, structured claim record, unstructured accident description text and accident scene image are integrated to construct a full-dimensional evaluation system. Based on natural language processing (NLP) technology, key semantic features (i.e. first accident features) are accurately extracted from the text description, and computer vision (CV) algorithm is used to analyze visual evidence features (i.e. second accident features) from image data. By fusing three types of data sources (text features, image features and original claim records), a machine learning model is used to dynamically generate a claim fraud risk assessment score of the target customer, and a scientific and quantitative anti-fraud decision is realized.

[0046] In an embodiment of the present application, a specific claim fraud score calculation scheme is provided, i.e. in S23, based on the first accident features, the second accident features and the claim records, a claim fraud score of the target customer is generated, specifically including steps S231-S233:

[0047] S231: Based on the first accident features and the second accident features, the intersection-over-union algorithm is used to calculate the intersection-over-union value between the accident description text and the accident image.

[0048] In this step, the first accident features and the second accident features are integrated, and the intersection-over-union algorithm is used to verify the consistency between the accident description text and the accident image in depth, and the intersection-over-union value is accurately calculated.

[0049] In the above manner, the consistency between the description text and the image is detected, which helps to grasp the authenticity of the accident from the matching degree of multi-modal information.

[0050] S232: Based on the claim records, the claim frequency and claim amount of the target customer are determined.

[0051] In this step, the claim records are comprehensively analyzed, and the claim frequency and claim amount of the target customer are accurately determined. These two key indicators can reflect the regularity and tendency of the target customer in claim behavior.

[0052] S233: Based on the intersection-over-union value, the claim frequency and the claim amount, a claim fraud score of the target customer is generated by a logistic regression model.

[0053] In this step, the intersection-over-union value, the claim frequency and the claim amount are used as important input parameters, and the logistic regression model is used for deep analysis and calculation to generate the claim fraud score of the target customer.

[0054] In the above manner, the calculation of the claim fraud score provides a scientific and objective reference for claim review, and helps to more accurately identify the claim fraud risk.

[0055] In an embodiment of the present application, a specific accident image authenticity detection scheme is provided, that is, after obtaining the claim health data set of the target customer, the following steps are specifically included:

[0056] Generating a fake image sample through a generative adversarial network model;

[0057] Training the residual network by taking the fake image sample and the historical claim accident image sample as training data to obtain a fake feature recognition model;

[0058] Extracting fake features of the accident image through the fake feature recognition model to obtain the authenticity probability of the accident image;

[0059] If the authenticity probability is less than or equal to a preset authenticity threshold, marking the claim event corresponding to the accident image as an abnormal event;

[0060] Generating an alarm information based on the abnormal event and sending the alarm information to the terminal of the target personnel.

[0061] In this embodiment, the powerful generation capability of the generative adversarial network (GAN) model is first used to synthesize fake image samples with high simulation degree, effectively expanding the anti-fraud training data set. Then the generated fake samples and historical real claim images are input into the residual network (ResNet) for adversarial training, and the subtle differences between real and fake images are accurately captured through deep feature learning to build a high-precision fake feature recognition model. Based on the model, the accident image is analyzed in multiple dimensions, and a quantitative image authenticity probability score is output. When the score is lower than the preset risk threshold, the system automatically marks the claim case as a high-risk abnormal event and generates a visual alarm report containing key risk indicators, which is pushed to the reviewer's work terminal in real time, realizing intelligent identification and rapid response of fraud risk.

[0062] In the above manner, the traces of tampering and abnormal features of the accident image are intelligently identified, and the proactive prevention and precise interception of fraud risk are realized.

[0063] In practical application scenarios, in the Java technology implementation, a deep forgery detection system is built based on the DL4J framework: first, a pre-trained GAN model (such as CycleGAN or StyleGAN) is used to generate synthetic medical image samples for training an image authenticity classifier with a ResNet-50 architecture. In the model inference stage, after inputting the accident scene image, the network outputs a probability value (interval [0, 1]) of authenticity based on the Sigmoid function. When the probability value is lower than the dynamic threshold (default 0.7), the system automatically triggers the manual review workflow. At the same time, a computer vision verification module is integrated, which uses OpenCV to extract the ROI region of the medical image (such as the fracture positioning frame), and calculates the spatial consistency score of the image region and the claim text description through an improved intersection over union algorithm. When the matching degree is lower than the preset threshold (such as 0.5), it is upgraded to a high-risk case. Finally, through an ensemble learning method (such as XGBoost weighted fusion), 12-dimensional features such as image authenticity probability and spatial matching degree are nonlinearly combined to generate a comprehensive fraud score, effectively improving the fraud identification accuracy.

[0064] S24: Determine the disease history of the target customer based on the medical records.

[0065] S25: Based on the basic portrait data, disease history and health monitoring data of the target customer, calculate the health risk score of the target customer through the gradient boosting tree model.

[0066] For steps S24-S25, to accurately quantify the health risk level of the target customer, first, the medical records are deeply mined and analyzed to accurately identify the target customer's past disease history and diagnosis and treatment trajectory. Subsequently, multi-dimensional data sources (including customer portrait attributes, disease history and real-time health monitoring indicators) are integrated, and a gradient boosting tree (GBDT) algorithm is used for high-precision modeling analysis. Through feature importance weighting and nonlinear relationship mining, the final output is the customer's personalized health risk dynamic score.

[0067] In practical application scenarios, a medical text intelligent analysis engine is built based on the Deeplearning4j (DL4J) framework, a domain pre-training model (BioBERT / PubmedBERT) weight file is loaded, and deep semantic parsing of the claim text is implemented through a Java Native API: including clinical entity recognition (ICD code extraction, drug dose detection), diagnosis and treatment event relation extraction (surgery-complication association), and clause compliance determination (classification model based on Attention mechanism). In the image forensics layer, a multi-stage processing is implemented by integrating the OpenCV-JavaCV cross-platform library: the original image is subjected to frequency domain analysis (FFT transformation), local anomaly detection (LBP texture matching), and deep forgery recognition based on a generative adversarial network (Java-JNI call integrating a StyleGAN2 discriminator module). Through the construction of a multi-modal reasoning workflow (Apache Beam), the text parsing confidence and image anomaly score are input into a Drools rule engine to perform cross-modal consistency checking based on a Bayesian network, and finally an interpretable fraud risk assessment report is generated (including evidence chain visualization and decision path tracing). After building a health risk assessment model based on XGBoost4j, multi-dimensional features are input: basic attributes (age, gender, BMI); medical history (previous medical history, family genetic diseases); claim features (annual claim frequency); dynamic health indicators (blood glucose fluctuation rate for 30 consecutive days, blood pressure diurnal difference). The model output adopts a three-class logic (low / medium / high risk) with a confidence score.

[0068] S30: Based on the claim fraud score, health risk score, and claim health data set, a health service recommendation plan for the target customer is generated through graph neural network technology.

[0069] In this step, the claim fraud score, health risk score, and claim health data set of the target customer are fully utilized, and powerful graph neural network technology is introduced. Graph neural networks can capture complex relationships and potential patterns in data. They perform deep analysis and learning of these key information, and mine the internal relationship between customer characteristics and various health services, so as to tailor a health service recommendation plan for each target customer that fits their actual situation. The recommendation plan covers personalized health rehabilitation experts / teams, professional health management courses, targeted exercise plans, nutrition plans, etc., aiming to provide a comprehensive and precise health service experience for the template customer, helping the customer to improve their health level and reduce health risks.

[0070] In an embodiment of the present application, as Figure 5As shown, a specific health service recommendation scheme is provided, S30, that is, based on the claim fraud score, health risk score and claim health data set, a health service recommendation scheme for the target customer is generated through the graph neural network technology, specifically including the following steps S31-S33:

[0071] S31: Construct a multi-dimensional relationship graph based on the graph neural network, wherein the multi-dimensional relationship graph includes the potential relationship between the customer, the health service item and the health and wellness team.

[0072] In this step, a multi-dimensional relationship graph is constructed based on the graph neural network (GNN), which deeply models the complex interaction network between the customer-health service item-health and wellness team. Through node feature learning and edge relationship weight calculation, this graph accurately captures the potential association patterns between entities, providing an interpretable relationship reasoning basis for intelligent recommendation and risk prediction.

[0073] S32: Input the claim fraud score, health risk score, basic portrait data and health monitoring data as node features into the multi-dimensional relationship graph to obtain at least one target health service item and target health service team that meets the target customer.

[0074] In this step, the pre-calculated claim fraud score and health risk score are combined with the customer's basic portrait and real-time health monitoring data as multi-dimensional node features embedded in the pre-constructed relationship graph. Based on the message passing and feature aggregation mechanism of the graph neural network (GNN), deep representation learning is performed on heterogeneous node information, and finally cross-modal feature fusion is realized through the graph attention network (GAT) to accurately output the health service item and health and wellness team recommendation combination with the highest matching degree to customer demand.

[0075] In actual application scenarios, a dynamic relationship graph is constructed through GraphX Java API, where nodes include customer nodes (attributes: rehabilitation stage, chronic disease type) and team nodes (attributes: professional qualifications, service capability rating), and edge weights are calculated based on historical service effect evaluation matrix. An improved PageRank algorithm (with a time decay factor) is used for iterative calculation of node importance, and after generating an initial recommendation list, a multi-dimensional secondary filtering is performed through the Drools rule engine: 1) Geofence matching (5 km radius priority); 2) Service schedule verification; 3) Price sensitivity adaptation, finally achieving high matching accuracy.

[0076] S33: Based on at least one target health service item and target health service team, a health service recommendation scheme for the target customer is generated.

[0077] In this step, based on the target health service project and health and wellness team matching results of intelligent screening, multi-dimensional evaluation indicators (including the current health status of the customer, risk level assessment and potential fraud risk coefficient) are fused to generate a dynamically optimized personalized health service recommendation scheme. This scheme quantitatively analyzes the adaptation degree of customer demand and service characteristics, constructs an optimal service combination strategy, and provides precise and differentiated health management solutions for target customers, achieving a full-process intelligent service closed loop from risk early warning to health intervention.

[0078] In an embodiment of the present application, a specific health service recommendation scheme updating scheme is provided, that is, based on the claim fraud score and health risk score, the health service recommendation scheme for the target customer is generated through the graph neural network, and the following steps are further included:

[0079] Obtaining real-time health monitoring data of the target customer;

[0080] Determining the health trend change of the target customer based on the real-time health monitoring data;

[0081] Judging whether the target customer has health index abnormalities based on the health trend change;

[0082] When it is detected that the target customer has health index abnormalities, updating the health risk score and the health service recommendation scheme based on the real-time health monitoring data.

[0083] In this embodiment, by continuously collecting real-time health monitoring data of the target customer, a time series analysis algorithm is used to dynamically monitor and trend analyze multi-dimensional physiological parameters, accurately identify the deviation state and evolution trend of health indicators, and realize real-time evaluation of the health status of the customer. When significant indicator fluctuations are detected, the system will automatically trigger the health risk score recalculation mechanism, update the risk assessment model parameters using incremental learning, and ensure that the score result dynamically reflects the latest health status of the customer. At the same time, based on the updated risk portrait, the health service recommendation strategy is optimized in real time through the reinforcement learning algorithm, keeping the service scheme and the customer's health demand in precise synchronization.

[0084] In practical application scenarios, an intelligent health monitoring rule base is constructed based on the Drools rule engine, and multi-dimensional trigger conditions are defined: physiological indicator threshold rules (such as a continuous 72-hour blood glucose value > 140 mg / dL triggering a diabetes management process); composite event rules (such as a heart rate variability decrease accompanied by sleep interruption triggering stress intervention). A real-time processing pipeline is constructed using Apache Flink, and event time semantics (EventTime) is used to process device data streams. Key implementations include: sliding window aggregation (1-hour window, 30-second sliding); KeyedState maintenance of customer health baseline; CEP complex event detection (such as SpO2 sudden drop pattern recognition). When an abnormal event is detected, the system automatically performs the following linkage operations: calls the GraphX API to update the edge weight matrix of the customer-service team graph in real time; triggers XGBoost model incremental training to update risk scores; generates dynamic adjustment suggestions (such as increasing the rehabilitation frequency from 2 times / week to 4 times / week), and finally generates a dynamically adjusted health care service recommendation plan (such as increasing rehabilitation frequency or changing the service team). The adjustment results are synchronized to the Elasticsearch and blockchain notarization modules to form a closed-loop feedback. Further, Flink stream processing and rule engine collaboration feed back health intervention data (such as customer rehabilitation progress) to the XGBoost risk model, and an Online Learning mechanism is implemented through Java to dynamically update premium discount policies. For example, if a target customer has met health indicators for 3 consecutive months, a premium discount rule is triggered, and a Java program automatically adjusts the policy terms by calling the insurance core system API, forming an intelligent closed loop of "data collection-intelligent analysis-risk control-service optimization" linkage.

[0085] As can be seen, in the above scheme, by integrating multi-dimensional data such as medical records, health monitoring data, and claim records of target customers, a complete customer health portrait data set is constructed. Based on this data set, accurate claim fraud detection and quantification of customer health risk levels are performed. Subsequently, the customer basic portrait data and double-score results (claim fraud score and health risk score) are jointly input into a graph neural network model, and a three-dimensional relationship network of "data collection-intelligent analysis-risk control-service optimization" is dynamically constructed to realize intelligent generation of personalized health service plans. Compared with the traditional mode of relying only on paper materials submitted by customers for manual assessment and recommendation, the present application realizes the integration of insurance claims and health services, providing customers with "risk protection + health management" dual value. Through cross-analysis and intelligent matching of multi-dimensional data, the fit degree of the final output health service plan and the actual needs of the customer is significantly improved, effectively improving the matching degree of the health service recommendation plan and the customer.

[0086] It should be understood that the size of the serial number of each step in the above embodiments does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0087] In an embodiment, a health service recommendation scheme generation apparatus is provided, which corresponds to the health service recommendation scheme generation method in the above embodiments. As shown in the figure, the health service recommendation scheme generation apparatus 100 includes an acquisition module 101, a first generation module 102, and a second generation module 103. The functions of each module are described in detail as follows: Figure 6

[0088] The acquisition module 101 is configured to acquire a claim health data set of a target customer, wherein the claim health data set includes basic portrait data, claim records, medical records, and health monitoring data.

[0089] The first generation module 102 is configured to perform claim fraud detection and health risk assessment on the target customer based on the claim health data set, to obtain a claim fraud score and a health risk score of the target customer.

[0090] The second generation module 103 is configured to generate a health service recommendation scheme for the target customer based on the claim fraud score, the health risk score, and the claim health data set through a graph neural network technology.

[0091] In an embodiment, the acquisition module 101 is specifically configured to:

[0092] Acquire multi-modal data of the target customer, wherein the multi-modal data includes structured data and unstructured data.

[0093] Integrate the multi-modal data based on a message queue to generate the claim health data set.

[0094] In an embodiment, the claim records include accident description text and accident images of a claim event, and the first generation module 102 is specifically configured to:

[0095] Obtain first accident features in the accident description text through natural language processing technology;

[0096] Obtain second accident features in the accident images through image processing method;

[0097] Generate a claim fraud score of the target customer based on the first accident features, the second accident features, and the claim records;

[0098] Determine a disease history of the target customer based on the medical records;

[0099] ​Based on the basic portrait data, disease history and health monitoring data of the target customer, the health risk score of the target customer is calculated through a gradient boosting tree model.

[0100] In an embodiment, the first generation module 102 is specifically further configured to:

[0101] Based on the first accident feature and the second accident feature, the intersection-over-union value between the accident description text and the accident image is calculated through an intersection-over-union algorithm;

[0102] Based on the claim record, the claim frequency and the claim amount of the target customer are determined;

[0103] Based on the intersection-over-union value, the claim frequency and the claim amount, the claim fraud score of the target customer is generated through a logistic regression model.

[0104] In an embodiment, the apparatus further comprises:

[0105] The third generation module is configured to generate a fake image sample through a generative adversarial network model;

[0106] The training module is configured to train the residual network by taking the fake image sample and the historical claim accident image sample as training data, to obtain a fake feature recognition model;

[0107] The fourth generation module is configured to extract the fake feature of the accident image through the fake feature recognition model, to obtain the authenticity probability of the accident image;

[0108] The comparison module is configured to mark the claim event corresponding to the accident image as an abnormal event if the authenticity probability is less than or equal to a preset authenticity threshold;

[0109] The fifth generation module is configured to generate an alarm information based on the abnormal event;

[0110] The sending module is configured to send the alarm information to a terminal of a target person.

[0111] In an embodiment, the second generation module 103 is specifically configured to:

[0112] Based on the graph neural network, a multi-dimensional relationship graph is constructed, wherein the multi-dimensional relationship graph includes the potential relationship among the customer, the health service item and the health care team;

[0113] The claim fraud score, the health risk score, the basic portrait data and the health monitoring data are input into the multi-dimensional relationship graph as node features, to obtain at least one target health service item and a target health service team that meet the target customer;

[0114] Based on the at least one target health service item and the target health service team, a health service recommendation scheme for the target customer is generated.

[0115] In an embodiment, the acquisition module 101 is further configured to acquire real-time health monitoring data of the target customer.

[0116] In an embodiment, the apparatus further comprises:

[0117] a determination module configured to determine a health trend change of the target customer based on the real-time health monitoring data;

[0118] a judgment module configured to judge whether the target customer has a health index abnormality based on the health trend change;

[0119] an updating module configured to update the health risk score and the health service recommendation scheme based on the real-time health monitoring data when it is detected that the target customer has the health index abnormality.

[0120] The present application provides a health service recommendation scheme generation apparatus 100, which integrates multi-dimensional data such as medical records, health monitoring data, and claim settlement records of a target customer to construct a complete customer health portrait data set. Based on the data set, accurate claim settlement fraud detection and quantification of customer health risk levels are performed. Subsequently, the customer basic portrait data and the double-score results (claim settlement fraud score and health risk score) are jointly input into a graph neural network model to dynamically construct a three-dimensional relationship network of “data collection-intelligent analysis-risk control-service optimization”, thereby realizing intelligent generation of personalized health service schemes. Compared with the traditional mode of relying only on paper materials submitted by customers for manual assessment, the present application realizes the integration of insurance claim settlement and health services, and provides customers with “risk protection + health management” dual value. Through cross-analysis and intelligent matching of multi-dimensional data, the fit degree of the final output health service scheme and the actual needs of the customer is significantly improved, effectively improving the matching degree of the health service recommendation scheme and the customer.

[0121] The specific limitations of the health service recommendation scheme generation apparatus can be referred to the limitations of the health service recommendation scheme generation method in the above, which will not be repeated here. Each module in the above health service recommendation scheme generation apparatus can be realized by software, hardware, and combinations thereof, in whole or in part. Each module described above can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to each module.

[0122] In one embodiment, a computer device is provided, which can be a server, and its internal structure diagram can be as shown in Figure 7As shown in the figure. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes non-volatile and / or volatile storage media, internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with the external client through the network connection. The computer program is executed by the processor to realize the functions or steps of the server side of the health service recommendation scheme generation method.

[0123] In one embodiment, a computer device is provided, which can be a client, and its internal structure diagram can be as shown in the figure. Figure 8 As shown in the figure. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes non-volatile storage media, internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with the external server through the network connection. The computer program is executed by the processor to realize the functions or steps of the client side of the health service recommendation scheme generation method.

[0124] In one embodiment, a computer device is provided, including a memory, a processor and a computer program stored on the memory and executable on the processor, and the processor executes the computer program to realize the following steps:

[0125] Obtain the target customer's claim health data set, wherein the claim health data set includes basic portrait data, claim records, medical records and health monitoring data;

[0126] Based on the claim health data set, the target customer is detected for claim fraud and health risk assessment to obtain the target customer's claim fraud score and health risk score;

[0127] Based on the claim fraud score, the health risk score and the claim health data set, a health service recommendation scheme for the target customer is generated through a graph neural network technology.

[0128] In one embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by the processor to realize the following steps:

[0129] obtaining a claim settlement health data set of the target customer, wherein the claim settlement health data set comprises basic portrait data, claim settlement records, medical records and health monitoring data;

[0130] performing claim fraud detection and health risk assessment on the target customer based on the claim settlement health data set to obtain a claim fraud score and a health risk score of the target customer;

[0131] generating a health service recommendation scheme for the target customer based on the claim fraud score, the health risk score and the claim settlement health data set through a graph neural network technology.

[0132] It should be noted that the functions or steps described above with respect to the computer-readable storage medium or the computer device can correspond to the related descriptions of the server side and the client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0133] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In the embodiments provided in the present application, any reference to memory, storage, database or other medium can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM) and memory bus dynamic RAM (RDRAM) and the like.

[0134] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is exemplified. In actual application, the above-mentioned functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above.

[0135] The above-described embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method of generating a health service recommendation plan, characterized by, The method comprises the following steps: obtaining a claim health data set of a target customer, wherein the claim health data set comprises basic portrait data, claim records, medical records and health monitoring data; based on the claim health data set, performing claim fraud detection and health risk assessment on the target customer to obtain a claim fraud score and a health risk score of the target customer; based on the claim fraud score, the health risk score and the claim health data set, generating a health service recommendation scheme for the target customer through a graph neural network technology.

2. The method of claim 1, wherein, The step of obtaining the claim health data set of the target customer specifically comprises: obtaining multi-modal data of the target customer, wherein the multi-modal data comprises structured data and unstructured data; integrating the multi-modal data based on a message queue to generate the claim health data set.

3. The method of claim 1, wherein, The claim records include accident description text and accident images of a claim event, and the step of performing claim fraud detection and health risk assessment on the target customer based on the claim health data set to obtain a claim fraud score and a health risk score of the target customer specifically comprises: obtaining first accident features in the accident description text through natural language processing technology; obtaining second accident features in the accident images through image processing method; generating a claim fraud score of the target customer based on the first accident features, the second accident features and the claim records; determining the disease history of the target customer based on the medical records; calculating the health risk score of the target customer through a gradient boosting tree model based on the basic portrait data, the disease history and the health monitoring data of the target customer.

4. The method of claim 3, wherein, The step of generating a claim fraud score of the target customer based on the first accident features, the second accident features and the claim records specifically comprises: calculating the intersection over union value between the accident description text and the accident images through an intersection over union algorithm based on the first accident features and the second accident features; determining the claim frequency and claim amount of the target customer based on the claim records; generating the claim fraud score of the target customer through a logistic regression model based on the intersection over union value, the claim frequency and the claim amount.

5. The method of claim 3, wherein, After obtaining the claim health data set of the target customer, the method further comprises the following steps: generating a fake image sample through a generative adversarial network model; training a residual network by taking the fake image sample and historical claim accident image sample as training data to obtain a fake feature recognition model; extracting fake features of the accident image through the fake feature recognition model to obtain a probability of authenticity of the accident image; if the probability of authenticity is less than or equal to a preset authenticity threshold, marking the claim event corresponding to the accident image as an abnormal event; generating an alarm information based on the abnormal event and sending the alarm information to a terminal of a target person.

6. The method of claim 1, wherein, The step of generating the health service recommendation scheme for the target customer based on the claim fraud score, the health risk score and the claim health data set through a graph neural network technology specifically includes: Constructing a multi-dimensional relationship graph based on a graph neural network, wherein the multi-dimensional relationship graph includes potential relationships among customers, health service items and health and wellness teams; Inputting the claim fraud score, the health risk score, the basic portrait data and the health monitoring data as node features into the multi-dimensional relationship graph to obtain at least one target health service item and a target health service team that meet the target customer; Generating the health service recommendation scheme for the target customer based on the at least one target health service item and the target health service team.

7. The method according to any one of claims 1 to 6, characterized in that, After the step of generating the health service recommendation scheme for the target customer based on the claim fraud score, the health risk score and the claim health data set through the graph neural network technology, the method further includes: Obtaining real-time health monitoring data of the target customer; Determining a health trend change of the target customer based on the real-time health monitoring data; Determining whether the target customer has a health index abnormality based on the health trend change; When it is detected that the target customer has a health index abnormality, updating the health risk score and the health service recommendation scheme based on the real-time health monitoring data.

8. A health service recommendation scheme generation apparatus characterized by comprising: The method includes: An acquisition module configured to obtain a claim health data set of a target customer, wherein the claim health data set includes basic portrait data, claim records, medical records and health monitoring data; A first generation module configured to perform claim fraud detection and health risk assessment on the target customer based on the claim health data set to obtain a claim fraud score and a health risk score of the target customer; A second generation module configured to generate a health service recommendation scheme for the target customer based on the claim fraud score, the health risk score and the claim health data set through a graph neural network technology.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the health service recommendation scheme generation method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the steps of the health service recommendation scheme generation method according to any one of claims 1 to 7.