Product recommendation method and device, computer equipment and storage medium

Abstract

The invention discloses a product recommendation method and device, computer equipment and a storage medium, belongs to the technical field of artificial intelligence, and is applied to health management and health insurance product recommendation scenes. The method comprises the following steps: firstly, carrying out precise modeling on dynamic changes of user physiological indexes by utilizing time sequence analysis, and obtaining an interpretable trend and an abnormal mode, so as to realize early recognition and quantitative evaluation of health risks; secondly, carrying out traceable and explainable multi-modal fusion on an abnormal index quantification result and individualized multi-dimensional information (family medical history, medical behaviors and the like), and outputting a comprehensive health risk score; thirdly, through efficient clustering analysis of historical risk data, a clear group portrait of similar risk groups is formed, and robustness and expandability of grouping intervention and personalized matching are improved; and finally, based on a matching algorithm of similar risk groups and a rich product library, personalized recommendation of health insurance products is realized, and the user satisfaction and the conversion rate are improved.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, specifically relating to a product recommendation method, apparatus, computer equipment, and storage medium. Background Technology

[0002] In the field of health management, regular physical examinations enable disease prevention and early intervention, playing a crucial role in improving individual health and reducing medical costs. With increasing public awareness of health, personalized physical examination package recommendations have become an important direction for industry development. Traditional physical examination package recommendations often rely on fixed templates or simple demographic information, making it difficult to accurately match users' health needs. This approach often fails to provide targeted solutions when facing complex and ever-changing health risks, resulting in low utilization rates of user examination results and ineffective health management. For example, when recommending health insurance products based on user examination results, the failure to fully explore the dynamic changes in users' physiological indicators and multi-dimensional health risk factors leads to a low match between the recommended health insurance products and the user's actual risk needs. This may result in insufficient protection for users or lead them to purchase unnecessary products, reducing user experience and conversion rates.

[0003] Existing recommendation methods have significant limitations when integrating user health information. Many solutions rely solely on health checkup data from a single point in time, ignoring the dynamic nature of health indicators over time. For example, if a user's indicator is abnormal in the short term but shows a stable long-term trend, traditional methods may incorrectly amplify the risk and recommend unnecessary tests. Furthermore, existing methods often lack the ability to comprehensively analyze multi-source data when processing personalized user information, such as effectively incorporating factors like family medical history, age changes, and medical history. This leads to a mismatch between recommended packages and the user's actual health risks, affecting the relevance and effectiveness of the health checkup.

[0004] At the technical level, the core challenge of dynamic recommendations lies in accurately quantifying health risks and achieving precise matching of service packages. Firstly, time-series analysis of abnormal indicators is a crucial technical attribute. Abnormalities in health indicators are not only related to current values ​​but also closely linked to trends over time. For example, if a user's blood sugar level gradually increases within a year but does not reach the abnormal threshold, traditional static analysis might overlook this potential risk, while dynamic analysis needs to quantify the importance of the trend through time decay factors. However, achieving this attribute introduces another technical challenge: how to generate a unified recommendation index by integrating multiple dimensions of factors (such as age, family history, and medical behavior) when quantifying risk. If the weighting of these factors is unreasonable, the recommendation index may deviate from actual needs. For example, a middle-aged user with a family history of diabetes but no recent medical intervention records might have their testing needs underestimated, missing early intervention opportunities, if the weighting is inappropriate.

[0005] Therefore, how to rationally integrate multi-dimensional risk factors based on dynamic analysis of health indicator trends, generate accurate recommendation indices, and match optimal physical examination packages has become a key issue in the field of health management. Summary of the Invention

[0006] The purpose of this application is to propose a product recommendation method, apparatus, computer equipment, and storage medium to address the key issue in the field of health management: how to rationally integrate multi-dimensional risk factors based on dynamic analysis of health indicator trends, generate an accurate recommendation index, and match the optimal health checkup package.

[0007] To address the aforementioned technical problems, this application provides a product recommendation method, employing the following technical solution: A product recommendation method, comprising: A time-series dataset is obtained by using users' historical physical examination records and current user health data. The time-series dataset includes a sequence of physiological indicator values ​​at consecutive time points. Time series analysis algorithms are applied to time series datasets to process changes in physiological index values, thereby obtaining dynamic trends in physiological index changes. Based on the dynamic trend of physiological indicator changes, the deviation range of physiological indicators is identified, and the abnormality pattern of indicators is determined according to the deviation range of physiological indicators, so as to obtain the quantitative results of abnormal indicators. The user's family medical history information and user medical behavior records are used as multi-dimensional factor inputs. The quantitative results of abnormal indicators are combined with multi-dimensional factors and processed by weighted fusion to obtain a comprehensive risk assessment score. Clustering algorithms were used to group and analyze comprehensive risk assessment scores and historical user risk data to identify similar health risk groups; Based on similar health risk groups, a pre-set matching algorithm is used to determine the most suitable personalized product combination from the product library for the user.

[0008] To address the aforementioned technical problems, this application also provides a product recommendation device, which employs the following technical solution: A product recommendation device, comprising: The time series module is used to obtain a time series dataset through the user's historical physical examination records and current user health data. The time series dataset includes a sequence of physiological indicator values ​​at consecutive time points. The dynamic indicators module is used to apply time series analysis algorithms to time series datasets to process changes in physiological indicator values ​​and obtain dynamic trends in physiological indicator changes. The abnormal pattern module is used to identify the deviation range of physiological indicators based on the dynamic trend of physiological indicator changes, and to determine the abnormal pattern of the indicators based on the deviation range of physiological indicators, so as to obtain the quantitative results of abnormal indicators. The risk assessment module is used to obtain users’ family medical history information and users’ medical behavior records as multi-dimensional factor inputs. The quantitative results of abnormal indicators are combined with multi-dimensional factors and weighted fusion processing is used to obtain a comprehensive risk assessment score. The group clustering module is used to perform group analysis on comprehensive risk assessment scores and historical user risk data using clustering algorithms to identify similar health risk groups; The product recommendation module is used to determine the most suitable personalized product combination for the user from the product library based on similar health risk groups using a preset matching algorithm.

[0009] To address the aforementioned technical problems, this application also provides a computer device that employs the following technical solution: A computer device includes a memory and a processor, the memory storing computer-readable instructions, the processor executing the computer-readable instructions to implement the steps of the product recommendation method as described in any of the preceding claims.

[0010] To address the aforementioned technical problems, this application also provides a computer-readable storage medium, employing the technical solution described below: A computer-readable storage medium storing computer-readable instructions that, when executed by a processor, implement the steps of the product recommendation method as described in any one of the preceding descriptions.

[0011] Compared with the prior art, the embodiments of this application have the following main advantages: This application discloses a product recommendation method, apparatus, computer equipment, and storage medium, belonging to the field of artificial intelligence technology, and applied to the recommendation scenario of health insurance products. First, it utilizes time-series analysis to accurately model the dynamic changes of users' physiological indicators, obtaining interpretable trends and abnormal patterns, thereby achieving early identification and quantitative assessment of health risks. Second, it performs traceable and interpretable multimodal fusion of the quantitative results of abnormal indicators with individualized multi-dimensional information (family medical history, medical behavior, etc.) to output a comprehensive health risk score. Third, through efficient cluster analysis of historical risk data, it forms a clear profile of similar risk groups, improving the robustness and scalability of group-based intervention and personalized matching. Finally, based on the matching algorithm for similar risk groups and a rich product library, it can output personalized product combinations highly consistent with individual health status, preferences, budgets, and intervention needs, possessing interpretability, traceability, and scalability, achieving personalized recommendations of health insurance products, and improving user satisfaction and conversion rates. Attached Figure Description

[0012] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 An exemplary system architecture diagram is shown, in which this application can be applied; Figure 2 A flowchart of one embodiment of the product recommendation method according to this application is shown; Figure 3 It shows Figure 2 A flowchart of an embodiment of step S303; Figure 4 A schematic diagram of one embodiment of the product recommendation device according to this application is shown; Figure 5 It shows Figure 4 A schematic diagram of a structure of an embodiment of the abnormal mode module 403; Figure 6 A schematic diagram of the structure of one embodiment of a computer device according to this application is shown. Detailed Implementation

[0014] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0015] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0016] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0017] like Figure 1As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables.

[0018] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0019] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer, and a desktop computer, etc.

[0020] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0021] It should be noted that the product recommendation method provided in this application embodiment is generally executed by a server / terminal device, and correspondingly, the product recommendation device is generally set in the server / terminal device.

[0022] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative; the system can have any number of terminal devices, networks, and servers depending on implementation needs.

[0023] Continue to refer to Figure 2 A flowchart illustrating an embodiment of a product recommendation method according to this application is shown. The product recommendation method includes the following steps: S201, obtain a time-series dataset by using the user's historical physical examination records and current user health data, wherein the time-series dataset includes a sequence of physiological indicator values ​​at consecutive time points; Specifically, in the stage of acquiring time-series datasets, it is necessary to clarify the data source, data type, and collection frequency, and design a unified data access and preprocessing pipeline. First, establish user profile identification and data segmentation logic to ensure that records for the same user across different data sources are aligned. Then, perform data fusion between historical physical examination records and current health data to resolve issues such as misaligned timestamps, inconsistent units, and differences in measurement units.

[0024] At the implementation level, time window join or timestamp-based interpolation alignment strategies can be adopted, with unit standardization and unification of dimensions (e.g., unifying units for indicators such as blood pressure, heart rate, and blood glucose) as necessary. Multiple imputation strategies are employed for missing data: simple imputation (e.g., forward imputation, backward imputation), statistical imputation (e.g., mean, median), and model imputation based on temporal correlation (e.g., Kalman filtering, temporal GAN). Robust handling of outliers is implemented, using methods such as box plots, Z-scores, and moving medians for initial detection, followed by setting thresholds and upper / lower limits based on domain knowledge. If necessary, a Bayesian anomaly detection model is introduced to distinguish between single-point anomalies and systemic anomalies. Regarding data quality control, monitoring of indicators such as data integrity, uniqueness, consistency, and real-time performance is established, triggering automated alarms and manual review processes.

[0025] Further, time-series data structures are constructed for physiological indicators, such as multivariate time-series data structures. Each time point includes metadata such as multidimensional indicator vectors, timestamps, data source identifiers, and acquisition device IDs to support feature extraction and model training. Regarding privacy compliance, localized data processing, anonymization, and minimal data collection principles are implemented, along with encrypted transmission and access control to ensure compliance with relevant regulations. Finally, a repeatable experimental grouping and evaluation framework is designed, splitting the old and new datasets into training, validation, and test sets to ensure that the temporal dependencies of the time-series data are appropriately handled during evaluation.

[0026] S202, apply time series analysis algorithms to the time series dataset to process changes in physiological index values ​​and obtain dynamic trends in physiological index changes; Specifically, when applying time-series analysis algorithms to extract dynamic physiological indicator trends from time-series datasets, a scalable analysis architecture needs to be constructed, covering feature engineering, model training, trend output, and uncertainty quantification. Key technical points include: first, establishing an independent time-series analysis pipeline for each physiological indicator, allowing parallel processing to improve throughput; second, developing indicator-level feature engineering schemes to extract features such as trend components, seasonality, periodicity, periodicity intensity, volatility, variation rate, and event-driven abrupt change points; and, when necessary, introducing methods such as wavelet transform, Fourier analysis, or empirical mode decomposition (EMD) to separate different frequency domain components and improve the robustness of trend modeling.

[0027] In terms of trend modeling, it supports multi-model fusion: for short-term trends, it uses classic methods such as autoregressive / moving average (AR / MAR / ARMA / ARIMA), and for long-term trends and nonlinear relationships, it uses deep time series networks such as LSTM / GRU, combined with autoregressive ensemble methods (such as ARIMA-GARCH combination) to capture conditional heteroscedasticity.

[0028] For model selection, a cross-validation-based model comparison framework is designed, using time series cross-validation (such as rolling window CV) to evaluate stability and generalization ability. Information criteria (AIC / BIC) can be used to assist model selection. The output needs to transform trend results into consumable dynamic indicators: trend direction, trend rate, prediction interval, confidence level, and predicted values ​​for future time points. Uncertainty is quantified, such as by providing prediction intervals, probability distributions, or confidence levels, and a visual representation of model uncertainty is retained. For outlier identification and labeling, residual analysis of the trend model and control chart methods such as CUSUM / SPC are combined to output the deviation strength, occurrence time period, and duration. Finally, a traceable computation graph and log system is designed to record the input data, model version, parameter settings, random seed, and results for each step, ensuring repeatability and auditability.

[0029] S203, Identify the deviation range of physiological indicators based on the dynamic trend of physiological indicator changes, determine the abnormal indicator pattern based on the deviation range of physiological indicators, and obtain the quantitative results of abnormal indicators. Specifically, in implementing the identification of deviations from physiological indicators and determining abnormal patterns based on dynamic physiological indicator trends, a multi-layered judgment framework needs to be established to ensure that deviation judgments are both sensitive and interpretable. First, a baseline for the deviation range is defined: individual historical data is used as the center, combined with population statistics (such as mean ± n standard deviations, tolerance intervals based on quantiles, etc.) to establish individualized deviation thresholds for each indicator. Second, trend results and thresholds are combined to design a dynamic threshold update mechanism, allowing the thresholds to adapt to changes in time and individual status. Deviation patterns are decomposed into several typical categories, such as the exceeding-threshold duration in a continuous upward trend, periodic dramatic events, and abnormal amplitudes in periodic fluctuations, and each pattern is assigned interpretable labels and conditions to facilitate risk communication and decision support.

[0030] The implementation can employ a combination of rule-based learning and statistical testing: statistical tests (such as slope significance test, CUSUM, SPC, and multivariate control charts) are performed on trend slope, residual distribution, and deviation duration, and the significance level, threshold, and event time window are recorded in a traceable decision log. During the identification process, multivariate coupling effects are considered, and linkage analysis (such as Granger causality, vector autoregression (VAR), and dynamic graphical models) is used to identify the mutual influence between indicators, thereby avoiding misjudgments caused by univariate thresholds. The output anomaly patterns are encoded to form a pattern library for pattern matching, risk stratification, and personalized intervention strategy recommendations. Finally, the output provides an interpretable report on the deviation patterns, including inferences about the causes of deviation, time-period clustering, and preliminary guidance on user health management strategies, ensuring that comprehensive risk assessment and product matching fully utilize this information.

[0031] S204: Obtain user's family medical history information and user medical behavior records as multi-dimensional factor inputs, combine the quantitative results of abnormal indicators with multi-dimensional factors and perform weighted fusion processing to obtain a comprehensive risk assessment score; Specifically, in achieving a comprehensive risk assessment score by combining the quantitative results of abnormal indicators with multi-dimensional factors and then weighted fusion, a flexible and interpretable weighted fusion framework needs to be designed to support dynamic adjustment and individualized configuration of weights. Key technical points include: First, defining independent structures for the indicator layer, factor layer, and interaction item layer, uniformly encoding the quantitative results of abnormal indicators, family medical history information, and medical behavior records (e.g., scores, probabilities, category labels) to form a unified multi-dimensional input vector. Second, designing a weight allocation mechanism, employing evidence-based adaptive weight update strategies, such as Bayesian weight updates, adaptive weight adjustment based on historical prediction errors, or learning optimal weights through a meta-learning framework.

[0032] For multi-source data fusion, safe and efficient fusion methods are employed, such as weighted linear combination, hierarchical fusion, or attention-based nonlinear fusion, to capture the relative contribution of different dimensions to risk and ensure the model possesses a certain degree of interpretability (through sensitivity analysis and local interpretability methods such as LIME / SHAP). The quantification results of outlier indicators are normalized to ensure that scores of different dimensions can be directly added or synthesized in probabilistic form; confidence levels or confidence intervals are introduced when necessary to express the uncertainty of the overall risk. A hierarchical risk scoring framework is established, mapping the overall risk score to risk levels (e.g., low, medium, high) and providing corresponding intervention recommendation strengths. Furthermore, a data-traceable assessment and calibration mechanism is designed, using historical real events to backtest the model output and label errors, ensuring the repeatability of weight adjustments. Finally, to meet privacy and regulatory requirements, input data is subject to minimum necessity, differential privacy injection, or peer-reviewed model transparency assessment to ensure the system achieves dynamic risk assessment and interpretation while remaining compliant.

[0033] S205 uses a clustering algorithm to group and analyze the comprehensive risk assessment scores and historical user risk data to identify similar health risk groups; Specifically, in identifying similar health risk groups through cluster analysis of comprehensive risk assessment scores and historical user risk data, a robust and scalable clustering framework needs to be established to handle high-dimensional, mixed-type data and time-related data. Key technical points include: First, data standardization and encoding are performed, transforming scores, risk labels, historical behaviors, etc., into consistent feature vectors. Dimensionality reduction techniques (such as PCA, t-SNE, UMAP, etc.) are used when necessary to reduce dimensionality, noise, and preserve important structure. Second, appropriate clustering algorithms are selected, and combinations can be used for different data characteristics: density-based clustering (such as DBSCAN, HDBSCAN) is advantageous for discovering clusters of arbitrary shapes, hierarchical clustering is beneficial for multi-scale group analysis, and model-based clustering (such as Gaussian Mixture Model, GMM) can provide the probability distribution and uncertainty assessment of clusters.

[0034] For time-related correlations, time series clustering or sliding window representation of historical risk data can be used to capture dynamic patterns and improve the stability of identifying similar health risk groups. To avoid bias caused by imbalanced samples, resampling techniques or weighted distance metrics are applied, and robust cluster structures are selected through multiple random initializations and stability assessments (such as silhouette coefficient, Davies-Bouldin index, and Gap statistic). Clustering results should include interpretable outputs: feature descriptions of each cluster, statistical distributions of members within the cluster, feature vectors of cluster centers, and explanations of the potential health risk similarity among members of the same cluster. The effectiveness of the clusters is evaluated using external validation methods, such as whether they improve the accuracy of the matching stage and whether they provide consistent recommendation patterns for similar populations. Finally, the clustering results are connected to the product matching algorithm, and the preference features and risk level distribution corresponding to each cluster are labeled to enable rapid matching of optimal combinations based on the user's cluster in the next step and support for personalized push notifications.

[0035] S206, based on similar health risk groups, uses a preset matching algorithm to determine the most suitable personalized product combination from the product library for the user.

[0036] Specifically, in implementing the process of determining personalized product combinations from a product library based on similar health risk groups using a pre-defined matching algorithm, an end-to-end intelligent matching and customized combination generation process needs to be established. Key technical points include: First, uniformly encoding the previously obtained information on similar risk groups, personal preferences, budget constraints, time factors (such as seasonal health check-up plans), and package attributes from the product library to form a request input vector. Second, designing a matching algorithm framework that combines rule-based matching and optimization solutions (such as combinatorial optimization / integer linear programming, constraint satisfaction problems CSP) with machine learning-assisted recommendation strategies (such as reinforcement learning-based strategies or hybrid recommendation models based on ranking learning) to select the most suitable personalized combination from the candidate package set that best matches the user's current health status and needs. To ensure interpretability, the output should include the matching reason, key matching indicators, and an explanation of the coverage of deviations from abnormal indicators. Furthermore, the system supports the automated generation of customized add-on items: based on user preferences and health risks, it dynamically generates optional supplementary medical examination items or services, such as additional laboratory tests, imaging examinations, and genetic counseling based on family history, and evaluates the necessity, expected benefits, and costs of each add-on. Implementation must consider scalable product library management to ensure that new packages, attribute updates, and price changes are reflected in the system in real time and remain consistent with the matching algorithm. Finally, an evaluation and feedback mechanism is designed to continuously optimize the matching strategy through A / B testing and offline + online evaluation methods, recording indicators such as user satisfaction, actual purchases, and health outcomes to periodically update the model and enhance its robustness.

[0037] Taking health insurance product recommendations as an example, simply put, the above-mentioned approach first calculates a comprehensive risk assessment score based on the user's temporal physiological indicator trends (such as persistently elevated blood sugar and abnormal blood pressure fluctuations) and the quantitative results of abnormal indicators (such as the duration of blood sugar deviation from the threshold and the risk weight of blood pressure fluctuations). This is combined with family medical history of diabetes and hypertension, as well as the user's medical history of regularly purchasing antihypertensive medications. This score is then clustered with the risk data of historical health insurance users to identify user groups with similar metabolic syndrome risk characteristics (e.g., the "middle-aged male high blood sugar and high blood pressure risk group"). Finally, a pre-set matching algorithm is used to select coverage packages from the health insurance product database that cover diabetes complication screening and hypertension-specific physical examinations. The deductible and sum insured are dynamically adjusted based on the average claims preferences of this group to generate a "personalized metabolic syndrome protection plan," along with targeted health management service recommendations (such as remote blood pressure monitoring services and diabetes diet guidance courses), achieving a precise match between health insurance products and the user's health risk needs.

[0038] Furthermore, the steps of applying time-series analysis algorithms to the time-series dataset to process changes in physiological indicator values ​​and obtain dynamic trends in physiological indicator changes specifically include: Preprocessing is performed on the time series dataset, including outlier detection, missing value imputation, and data smoothing. Feature extraction is performed on the preprocessed time series dataset to obtain multi-dimensional time series feature parameters; Based on multi-dimensional temporal feature parameters, a temporal modeling algorithm is used to dynamically model the user's physiological index sequence to obtain the prediction results of physiological index value changes. Based on the predicted results of changes in physiological indicators, the rate and direction of change of physiological indicators are calculated, and a trend curve is generated to describe the dynamic changes in individual health, thus obtaining the dynamic trend of physiological indicator changes.

[0039] In this embodiment, preprocessing the time-series dataset significantly improves the robustness and reproducibility of the analysis. Outlier detection requires combining statistical thresholds with domain knowledge, distinguishing between local and global outliers, and employing robust statistical methods (such as MAD and quantile methods) and weight adjustments based on device characteristics to reduce the impact of sensor drift or short-term noise. Missing value imputation considers not only temporal continuity but also physiological significance, prioritizing time-series model-driven imputation (such as Kalman filtering, higher-order Kalman filters, and temporal GANs), and incorporating constraints based on individual historical patterns and the distribution of homologous populations. Data smoothing should preserve trends and points of change, employing multi-scale smoothing strategies (such as Loess, Savitzky-Golay, and wavelet denoising) to distinguish between short-term noise and genuine physiological fluctuations.

[0040] In the feature extraction stage, time-series feature vectors need to be constructed at multiple dimensions, covering trends, seasonality, periodicity, volatility, peak and trough characteristics, variation rate, and event-driven mutation points. Feature dimensionality reduction and feature selection mechanisms are used to retain the components most explanatory to health dynamics. In the dynamic modeling stage, a parallel modeling pipeline is established, allowing the application of model combinations such as ARIMA, Prophet, and LSTM / GRU to each physiological indicator. Model ensemble strategies (such as weighted fusion, stacking, and Bayesian model averaging) are introduced to improve prediction robustness. The output trend curve provides not only point predictions but also interval predictions and uncertainty measures, ensuring that decision-making risks are controllable. Finally, the prediction results are linked with deviation detection, individualized thresholds, and historical health trajectories to form a traceable, interpretable, and scalable dynamic health monitoring and intervention plan generation system.

[0041] Through the above steps, high-precision modeling and interpretable trend output of dynamic changes in physiological indicators can be achieved.

[0042] Further, please refer to Figure 3 The steps involved in identifying deviations from physiological indicators based on dynamic trends, determining abnormal patterns based on these deviations, and obtaining quantitative results for the abnormal indicators include: S301, Obtain the standard reference range and individual baseline value of each physiological indicator, and establish a threshold model for the normal range of the indicators. The threshold model is used to determine the degree of deviation of each physiological indicator. S302 compares and analyzes the dynamic physiological index change trend with the threshold model, and calculates the deviation of the physiological index from the reference interval and the duration at each time point. S303, Identify the abnormal pattern type based on the deviation magnitude and duration, where the abnormal pattern types include short-term fluctuation type, continuous rise type, continuous decline type and periodic abnormal type; S304, perform quantitative scoring on each abnormal pattern, and generate an abnormal quantitative score for the indicator based on the degree of deviation, frequency of occurrence and duration. S305 normalizes and weights the abnormal quantitative scores of multiple physiological indicators to obtain the quantitative results of abnormal indicators.

[0043] In this embodiment, constructing an individualized threshold model by obtaining the standard reference interval and individual baseline value for each physiological indicator is a key prerequisite for achieving accurate deviation detection. The reference interval can be derived from a multimodal baseline combining population statistical distribution (such as the mean and standard deviation, quantile intervals, etc. of a normal distribution) and expert knowledge. To adapt to individual differences, the threshold model must not only cover static intervals but also incorporate temporal dynamics: the threshold can be adaptively adjusted as factors such as age, gender, past medical history, and lifestyle change. The establishment of baseline values ​​requires robust statistics based on the most recent N historical observations, such as using the median and MAD (median absolute deviation) to mitigate the influence of outliers, and performing sensitivity analysis on observations before and after the health event window to avoid threshold drift due to temporary anomalies.

[0044] The system can also assign weights or confidence levels to the threshold model by clearly defining a signal quality assessment mechanism at the device level, combining factors such as sensor drift, data packet loss rate, and sampling frequency changes, to ensure that deviation judgments are based on high-quality data. Subsequently, the threshold model is combined with dynamic physiological indicator trends to form a time-series-aware deviation assessment framework, enabling comparisons of the degree of deviation of the same indicator at different time points on the same scale. To achieve traceability, a versioned threshold library is designed, recording the reason for each threshold update, data source, statistical hypothesis, and validation results. Finally, the system can also establish a scalable architecture to support threshold learning and rapid integration of new physiological indicators, adapting to the expanded needs of future health monitoring scenarios.

[0045] Through the above steps, dynamic identification and quantitative scoring of deviations in multiple indicators can be achieved, providing unified and interpretable quantitative results of abnormal indicators to support personalized health intervention and precise matching.

[0046] Furthermore, the process of quantifying and scoring each abnormal pattern, generating quantitative scores for the indicators based on the degree of deviation, frequency of occurrence, and duration, specifically includes: The deviation of each physiological indicator is numerically quantified and calculated. The deviation score is determined based on the difference between the actual measured value and the central value of the reference interval. Statistically analyze the frequency of abnormal events within a specified time window, and calculate the frequency score based on the number of occurrences and the time distribution pattern of abnormal events; The duration of abnormal states is detected, and the length of time that continuously exceeds the normal threshold is converted into a duration score. The abnormality score of the indicator is obtained by weighting and fusing the deviation score, frequency score and duration score.

[0047] In this embodiment, when quantifying and scoring each abnormal pattern, a unified and traceable score calculation framework needs to be established to ensure the comparability and interpretability of different physiological indicators under the same benchmark. The numericalization of deviation magnitude first requires clarifying the scaling relationship between the "reference interval center value" and the "measured value." It is recommended to use a standardized deviation metric, such as dividing the difference by the adaptive standard deviation of the individual baseline or the width of the quantile interval, to obtain a scaled deviation score (e.g., a score range of 0 to 100). The same score mapping rules should be set for different indicators to avoid weight bias caused by scale inconsistencies.

[0048] To improve robustness, the impact of outliers should be separated. For example, robust correction can be applied to extreme values ​​when calculating deviation magnitude, or outliers can be clustered and labeled within a sliding time window to prevent score calculation from being distorted by isolated events. For the calculation of the "frequency score," a time window (e.g., the past 14 or 28 days) needs to be defined, and the number of outliers within that window recorded. The weights should be adjusted based on the temporal distribution characteristics of the events (e.g., whether they are concentrated in a specific time period or exhibit a periodic pattern). Statistical models such as Poisson or negative binomial distributions can be introduced to estimate the expected frequency of occurrence, and a score consistent with historical observations can be given accordingly. The duration score requires mapping the duration of continuous exceedances to a score. During the mapping process, the continuity and distribution stability of the events should be considered. For example, the length of continuous segments can be segmented and weighted, with lower weights given to short-term, brief exceedances, while significant gains are given to long-term, sustained exceedances.

[0049] When weighting and fusing the three types of scores, an interpretable weighting mechanism should be adopted, such as based on feature importance analysis (SHAP / LIME, etc.), the contribution of historical events to prediction errors, and the empirical weight of physician domain knowledge. The weights should be dynamically adjusted to adapt to different populations and monitoring scenarios. To ensure system traceability, a versioned scoring rule library should be established, recording the reasons for each rule update, data sources, statistical hypotheses, and validation results, and providing a replay function to reproduce the historical scoring process. Finally, the output should not only provide a single score but also a detailed table of score decomposition, listing the deviation magnitude, frequency, and duration scores, along with their respective weights and interval mapping relationships, facilitating interpretive communication and intervention decisions by health management personnel. If multi-source data is involved, the weights of different sources should be adjusted based on temporal consistency, instrument calibration status, and data quality to ensure that the final score comprehensively considers data credibility. The entire process should be scalable, facilitating the inclusion of new abnormal patterns and novel detection indicators, and seamlessly integrated with comprehensive risk assessment and personalized product matching modules.

[0050] Through the above steps, a multi-dimensional and interpretable quantitative score for abnormal patterns can be achieved.

[0051] Furthermore, the steps of obtaining user family medical history information and user medical behavior records as multi-dimensional factor inputs, and combining the quantitative results of abnormal indicators with multi-dimensional factors through weighted fusion processing to obtain a comprehensive risk assessment score, specifically include: Obtain user's family medical history information and perform feature encoding to obtain a family medical history risk vector; Collect user medical behavior records and establish a medical behavior risk vector based on time dimension and behavioral characteristics; The abnormal indicator quantification results, family medical history risk vector, and medical behavior risk vector are subjected to feature alignment and normalization. A weighted fusion model is used to fuse the normalized abnormal indicator quantification results, family medical history risk vector, and medical behavior risk vector to output a preliminary health risk score. The weighted fusion model includes linear weighting, hierarchical weighting, or machine learning-based adaptive weighting algorithms. The model is calibrated and confidence is assessed on the preliminary health risk score. Bias is corrected based on historical population sample data, and the final comprehensive risk assessment score is output.

[0052] In this embodiment, the core of fusing the quantitative results of abnormal indicators with multi-dimensional factors to output a comprehensive health risk score lies in establishing a flexible and interpretable multi-source data fusion framework. For encoding family medical history information, qualitative and quantitative information needs to be unified into a vectorized representation. Hierarchical feature engineering is employed to map specific disease categories, genetic predispositions, and family age distribution of onset to numerical features. Incomplete information processing strategies are introduced, such as proxy variables for missing values ​​or uncertainty weights, to avoid the significant impact of incomplete data on risk scoring. The time dimension processing of medical behavior records focuses on extracting behavioral patterns, such as regular physical examinations, medication adherence, and changes in treatment pathways. These are transformed into risk vectors using methods such as time series clustering and behavioral sequence embedding. Different weights are assigned to behavioral differences at different time periods to reflect the intensity of their impact on current health risk.

[0053] Aligning the quantified results of abnormal indicators, family medical history vectors, and medical behavior vectors typically requires standardizing feature scales (e.g., Z-score normalization, interval scaling) and addressing the coupling relationships between features. Interaction features should be introduced when necessary to capture potential synergistic effects. Weighted fusion models can employ linear weighting for transparency or introduce non-linear fusion with attention mechanisms to enhance adaptability, and should provide a weight update mechanism to adjust according to historical data and population evolution. The initial risk score from the fusion output needs statistical calibration, such as using calibration curves or Bayesian calibration, to correct systematic biases using historical population sample data, and to provide confidence intervals and uncertainty measures. The entire process requires a versioned feature dictionary and fusion rule base to ensure traceability, reproducibility, and rapid expansion to new features and populations. Finally, an interpretability report should be established, clearly defining the contribution of each input factor to the final score and potential intervention suggestions for use in the personalized product matching stage.

[0054] Through the above steps, interpretability risk scoring and robust integration of multidimensional inputs are achieved.

[0055] Furthermore, the steps of using a weighted fusion model to fuse the normalized abnormal indicator quantification results, family medical history risk vector, and medical behavior risk vector to output a preliminary health risk score include: Initial weight parameters are set based on the medical relevance and importance of different risk factors. These initial weight parameters include physiological indicator weights, family genetic risk weights, and behavioral risk weights. These initial weight parameters are used to guide the weight allocation of the fusion model. The normalized abnormal indicator quantification results, family medical history risk vector, and medical behavior risk vector are input into the weighted fusion model. The weighted fusion model is used to perform weighted product and feature interaction calculation on each input factor to obtain the risk contribution value of each dimension and generate a weighted intermediate result matrix. The weighted intermediate results are subjected to nonlinear transformation and normalization, and abnormal weight bias is suppressed by activation function or regularization mechanism. A preliminary health risk score is output based on the weighted and fused multidimensional risk characteristics.

[0056] In this embodiment, the initial health risk score is output by fusing the normalized abnormal indicator quantification results, family medical history risk vector, and medical behavior risk vector. The core of this approach lies in constructing a flexible and interpretable multimodal fusion framework. The initial weight parameters should be set not only based on medical knowledge and domain expert experience but also combined with data-driven statistical evidence, such as prediction errors of historical events, correlation analysis with the current population, and feature importance ranking, to dynamically adjust weights for different populations and scenarios. The implementation can employ a two-stage fusion approach: the first stage performs feature alignment, scale unification, and interactive feature construction to ensure comparability between different modalities within the same vector space; the second stage introduces verifiable weighted aggregation, such as linear weighting, attention mechanisms, or tree-based nonlinear fusion, to capture the complex coupling relationships between input factors.

[0057] To ensure the stability and robustness of the weights, regularization terms, interval constraints, and strategies to suppress extreme weights should be introduced to prevent bias caused by some inputs dominating the scoring. The generation of the intermediate result matrix should be traceable, recording weight configurations, input vectors, interaction terms, and intermediate layer outputs to support interpretable analysis. During the nonlinear transformation and normalization stages, appropriate activation functions (such as ReLU and GELU) should be selected, combined with normalization / noise control strategies, to ensure that the output scores are within an acceptable range and are comparable. Finally, an adaptive threshold calibration mechanism should be designed to maintain the stability of the initial risk score when new data is added and to provide confidence interval outputs to express uncertainty.

[0058] Through the above steps, the entire process should support versioned weights and integrated rule bases to facilitate rapid deployment across different population groups and the auditability of the model.

[0059] Furthermore, the steps of using clustering algorithms to group and analyze comprehensive risk assessment scores and historical user risk data to identify similar health risk groups specifically include: Construct a historical user risk dataset, which includes comprehensive risk assessment scores, scores of major abnormal indicators, family medical history factors, and medical behavior characteristics, to characterize individual health risk characteristics; Feature extraction is performed on the historical user risk dataset to obtain a set of user risk feature vectors; The user risk feature vector set is standardized and dimensionality reduced to obtain a dimensionality-reduced risk feature vector set. Select and train a clustering model, input the comprehensive risk assessment score and the dimensionality-reduced risk feature vector set into the clustering model, and obtain the clustering results; Based on the clustering results, the health risk center value and characteristic distribution of each group are calculated, and the Euclidean distance or cosine similarity between clustered samples are evaluated to identify similar health risk groups.

[0060] In this embodiment, when performing cluster analysis to identify similar health risk groups, a comprehensive design is needed, encompassing data preparation, feature engineering, dimensionality reduction, and clustering algorithm selection. The historical user risk dataset should not only include comprehensive risk assessment scores but also incorporate time-series statistics of key anomaly indicators, family medical history factors, and medical behavior characteristics (such as the mean, fluctuation range, and trend slope of the most recent 12 weeks) to form a rich vector representation. The feature extraction stage can employ self-supervised or supervised feature learning, combining domain knowledge to design interaction terms to capture the coupling effects between indicators.

[0061] In terms of dimensionality reduction, priority should be given to preserving cluster structure information, such as using t-SNE / UMAP for visualization preprocessing. However, in actual cluster training, PCA, ICA, or sparse autoencoders are preferred to maintain linear interpretability and computational stability. The choice of clustering algorithm should be based on prior assumptions about data distribution characteristics and cluster shape: for cases with irregular cluster shapes and large density differences, density clustering such as DBSCAN / HDBSCAN has advantages; if interpretability and probabilistic properties of inter-cluster distance are required, Gaussian Mixture Models (GMM) can be tried. To evaluate cluster quality, in addition to silhouette coefficients and Davies-Bouldin metrics, external validation should be combined, such as the consistency of health intervention effects within clusters, alignment with existing risk stratification, and the improvement effect on product recommendations.

[0062] To improve scalability, a modular framework is designed: the input layer supports batch updates of incremental data, versioned management of feature dictionaries, and the clustering model can be rolled back and retrained; the output provides cluster descriptions, center vectors, intra-cluster statistics, and explanations of cluster assignments for individual users, facilitating the rapid implementation of personalized push and intervention strategies.

[0063] Through the above steps, clustering and interpretation of cluster features of high-dimensional risk data can be achieved.

[0064] In the above embodiments, this application discloses a product recommendation method, belonging to the field of artificial intelligence technology, applied to the recommendation scenario of health insurance products. First, time-series analysis is used to accurately model the dynamic changes of users' physiological indicators, obtaining interpretable trends and abnormal patterns, thereby achieving early identification and quantitative assessment of health risks. Second, the quantitative results of abnormal indicators are fused with individualized multi-dimensional information (family medical history, medical behavior, etc.) in a traceable and interpretable multimodal manner to output a comprehensive health risk score. Third, through efficient clustering analysis of historical risk data, a clear profile of similar risk groups is formed, improving the robustness and scalability of group intervention and personalized matching. Finally, based on the matching algorithm for similar risk groups and a rich product library, a personalized product combination highly consistent with an individual's health status, preferences, budget, and intervention needs can be output, possessing interpretability, traceability, and scalability, realizing personalized recommendations of health insurance products, and improving user satisfaction and conversion rates.

[0065] In this embodiment, the product recommendation method operates on an electronic device (e.g., Figure 1 The server shown can receive instructions or acquire data via wired or wireless connection. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future wireless connection methods.

[0066] It should be emphasized that, to further ensure the privacy and security of the aforementioned product recommendation information, this information can also be stored in a blockchain node.

[0067] The blockchain referred to in this application is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0068] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0069] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0070] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0071] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by 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 accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0072] Further reference Figure 4 As a response to the above Figure 2 To implement the method shown, this application provides an embodiment of a product recommendation device, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0073] like Figure 4 As shown, the product recommendation device 400 described in this embodiment includes: The time series module 401 is used to obtain a time series dataset through the user's historical physical examination records and current user health data, wherein the time series dataset includes a sequence of physiological indicator values ​​at consecutive time points; The dynamic indicator module 402 is used to apply time series analysis algorithms to the time series dataset to process changes in physiological indicator values ​​and obtain dynamic trends in physiological indicator changes. The abnormal mode module 403 is used to identify the deviation range of physiological indicators based on the dynamic physiological indicator change trend, and to determine the abnormal mode of the indicators based on the deviation range of physiological indicators, so as to obtain the quantitative result of the abnormal indicators. Risk assessment module 404 is used to obtain user family medical history information and user medical behavior records as multi-dimensional factor inputs. It combines the quantitative results of abnormal indicators with multi-dimensional factors and performs weighted fusion processing to obtain a comprehensive risk assessment score. The group clustering module 405 is used to perform group analysis on the comprehensive risk assessment score and historical user risk data using a clustering algorithm to identify similar health risk groups; The product recommendation module 406 is used to determine the most suitable personalized product combination for the user from the product library based on similar health risk groups through a preset matching algorithm.

[0074] Furthermore, the dynamic indicator module 402 specifically includes: The preprocessing unit is used to preprocess the time series dataset, including outlier detection, missing value imputation, and data smoothing. The feature extraction unit is used to extract features from the preprocessed time series dataset to obtain multi-dimensional time series feature parameters; The dynamic modeling unit is used to dynamically model the user's physiological index sequence based on multi-dimensional time-series feature parameters and time-series modeling algorithms to obtain the prediction results of physiological index value changes. The trend prediction unit is used to calculate the rate and direction of change of physiological indicators based on the prediction results of changes in physiological indicator values, and generate a trend curve to describe the dynamic changes in individual health, thereby obtaining the dynamic trend of physiological indicator changes.

[0075] Further, please refer to Figure 5 The exception mode module 403 specifically includes: Threshold model unit 501 is used to obtain the standard reference range and individual baseline value of each physiological indicator and establish a threshold model for the normal range of the indicator. The threshold model is used to determine the degree of deviation of each physiological indicator. The comparative analysis unit 502 is used to compare and analyze the changing trends of dynamic physiological indicators with the threshold model, and to calculate the deviation of physiological indicators from the reference interval and the duration at each time point. Anomaly identification unit 503 is used to identify anomaly pattern types based on deviation magnitude and duration, wherein the anomaly pattern types include short-term fluctuation type, continuous rise type, continuous decrease type and periodic anomaly type; The quantitative scoring unit 504 is used to perform quantitative scoring on each abnormal pattern, and generates an abnormal quantitative score for the indicator based on the degree of deviation, frequency of occurrence and duration. The scoring weighting unit 505 is used to normalize and weight the abnormal quantitative scores of multiple physiological indicators to obtain the quantitative results of abnormal indicators.

[0076] Furthermore, the quantitative scoring unit 504 specifically includes: The deviation quantification subunit is used to perform numerical quantification calculation on the deviation of each physiological indicator and determine the deviation score based on the difference between the actual measured value and the central value of the reference interval. The frequency quantization subunit is used to count the frequency of abnormal events within a specified time window and calculate the frequency score based on the number of abnormal events and the time distribution pattern. The duration quantization subunit is used to detect the duration of abnormal states, converting the duration of continuous exceedance of the normal threshold into a duration score. The score-weighted fusion subunit is used to perform weighted fusion based on deviation score, frequency score, and duration score to obtain the quantitative score of indicator anomaly.

[0077] Furthermore, risk assessment module 404 specifically includes: The first encoding unit is used to acquire user family medical history information and perform feature encoding to obtain a family medical history risk vector. The second encoding unit is used to collect user medical behavior records and establish a medical behavior risk vector based on the time dimension and behavioral characteristics. The normalization unit is used to perform feature alignment and normalization on the quantitative results of abnormal indicators, the risk vector of family medical history, and the risk vector of medical behavior. The risk fusion calculation unit is used to perform fusion calculation on the normalized abnormal indicator quantification results, family medical history risk vector and medical behavior risk vector using a weighted fusion model, and output a preliminary health risk score. The weighted fusion model includes linear weighting, hierarchical weighting or machine learning-based adaptive weighting algorithm. The confidence assessment unit is used to calibrate the model and assess the confidence of the preliminary health risk score, correct for biases based on historical population sample data, and output the final comprehensive risk assessment score.

[0078] Furthermore, the risk fusion computing unit specifically includes: The parameter setting subunit is used to set initial weight parameters based on the medical relevance and importance of different risk factors. The initial weight parameters include physiological indicator weights, family genetic risk weights, and behavioral risk weights. The initial weight parameters are used to guide the weight allocation of the fusion model. The weighted product and feature interaction subunit is used to input the normalized abnormal indicator quantification results, family medical history risk vector and medical behavior risk vector into the weighted fusion model. The weighted fusion model performs weighted product and feature interaction calculations on each input factor to obtain the risk contribution value of each dimension and generate a weighted intermediate result matrix. The nonlinear transformation subunit is used to perform nonlinear transformation and normalization on the weighted intermediate results, and to suppress abnormal weight bias through activation functions or regularization mechanisms. The scoring output sub-unit is used to output a preliminary health risk score based on the weighted and fused multidimensional risk characteristics.

[0079] Furthermore, the group clustering module 405 specifically includes: The historical data unit is used to construct a historical user risk dataset, which includes a comprehensive risk assessment score, scores of major abnormal indicators, family medical history factors, and medical behavior characteristics, used to characterize individual health risk characteristics. The historical feature extraction unit is used to extract features from the historical user risk dataset to obtain a set of user risk feature vectors. The dimension reduction processing unit is used to standardize and reduce the dimension of the user risk feature vector set to obtain a dimension-reduced risk feature vector set. The clustering processing unit is used to select and train the clustering model. It inputs the comprehensive risk assessment score and the dimensionality-reduced risk feature vector set into the clustering model to obtain the clustering results. The similarity assessment unit is used to calculate the health risk center value and characteristic distribution of each group based on the clustering results, and to evaluate the Euclidean distance or cosine similarity between clustered samples to identify similar health risk groups.

[0080] In the above embodiments, this application discloses a product recommendation device, belonging to the field of artificial intelligence technology, applied to the recommendation scenario of health insurance products. First, it uses time-series analysis to accurately model the dynamic changes of users' physiological indicators, obtaining interpretable trends and abnormal patterns, thereby achieving early identification and quantitative assessment of health risks. Second, it performs traceable and interpretable multimodal fusion of the quantitative results of abnormal indicators with individualized multi-dimensional information (family medical history, medical behavior, etc.) to output a comprehensive health risk score. Third, through efficient clustering analysis of historical risk data, it forms a clear profile of similar risk groups, improving the robustness and scalability of group intervention and personalized matching. Finally, based on the matching algorithm for similar risk groups and a rich product library, it can output personalized product combinations highly consistent with individual health status, preferences, budgets, and intervention needs, possessing interpretability, traceability, and scalability, realizing personalized recommendations of health insurance products, and improving user satisfaction and conversion rates.

[0081] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 6 , Figure 6 This is a basic structural block diagram of the computer device in this embodiment.

[0082] The computer device 6 includes a memory 61, a processor 62, and a network interface 63 that are interconnected via a system bus. It should be noted that only the computer device 6 with memory 61, processor 62, and network interface 63 is shown in the figure; however, it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0083] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0084] The memory 61 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 61 may be an internal storage unit of the computer device 6, such as the hard disk or memory of the computer device 6. In other embodiments, the memory 61 may also be an external storage device of the computer device 6, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 6. Of course, the memory 61 may also include both the internal storage unit and its external storage device of the computer device 6. In this embodiment, the memory 61 is typically used to store the operating system and various application software installed on the computer device 6, such as computer-readable instructions for product recommendation methods. In addition, the memory 61 can also be used to temporarily store various types of data that have been output or will be output.

[0085] In some embodiments, the processor 62 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip. The processor 62 is typically used to control the overall operation of the computer device 6. In this embodiment, the processor 62 is used to execute computer-readable instructions stored in the memory 61 or to process data, such as executing computer-readable instructions for the product recommendation method.

[0086] The network interface 63 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 6 and other electronic devices.

[0087] This application also provides an embodiment, namely, a computer device including a memory and a processor. The memory stores computer-readable instructions, and the processor, when executing the computer-readable instructions, implements the steps of the product recommendation method described above, that is, implements: A product recommendation method, comprising: A time-series dataset is obtained by using users' historical physical examination records and current user health data. The time-series dataset includes a sequence of physiological indicator values ​​at consecutive time points. Time series analysis algorithms are applied to time series datasets to process changes in physiological index values, thereby obtaining dynamic trends in physiological index changes. Based on the dynamic trend of physiological indicator changes, the deviation range of physiological indicators is identified, and the abnormality pattern of indicators is determined according to the deviation range of physiological indicators, so as to obtain the quantitative results of abnormal indicators. The user's family medical history information and user medical behavior records are used as multi-dimensional factor inputs. The quantitative results of abnormal indicators are combined with multi-dimensional factors and processed by weighted fusion to obtain a comprehensive risk assessment score. Clustering algorithms were used to group and analyze comprehensive risk assessment scores and historical user risk data to identify similar health risk groups; Based on similar health risk groups, a pre-set matching algorithm is used to determine the most suitable personalized product combination from the product library for the user.

[0088] This application also provides another embodiment, namely, a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the product recommendation method described above, i.e., to implement: A product recommendation method, comprising: A time-series dataset is obtained by using users' historical physical examination records and current user health data. The time-series dataset includes a sequence of physiological indicator values ​​at consecutive time points. Time series analysis algorithms are applied to time series datasets to process changes in physiological index values, thereby obtaining dynamic trends in physiological index changes. Based on the dynamic trend of physiological indicator changes, the deviation range of physiological indicators is identified, and the abnormality pattern of indicators is determined according to the deviation range of physiological indicators, so as to obtain the quantitative results of abnormal indicators. The user's family medical history information and user medical behavior records are used as multi-dimensional factor inputs. The quantitative results of abnormal indicators are combined with multi-dimensional factors and processed by weighted fusion to obtain a comprehensive risk assessment score. Clustering algorithms were used to group and analyze comprehensive risk assessment scores and historical user risk data to identify similar health risk groups; Based on similar health risk groups, a pre-set matching algorithm is used to determine the most suitable personalized product combination from the product library for the user.

[0089] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0090] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0091] It should be noted that the software tools or components not belonging to this company that appear in the various embodiments of this application are merely illustrative examples and do not represent actual use.

[0092] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A product recommendation method, characterized in that, include: A time-series dataset is obtained by using the user's historical physical examination records and current user health data, wherein the time-series dataset includes a sequence of physiological indicator values ​​at consecutive time points; The time series dataset is processed using time series analysis algorithms to analyze changes in physiological index values, thereby obtaining dynamic trends in physiological index changes. Based on the dynamic physiological indicator change trend, the deviation range of the physiological indicator is identified, and the abnormal indicator pattern is determined according to the deviation range of the physiological indicator, so as to obtain the quantitative result of the abnormal indicator. The user's family medical history information and user medical behavior records are used as multi-dimensional factor inputs. The quantitative results of the abnormal indicators are combined with the multi-dimensional factors and weighted fusion processing is used to obtain a comprehensive risk assessment score. Clustering algorithms are used to group and analyze the comprehensive risk assessment scores and historical user risk data to identify similar health risk groups. Based on the similar health risk groups, a pre-set matching algorithm is used to determine the most suitable personalized product combination for the user from the product library.

2. The product recommendation method as described in claim 1, characterized in that, The step of applying a time-series analysis algorithm to the time-series dataset to process changes in physiological index values ​​and obtain dynamic trends in physiological index changes specifically includes: The time series dataset is preprocessed, wherein the preprocessing includes outlier detection, missing value imputation, and data smoothing. Feature extraction is performed on the preprocessed time series dataset to obtain multi-dimensional time series feature parameters; Based on the aforementioned multi-dimensional temporal feature parameters, a temporal modeling algorithm is used to dynamically model the user's physiological indicator sequence to obtain the prediction results of physiological indicator value changes. Based on the predicted results of the changes in the physiological indicators, the rate and direction of change of the physiological indicators are calculated, and a trend curve is generated to describe the dynamic changes in individual health, thereby obtaining the trend of the dynamic physiological indicators.

3. The product recommendation method as described in claim 1, characterized in that, The steps of identifying the deviation range of physiological indicators based on the changing trend of the dynamic physiological indicators, determining the abnormal pattern of the indicators based on the deviation range of the physiological indicators, and obtaining the quantitative results of the abnormal indicators specifically include: Obtain the standard reference range and individual baseline value of each physiological indicator, and establish a threshold model for the normal range of the indicators. The threshold model is used to determine the degree of deviation of each physiological indicator. The dynamic physiological index change trend is compared and analyzed with the threshold model to calculate the deviation and duration of the physiological index relative to the reference interval at each time point; The abnormal pattern type is identified based on the deviation magnitude and the duration, wherein the abnormal pattern type includes short-term fluctuation type, continuous rise type, continuous fall type and periodic abnormal type; Each abnormal pattern is quantitatively scored, and an abnormal quantitative score is generated based on the degree of deviation, frequency of occurrence, and duration. The abnormal quantitative scores of multiple physiological indicators are normalized and weighted to obtain the quantitative results of the abnormal indicators.

4. The product recommendation method as described in claim 3, characterized in that, The step of quantifying and scoring each abnormal pattern, and generating anomaly scores based on the degree of deviation, frequency of occurrence, and duration, specifically includes: The deviation of each physiological indicator is numerically quantified and calculated. The deviation score is determined based on the difference between the actual measured value and the central value of the reference interval. Statistically analyze the frequency of abnormal events within a specified time window, and calculate the frequency score based on the number of occurrences and the time distribution pattern of abnormal events; The duration of abnormal states is detected, and the length of time that continuously exceeds the normal threshold is converted into a duration score. The abnormality score of the indicator is obtained by weighting and fusing the deviation score, the frequency score, and the duration score.

5. The product recommendation method as described in claim 1, characterized in that, The step of obtaining user family medical history information and user medical behavior records as multi-dimensional factor inputs, and combining the quantitative results of the abnormal indicators with the multi-dimensional factors through weighted fusion processing to obtain a comprehensive risk assessment score, specifically includes: The user's family medical history information is obtained and its features are encoded to obtain a family medical history risk vector; Collect the user's medical behavior records and establish a medical behavior risk vector based on the time dimension and behavioral characteristics; The abnormal indicator quantification results, the family medical history risk vector, and the medical behavior risk vector are subjected to feature alignment and normalization. A weighted fusion model is used to fuse the normalized abnormal indicator quantification results, the family medical history risk vector, and the medical behavior risk vector to calculate a preliminary health risk score. The weighted fusion model includes linear weighting, hierarchical weighting, or machine learning-based adaptive weighting algorithms. The preliminary health risk score is model calibrated and confidence level assessed, biases are corrected based on historical population sample data, and the final comprehensive risk assessment score is output.

6. The product recommendation method as described in claim 5, characterized in that, The step of using a weighted fusion model to fuse the normalized abnormal indicator quantification results, the family medical history risk vector, and the medical behavior risk vector to output a preliminary health risk score specifically includes: Initial weight parameters are set according to the medical relevance and importance of different risk factors. The initial weight parameters include physiological indicator weights, family genetic risk weights, and behavioral risk weights. The initial weight parameters are used to guide the weight allocation of the fusion model. The normalized abnormal indicator quantification results, the family medical history risk vector, and the medical behavior risk vector are input into the weighted fusion model. The weighted fusion model is used to perform weighted product and feature interaction calculation on each input factor to obtain the risk contribution value of each dimension and generate a weighted intermediate result matrix. The weighted intermediate results are subjected to nonlinear transformation and normalization, and abnormal weight biases are suppressed by activation functions or regularization mechanisms. A preliminary health risk score is output based on the weighted and fused multidimensional risk characteristics.

7. The product recommendation method as described in claim 1, characterized in that, The step of using a clustering algorithm to group and analyze the comprehensive risk assessment score and historical user risk data to determine similar health risk groups specifically includes: Construct a historical user risk dataset, which includes a comprehensive risk assessment score, scores of major abnormal indicators, family medical history factors, and medical behavior characteristics, to characterize individual health risk characteristics; Feature extraction is performed on the historical user risk dataset to obtain a user risk feature vector set; The user risk feature vector set is standardized and dimensionality reduced to obtain a dimensionality-reduced risk feature vector set; Select and train a clustering model, input the comprehensive risk assessment score and the dimensionality-reduced risk feature vector set into the clustering model, and obtain the clustering results; Based on the clustering results, the health risk center value and characteristic distribution of each group are calculated, and the Euclidean distance or cosine similarity between clustered samples is evaluated to determine the similar health risk groups.

8. A product recommendation device, characterized in that, include: The time-series module is used to obtain a time-series dataset through the user's historical physical examination records and current user health data, wherein the time-series dataset includes a sequence of physiological indicator values ​​at consecutive time points; The dynamic indicator module is used to apply time series analysis algorithms to the time series dataset to process changes in physiological indicator values ​​and obtain dynamic trends in physiological indicator changes. An abnormal mode module is used to identify the deviation range of the physiological indicators based on the changing trend of the dynamic physiological indicators, and to determine the abnormal mode of the indicators according to the deviation range of the physiological indicators, so as to obtain the quantitative result of the abnormal indicators. The risk assessment module is used to obtain user family medical history information and user medical behavior records as multi-dimensional factor inputs. The quantitative results of the abnormal indicators are combined with the multi-dimensional factors and weighted fusion processing is used to obtain a comprehensive risk assessment score. The group clustering module is used to perform group analysis on the comprehensive risk assessment score and historical user risk data using a clustering algorithm to identify similar health risk groups; The product recommendation module is used to determine the most suitable personalized product combination for the user from the product library based on the similar health risk groups using a preset matching algorithm.

9. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the product recommendation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions that, when executed by a processor, implement the steps of the product recommendation method as described in any one of claims 1 to 7.

Images