Group insurance product recommendation algorithm, device and computer equipment based on big data

Abstract

The application discloses a group insurance product recommendation algorithm and device based on big data and a computer device, and particularly relates to the field of insurance technology, and comprises feature fusion and standardization, space-time adaptation degree calculation, group preference propagation, dynamic demand prediction, multi-objective optimization sorting and feedback self-adaptation.The application realizes deep integration of multi-source data through multi-modal feature fusion and encryption standardization, dynamically captures service coverage radius change and policy period attenuation characteristics by using a three-dimensional space-time quantization model, improves recommendation timeliness in combination with a group preference propagation mechanism, cooperates enterprise profit and customer utility targets by means of an innovatively designed double-target Pareto optimization strategy, dynamically adjusts space-time adaptation coefficients and propagation weights by means of real-time collection of business indexes, and forms a closed-loop feedback system.The technology effectively solves the problems of single data dimension, static rule lag and interest imbalance in traditional methods, and has the advantages of privacy protection, dynamic response and multi-objective balance.

Description

Technical Field

[0001] This invention relates to the field of insurance technology, and more specifically, to a group insurance product recommendation algorithm, device, and computer equipment based on big data. Background Technology

[0002] Intelligent recommendation of group insurance products is an important research direction in the field of insurance technology. It aims to achieve precise product matching through data analysis. Existing technologies usually adopt a static matching mechanism between basic customer information and product terms. Its operation process consists of three steps: First, extracting structured data such as customer occupation type, company size, and historical policies from the business system; second, calculating the matching degree between products and customer tags through a predefined rule base; and finally generating a recommendation list based on a single dimension and relying on manual underwriting to adjust the results.

[0003] The existing technology has three significant drawbacks: First, it lacks the ability to process multi-source heterogeneous data, and cannot effectively integrate unstructured data such as spatial location and temporal behavior, resulting in a single dimension of feature representation; second, it lacks a dynamic spatiotemporal adaptation mechanism, and does not consider the impact of service outlet coverage radius, time-varying characteristics of occupational risks, and policy cycle decay patterns on recommendation effectiveness; third, it adopts a fixed weight ranking strategy, which makes it difficult to coordinate the balance between insurance institutions' profit targets and customers' actual needs, and lacks online feedback and optimization capabilities. Summary of the Invention

[0004] To overcome the aforementioned deficiencies in the prior art, this invention provides a group insurance product recommendation algorithm, apparatus, and computer equipment based on big data. The following solutions address the problems mentioned in the background art, such as insufficient processing of multi-source heterogeneous data, lack of dynamic spatiotemporal adaptation mechanism, difficulty in balancing the interests of multiple parties with fixed weight strategies, and lack of online feedback and optimization capabilities.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a group insurance product recommendation algorithm based on big data, comprising:

[0006] S1: Feature Fusion and Standardization: Integrate heterogeneous data sources such as product database, customer ledger and salesperson profile, construct cross-domain feature matrix, convert geographical location into raster coordinates through spatial encoding, eliminate dimensional differences by standardization method, encrypt sensitive fields, and output a set of feature vectors with unified spatiotemporal granularity.

[0007] S2: Spatiotemporal Adaptability Calculation: Comprehensively evaluate the coverage and matching degree between the customer's geographical location and the scope of product services, the compatibility between occupational risk categories and insurance terms, and the fit between the policy period and time sensitivity to establish a three-dimensional spatial-attribute-time customer product adaptability quantitative model.

[0008] S3: Group Preference Propagation: Based on customer feature similarity calculation and behavioral pattern correlation analysis, a customer relationship graph is constructed. Based on graph neural network, a preference propagation equation is designed to spread explicit purchase behavior data along the network topology in multiple hops, mine potential group selection preferences, and form an implicit demand correlation mapping between customer nodes.

[0009] S4: Dynamic Demand Forecasting: By integrating historical renewal cycle patterns and the influencing factors of real-time market events, a time series forecasting model is constructed to analyze the fluctuation characteristics of customers' potential insurance demand at different time points.

[0010] S5: Multi-objective optimization ranking: Construct a dual-objective optimization function that takes into account both corporate profit goals and customer utility value, set differentiated strategy parameters according to user role type, and generate a recommended product ranking sequence that satisfies the balance of interests of all parties;

[0011] S6: Feedback Adaptation: By collecting click-through rate and conversion rate data of recommendation results in real time, an online learning mechanism is established to dynamically adjust model parameters and achieve continuous iterative optimization of recommendation strategies.

[0012] Preferred, a group insurance product recommendation device based on big data includes:

[0013] Feature Fusion and Standardization Module: Integrates heterogeneous data sources such as product library, customer ledger and salesperson profile, constructs cross-domain feature matrix, converts geographical location into raster coordinates through spatial encoding, eliminates dimensional differences through standardization methods, encrypts sensitive fields, and outputs a set of feature vectors with unified spatiotemporal granularity.

[0014] Spatiotemporal Adaptability Calculation Module: Comprehensively evaluates the coverage and matching degree between customer's geographical location and product service scope, the compatibility between occupational risk category and insurance terms, and the degree of fit between policy period and time sensitivity, and establishes a three-dimensional spatial-attribute-time customer product adaptability quantitative model;

[0015] Group Preference Propagation Module: Based on customer feature similarity calculation and behavioral pattern association analysis, a customer relationship graph is constructed. Based on graph neural network, a preference propagation equation is designed to spread explicit purchase behavior data along the network topology in multiple hops, mine potential group selection preferences, and form an implicit demand association mapping between customer nodes.

[0016] Dynamic demand forecasting module: Integrates historical renewal cycle patterns and real-time market event influencing factors to build a time series forecasting model and analyze the fluctuation characteristics of customers' potential insurance demand at different time points;

[0017] Multi-objective optimization ranking module: Constructs a dual-objective optimization function that takes into account both corporate profit goals and customer utility value, sets differentiated strategy parameters according to user role type, and generates a recommended product ranking sequence that satisfies the balance of interests of all parties;

[0018] Feedback Adaptive Module: By collecting click-through rate and conversion rate data of recommendation results in real time, an online learning mechanism is established to dynamically adjust model parameters and achieve continuous iterative optimization of recommendation strategies.

[0019] Preferably, a computer device stores a group insurance product recommendation algorithm based on big data, which can be loaded and executed by a processor. The medium of the computer device includes various media that can store program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0020] The technical effects and advantages of this invention are as follows:

[0021] This solution integrates heterogeneous data from product databases, customer ledgers, and salesperson profiles through multimodal feature fusion and encrypted standardization. It converts geographical locations into a computable latitude and longitude matrix using spatial rasterization coding and combines dynamic salt hash encryption to improve data consistency while protecting privacy. This solution solves the problem of incomplete feature representation caused by the single data dimension in traditional methods.

[0022] The spatiotemporal adaptation mechanism based on the three-dimensional quantitative model innovatively integrates spatial compatibility index, occupational risk matching degree and behavioral cycle fit degree. By constructing a customer relationship map, it realizes the propagation of group preferences, effectively captures the dynamic changes in the service outlet coverage, the time-varying law of occupational risk coefficient and the decay characteristics of the policy effective period, and overcomes the shortcomings of static matching mechanism in responding to spatiotemporal elements with lag.

[0023] By employing a strategy that combines a dual-objective optimization function with online feedback tuning, the goal of balancing corporate profit and customer utility is achieved within the Pareto optimal solution set. By dynamically adjusting the spatiotemporal adaptation coefficient, propagation weight, and optimization parameters through real-time collection of business indicators, a closed-loop adaptive system is formed, overcoming the technical bottleneck of traditional fixed-weight strategies that struggle to balance the interests of multiple parties and lack self-optimization. Attached Figure Description

[0024] Fig. 1 This is a schematic diagram of the overall process structure of the present invention.

[0025] Fig. 2 This is a schematic diagram of the complete embodiment of the present invention.

[0026] Fig. 3 This is a schematic diagram of the data flow topology of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] refer to Figs. 1-3 The group insurance product recommendation algorithm based on big data shown includes...

[0029] S1: Feature Fusion and Standardization: Integrate heterogeneous data sources such as product database, customer ledger and salesperson profile, construct cross-domain feature matrix, convert geographical location into raster coordinates through spatial encoding, eliminate dimensional differences by standardization method, encrypt sensitive fields, and output a set of feature vectors with unified spatiotemporal granularity.

[0030] The S1 integrates three types of raw inputs: product feature vectors, customer feature vectors, and behavioral time-series data. It converts geographical locations into latitude and longitude coordinates through spatial rasterization encoding, eliminates numerical differences between features of different dimensions using Z-score standardization, implements dynamic salt hash encryption for sensitive fields, and uses time window alignment technology to unify the temporal granularity of multi-source data, generating a unified feature matrix that includes spatial encoding, standardized values, and desensitized fields.

[0031] The product feature vector is extracted from the structured data storage of the product table. A full data synchronization is performed daily at midnight by a scheduled task system. Database snapshot technology is used to capture the complete status of product parameters, and database triggers are used to monitor product information change events in real time. The customer feature vector is constructed based on the historical transaction records of the ledger table. Data is incrementally extracted from the business database every hour using ETL tools, and customer status changes are captured by transaction log parsing technology. The behavioral time-series data is extracted from the effective date and renewal operation timestamp fields of the ledger table, as well as the mass mailing records and tracking logs of the salesperson table.

[0032] The product feature vector is represented as follows: The customer feature vector is represented as: The behavioral time series data is represented as follows: The standardization method is as follows: x i μ represents the original characteristic element. x σ represents the mean of the field. x The standard deviation is represented; the dynamic salt hash encryption is represented as: Salt represents a dynamic salt value, updated every 24 hours according to the system time; Hash represents a composite hash function; and s represents the plaintext of the sensitive field. This indicates a bitwise XOR operation.

[0033] S2: Spatiotemporal Adaptability Calculation: Comprehensively assess the coverage and matching degree between the customer's geographical location and the scope of product services, the compatibility between occupational risk categories and insurance terms, and the degree of fit between the policy period and time sensitivity to establish a three-dimensional spatial-attribute-time customer product adaptability quantitative model.

[0034] The S2 uses a spatial compatibility index to assess the degree of matching between the customer's geographical location and the coverage radius of the product service outlets, uses occupational risk matching to measure the compatibility between the customer's occupational category and the underwriting terms of the insurance product, and combines behavioral cycle fit analysis to analyze the correlation between the customer's policy cycle and the current time node, thus constructing a multi-dimensional dynamic adaptation assessment system to comprehensively judge the spatiotemporal matching characteristics between the customer and the insurance product.

[0035] The spatial compatibility index is calculated by obtaining the customer's geographical location coordinates from the ledger data and the service outlet coordinates marked in the product table, calculating the spherical distance between the two, normalizing the actual distance value with the maximum service radius of the product's historical underwriting cases, and finally generating a matching metric value in the range of 0-1. The closer the value is to 1, the higher the matching degree between the customer's geographical location and the service coverage of the product.

[0036] The spatial compatibility index is specifically expressed as follows: L c L represents the customer's coordinates. p R represents the coordinates of product service outlets. max Indicates the product's maximum coverage radius.

[0037] The occupational risk matching degree is determined by comparing the customer's occupational category with the occupational category requirements specified in the insurance product's underwriting terms. Occupational categories that fully meet the coverage scope are assigned a full matching value, while occupational categories with overlapping risk levels but not explicitly excluded are assigned a partial matching value. Finally, the weighted average of the matching results for all occupational categories is taken.

[0038] The occupational risk matching degree is specifically expressed as follows: o c Indicates the customer's actual occupational category, o p This represents the occupational categories allowed under the insurance product's underwriting terms, where j represents the j-th occupational category, δ represents the occupational category matching function (δ=1 for a perfect match, δ=0.5 for a partial match), and J represents the number of occupational categories required by the product.

[0039] The behavioral cycle fit is calculated based on the date difference between the policy validity period and the current system time to determine time sensitivity. An exponential decay model is used to quantify the timeliness characteristics of customer renewal needs, where the decay coefficient is set according to the historical renewal cycle distribution pattern.

[0040] The behavioral cycle fit is specifically expressed as follows: d expire Indicates the expiration date of the customer's current policy, d current This represents the current date when the system performs the calculation, and λ represents the attenuation coefficient.

[0041] S3: Group Preference Propagation: Based on customer feature similarity calculation and behavioral pattern association analysis, a customer relationship graph is constructed. Based on graph neural network, a preference propagation equation is designed to spread explicit purchase behavior data along the network topology in multiple hops, mine potential group selection preferences, and form an implicit demand association mapping between customer nodes.

[0042] The customer relationship graph is based on the cosine similarity of customer static feature vectors to calculate the basic correlation. The cosine similarity of time-series behavior vectors is introduced as a dynamic correction term. The two types of similarity are weighted and fused through the time-series weight coefficient optimized by grid search to construct the comprehensive correlation weight between customer nodes. Static features reflect the degree of matching of inherent customer attributes, time-series behavior captures the similarity of dynamic interaction patterns, and the weight coefficient balances the contribution ratio of the two types of information.

[0043] The customer relationship graph is specifically represented as follows: w ij C represents the composite similarity weight in a customer relationship network. i / C j T represents the customer's static feature vector. i / T j Let α represent the temporal behavior vector, and let α represent the temporal behavior weight coefficient. The value of α ranges from 0 to 1, and α = 0.35 is determined through grid search.

[0044] The preference propagation equation constructs a weighted relationship graph based on customer feature similarity and behavioral correlation. The node connection strength is represented by the Laplace matrix. The initial purchase probability matrix and the identity matrix are offset by inverse compensation. The propagation coefficient set by experience controls the information diffusion range. Finally, the linear equation system is solved to obtain the steady-state preference distribution.

[0045] The preference propagation equation is specifically expressed as follows: P new Let P represent the preference probability matrix, L represent the Laplace matrix, β represent the propagation coefficient, and P represent the probability matrix. init This represents the initial purchase probability matrix.

[0046] The wij As elements of the edge weight matrix in the customer relationship graph, P encodes the explicit similarity between customers, which is then transformed into structural constraints in the topological space using the Laplace matrix L; new This is the steady-state solution under the constraint, preserving the initial purchasing tendency P. init Based on this, the preference values ​​of similar customer nodes are mutually permeated through weighted connection channels, ultimately forming a new probability distribution that reflects both individual characteristics and group influence.

[0047] S4: Dynamic Demand Forecasting: By integrating historical renewal cycle patterns and real-time market event influencing factors, a time series forecasting model is constructed to analyze the fluctuation characteristics of customers' potential insurance demand at different time points.

[0048] The S4 establishes basic demand forecasting by analyzing the time series patterns of customers' historical insurance behavior through a baseline prediction model. It introduces event impact factors to quantify the immediate disturbance effect of external events on insurance demand, combines time series trend forecasting with emergency response, and constructs a dynamic demand forecasting model with environmental awareness, which outputs the intensity of customers' potential insurance demand in different scenarios in real time.

[0049] The baseline prediction model is based on historical policy effective dates and renewal cycle data. It establishes a basic prediction framework using autoregressive time series analysis, determines the optimal lag period through autocorrelation function analysis, uses the customer's historical insurance application time and renewal interval as core input features, and introduces real-time events as external intervention variables. It uses maximum likelihood estimation to calculate the regression coefficients of each feature and constructs a highly interpretable demand prediction benchmark model.

[0050] The baseline prediction model is specifically represented as follows: , y represents the predicted demand value at time t. t-1 y represents the actual demand observation value one day ago. t-7 This represents the actual demand observation value 7 days ago, where φ1 and φ2 represent y. t-1 and y t-7 The autoregressive coefficients ε at two historical points in time t This represents the random error term.

[0051] The time-based impact factor defines market policy changes and marketing campaign events, quantifies the average impact of different events on insurance demand based on historical data regression analysis, uses a Boolean identifier function within the event occurrence time window to capture the event status in real time, linearly weights and sums the impact coefficients of each event type with its occurrence status, and dynamically adjusts the baseline forecast value to reflect the impact of sudden market factors on customer demand.

[0052] The time-related factor is specifically expressed as follows: Dy represents the incremental impact of external events on insurance demand, and γ represents the value of the impact of external events on insurance demand. k Indicates the intensity of an event's impact; when the event is a policy adjustment, γ... k =1.2, when the event is a promotional activity, γ k =0.8, I(event) k ) represents the event monitoring function.

[0053] The S4 is passed Together with Dy, they constitute a two-level correction mechanism for the prediction system, specifically expressed as follows: Y t This represents the final demand forecast.

[0054] S5: Multi-objective optimization ranking: Construct a dual-objective optimization function that takes into account both corporate profit goals and customer utility value, set differentiated strategy parameters according to user role types, and generate a recommended product ranking sequence that satisfies the balance of interests of all parties.

[0055] The S5 constructs a dual objective function that simultaneously optimizes the enterprise's profit target and the customer's utility value, designs constraints including a maximum recommendation quantity limit and product selection variable constraints, and uses the Pareto front solution method to generate a recommended product ranking sequence that satisfies the balance between business strategy and customer needs. The objective weights are dynamically configured according to the user role type.

[0056] The dual objective function is based on the enterprise's historical premium profit data to construct a profit maximization objective function, and at the same time, it combines the client's product suitability index to establish a utility value evaluation function. The two types of objectives are integrated into a unified optimization framework through parameterized weighting. The weight parameters are dynamically configured according to the business scenario, and additional restrictions on the number of recommendations and product mutual exclusion constraints are added to form a complete mathematical programming model. Finally, a non-dominated sorting algorithm is used to generate the recommendation solution set.

[0057] The dual objective function is specifically expressed as follows: Profit represents the profit target, Utility represents the utility target, and θ represents the adjustment parameter, determined according to the role type mapping table: back-office staff = 0.7, salesperson = 0.3, r i u represents the unit profit contribution value of the i-th insurance product, calculated by subtracting operating costs from the product's premium income in historical ledger data. j Let x represent the customer utility score for the j-th insurance product. i / x j Denotes the decision variable, when x i When x = 1, it means recommending the i-th product. j =1 indicates that the j-th product is recommended.

[0058] The u jSpecifically, it is expressed as follows: Dt represents the time interval between the current time and the baseline point for customer demand forecasting.

[0059] The constraint design is based on actual business rules and system processing capabilities. The risk of information overload is controlled by setting a maximum recommendation limit, and the threshold is determined comprehensively based on the user interface capacity and customer decision-making efficiency. The binary variable constraint of product selection is adopted to ensure the discretized decision-making characteristics of the recommendation results, with each product having only two clear states: recommended or not recommended.

[0060] The constraints are specifically expressed as follows: N max This indicates the maximum number of recommendations.

[0061] S6: Feedback Adaptation: By collecting click-through rate and conversion rate data of recommendation results in real time, an online learning mechanism is established to dynamically adjust model parameters and achieve continuous iterative optimization of recommendation strategies.

[0062] The S6 uses real-time tracking of the click-through rate and conversion rate of recommended products as core performance evaluation indicators, establishes an online learning mechanism based on gradient descent, and dynamically adjusts the decay coefficient in spatiotemporal adaptation calculation, the temporal behavior weight coefficient in group preference propagation, and the adjustment parameters in multi-objective optimization according to changes in the indicators. This enables automatic iterative optimization of recommendation strategy parameters. The parameter update process uses a sliding window mechanism to retain the timeliness characteristics of recent behavioral data, and corrects the calculation parameters of each process through backpropagation of the loss function, forming a closed-loop feedback adjustment.

[0063] The performance evaluation metrics construct a two-dimensional evaluation system by collecting real-time user click behavior data and insurance conversion data for recommended products. The click-through rate (CTR) metric calculates the ratio of the number of times recommended products are exposed to the number of times users actually click on them to measure the attractiveness of the content. The conversion rate metric counts the percentage of users who complete the insurance purchase process from viewing product details to assess the accuracy of demand matching. The data sources for the two types of metrics are front-end interaction logs and transaction records from the ledger system, respectively.

[0064] The specific performance evaluation indicators include: , CTR stands for Click-through rate, Conversion stands for Conversion rate, Clicks stands for Number of clicks, Impressions stands for Number of impressions, Details_Views stands for Number of details viewed, and Purchases stands for Number of purchases.

[0065] The parameter updates are based on real-time collected click-through rate and conversion rate data of recommendation results. The direction of parameter adjustment is determined by calculating the partial derivatives of the performance indicators with respect to the model parameters. The update step size is controlled by a dynamic learning rate. The learning rate decays exponentially over time to balance the convergence speed and stability. At the same time, parameter threshold boundary constraints are set to prevent over-adjustment. A sliding window smoothing mechanism is activated for abnormal fluctuation data.

[0066] The update rules for the parameters are specifically expressed as follows: α new Indicates the updated parameter value, α old Indicates the current parameter value. η represents the partial derivative of the click-through rate with respect to the parameter, and η represents the learning rate, which is initially 0.01 and decays over time.

[0067] This application also discloses a group insurance product recommendation device based on big data, including:

[0068] Feature Fusion and Standardization Module: Integrates heterogeneous data sources such as product library, customer ledger and salesperson profile, constructs cross-domain feature matrix, converts geographical location into raster coordinates through spatial encoding, eliminates dimensional differences through standardization methods, encrypts sensitive fields, and outputs a set of feature vectors with unified spatiotemporal granularity.

[0069] Spatiotemporal Adaptability Calculation Module: Comprehensively evaluates the coverage and matching degree between customer's geographical location and product service scope, the compatibility between occupational risk category and insurance terms, and the degree of fit between policy period and time sensitivity, and establishes a three-dimensional spatial-attribute-time customer product adaptability quantitative model;

[0070] Group Preference Propagation Module: Based on customer feature similarity calculation and behavioral pattern association analysis, a customer relationship graph is constructed. Based on graph neural network, a preference propagation equation is designed to spread explicit purchase behavior data along the network topology in multiple hops, mine potential group selection preferences, and form an implicit demand association mapping between customer nodes.

[0071] Dynamic demand forecasting module: Integrates historical renewal cycle patterns and real-time market event influencing factors to build a time series forecasting model and analyze the fluctuation characteristics of customers' potential insurance demand at different time points;

[0072] Multi-objective optimization ranking module: Constructs a dual-objective optimization function that takes into account both corporate profit goals and customer utility value, sets differentiated strategy parameters according to user role type, and generates a recommended product ranking sequence that satisfies the balance of interests of all parties;

[0073] Feedback Adaptive Module: By collecting click-through rate and conversion rate data of recommendation results in real time, an online learning mechanism is established to dynamically adjust model parameters and achieve continuous iterative optimization of recommendation strategies.

[0074] This application also discloses a computer device that stores a group insurance product recommendation algorithm based on big data, which can be loaded and executed by a processor. The computer device medium includes various media that can store program code, such as USB flash drive, mobile hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk.

[0075] This invention first integrates heterogeneous data sources such as product databases, customer ledgers, and salesperson profiles. It converts geographic locations into latitude and longitude coordinates through spatial rasterization encoding, eliminates dimensional differences using Z-score standardization, and implements dynamic salt-value hash encryption to generate a feature matrix with unified spatiotemporal granularity. Next, it calculates the spherical distance matching degree between customer geographic location and product service outlets, the matching weight between occupation category and underwriting terms, and the exponential decay fit between policy validity and current time using a three-dimensional quantification model. Subsequently, it constructs a customer relationship graph based on feature similarity and temporal behavioral correlation, and designs a preference propagation equation using a weighted Laplace matrix to reflect initial purchases. Probability spreads along the network topology to form a group preference distribution; then, an autoregressive time series model and event influence factors are integrated to perform dynamic demand prediction, generating demand intensity values ​​that include baseline trends and event disturbances; subsequently, a bi-objective optimization function is constructed and a maximum recommendation quantity constraint is applied, generating a ranking sequence that balances enterprise profits and customer utility through Pareto front solving; finally, click-through rate and conversion rate data are collected in real time, and the spatiotemporal adaptation decay coefficient, preference propagation weight, and multi-objective parameters are adjusted online through gradient descent to form a closed-loop feedback optimization, completing the entire recommendation process from data preprocessing, adaptation evaluation, group propagation, demand prediction, optimized ranking to parameter adaptation.

[0076] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0077] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A group insurance product recommendation algorithm based on big data, characterized in that, include: S1: Feature Fusion and Standardization: Integrate heterogeneous data sources such as product database, customer ledger and salesperson profile, construct cross-domain feature matrix, convert geographical location into raster coordinates through spatial encoding, eliminate dimensional differences by standardization method, encrypt sensitive fields, and output a set of feature vectors with unified spatiotemporal granularity. S2: Spatiotemporal Adaptability Calculation: Comprehensively evaluate the coverage and matching degree between the customer's geographical location and the scope of product services, the compatibility between occupational risk categories and insurance terms, and the fit between the policy period and time sensitivity to establish a three-dimensional spatial-attribute-time customer product adaptability quantitative model. S3: Group Preference Propagation: Based on customer feature similarity calculation and behavioral pattern correlation analysis, a customer relationship graph is constructed. Based on graph neural network, a preference propagation equation is designed to spread explicit purchase behavior data along the network topology in multiple hops, mine potential group selection preferences, and form an implicit demand correlation mapping between customer nodes. The customer relationship graph calculates the basic association degree based on the cosine similarity of customer static feature vectors, and introduces the cosine similarity of temporal behavior vectors as a dynamic correction term. The two types of similarity are weighted and fused using temporal weight coefficients optimized by grid search to construct a comprehensive association weight between customer nodes. Static features reflect the degree of matching of customers' inherent attributes, time-series behavior captures the similarity of dynamic interaction patterns, and weight coefficients balance the contribution ratio of the two types of information. The preference propagation equation constructs a weighted relationship graph based on customer feature similarity and behavioral correlation. The node connection strength is represented by the Laplace matrix. The initial purchase probability matrix and the identity matrix are offset by inverse compensation. The propagation coefficient set by experience controls the information diffusion range. Finally, the linear equation system is solved to obtain the steady-state preference distribution. S4: Dynamic Demand Forecasting: By integrating historical renewal cycle patterns and the influencing factors of real-time market events, a time series forecasting model is constructed to analyze the fluctuation characteristics of customers' potential insurance demand at different time points. S5: Multi-objective optimization ranking: Construct a dual-objective optimization function that takes into account both corporate profit goals and customer utility value, set differentiated strategy parameters according to user role type, and generate a recommended product ranking sequence that satisfies the balance of interests of all parties; S6: Feedback Adaptive: By collecting click-through rate and conversion rate data of recommendation results in real time, an online learning mechanism is established to dynamically adjust model parameters and achieve continuous iterative optimization of recommendation strategies; The S1 integrates three types of raw inputs: product feature vector, customer feature vector, and behavioral time series data. It converts geographical location into latitude and longitude coordinates through spatial rasterization encoding, eliminates numerical differences between features of different dimensions using Z-score standardization, implements dynamic salt value hash encryption for sensitive fields, and uses time window alignment technology to unify the time series granularity of multi-source data, generating a unified feature matrix that includes spatial encoding, standardized values, and desensitized fields. The product feature vector is extracted from the structured data storage of the product table. A full data synchronization is performed daily at midnight via a scheduled task system, and database snapshot technology is used to capture the complete state of product parameters. Simultaneously, database triggers are used to monitor product information change events in real time. The customer feature vector is constructed based on historical transaction records in the ledger table. Data is incrementally extracted from the business database hourly using ETL tools, combined with transaction log parsing technology to capture changes in customer status. The behavioral time-series data extracts time-series information from the effective date and renewal operation timestamp fields of the ledger table, as well as from the salesperson table's mass messaging records and tracking logs. The S2 uses a spatial compatibility index to assess the degree of matching between the customer's geographical location and the coverage radius of the product service outlets, uses occupational risk matching to measure the compatibility between the customer's occupational category and the insurance product underwriting terms, and combines behavioral cycle fit to analyze the correlation between the customer's policy cycle and the current time node, thus constructing a multi-dimensional dynamic adaptation assessment system to comprehensively judge the spatiotemporal matching characteristics between the customer and the insurance product. The spatial compatibility index is calculated by obtaining the customer's geographical location coordinates from the ledger data and the service outlet coordinates marked in the product table, calculating the spherical distance between the two, normalizing the actual distance value with the maximum service radius of the product's historical underwriting cases, and finally generating a matching metric value in the range of 0-1. The closer the value is to 1, the higher the matching degree between the customer's geographical location and the service coverage of the product. The occupational risk matching degree is obtained by comparing the customer's occupational category with the occupational category requirements specified in the insurance product's underwriting terms. Occupational categories that fully meet the coverage scope are assigned a full matching value, while occupational categories with overlapping risk levels but not explicitly excluded are assigned a partial matching value. Finally, the weighted average of the matching results of all occupational categories is taken. The behavioral cycle fit is calculated based on the date difference between the policy validity period and the current system time to determine time sensitivity. An exponential decay model is used to quantify the timeliness characteristics of customer renewal needs, where the decay coefficient is set according to the historical renewal cycle distribution pattern. The S4 establishes basic demand forecasting by analyzing the time series patterns of customers' historical insurance behavior through a baseline prediction model, introduces event impact factors to quantify the immediate disturbance effect of external events on insurance demand, combines time series trend forecasting with emergency response, constructs a dynamic demand forecasting model with environmental awareness, and outputs the intensity of customers' potential insurance demand in different scenarios in real time. The baseline prediction model is based on historical policy effective dates and renewal cycle data. It uses autoregressive time series analysis to establish a basic prediction framework, determines the optimal lag period through autocorrelation function analysis, takes the customer's historical insurance application time and renewal interval as core input features, introduces real-time events as external intervention variables, and uses maximum likelihood estimation to calculate the regression coefficients of each feature to construct a demand prediction benchmark model with strong interpretability. The time impact factor defines market policy changes and marketing campaign events, quantifies the average impact of different events on insurance demand based on historical data regression analysis, uses a Boolean identifier function within the event occurrence time window to capture the event status in real time, linearly weights and sums the impact coefficients of each event type with their occurrence status, and dynamically adjusts the baseline forecast value to reflect the impact of sudden market factors on customer demand. The S5 constructs a dual objective function that simultaneously optimizes the enterprise's profit target and the customer's utility value, designs constraints including the maximum number of recommendations and product selection variable constraints, and uses the Pareto front method to generate a recommended product ranking sequence that satisfies the balance between business strategy and customer needs. The objective weights are dynamically configured according to the user role type. The dual objective function is based on the historical premium profit data of the enterprise to construct a profit maximization objective function, and at the same time, it combines the product suitability index of the client to establish a utility value evaluation function. The two types of objectives are integrated into a unified optimization framework through parameterized weighting. The weight parameters are dynamically configured according to the business scenario, and additional restrictions on the number of recommendations and product mutual exclusion constraints are added to form a complete mathematical programming model. Finally, a non-dominated sorting algorithm is used to generate the recommendation solution set. The constraint design is based on actual business rules and system processing capabilities. The risk of information overload is controlled by setting a maximum recommendation limit. The threshold is determined comprehensively based on the user interface carrying capacity and customer decision-making efficiency. The product selection binary variable constraint ensures the discretized decision-making characteristics of the recommendation results, with each product having only two clear states: recommended or not recommended. The S6 uses real-time tracking of the click-through rate and conversion rate of recommended products as the core performance evaluation indicators, establishes an online learning mechanism based on gradient descent, and dynamically adjusts the decay coefficient in spatiotemporal adaptation calculation, the temporal behavior weight coefficient in group preference propagation, and the adjustment parameters in multi-objective optimization according to the changes in the indicators. This enables automatic iterative optimization of the recommendation strategy parameters. The parameter update process uses a sliding window mechanism to retain the timeliness characteristics of recent behavioral data, and corrects the calculation parameters of each process through backpropagation of the loss function, forming a closed-loop feedback adjustment. The performance evaluation metrics construct a two-dimensional evaluation system by collecting real-time user click behavior data and insurance conversion data for recommended products. The click-through rate metric calculates the ratio of the number of times the recommended product is exposed to the number of times the user actually clicks to measure the attractiveness of the content. The conversion rate metric counts the conversion ratio from viewing product details to completing the insurance purchase to evaluate the accuracy of demand matching. The data sources for the two types of metrics are front-end interaction logs and transaction records from the ledger system, respectively. The parameter updates are based on real-time collected click-through rate and conversion rate data of recommendation results. The direction of parameter adjustment is determined by calculating the partial derivatives of the performance indicators with respect to the model parameters. The update step size is controlled by a dynamic learning rate. The learning rate decays exponentially over time to balance the convergence speed and stability. At the same time, parameter threshold boundary constraints are set to prevent over-adjustment. A sliding window smoothing mechanism is activated for abnormal fluctuation data.

2. A group insurance product recommendation device based on big data, used to implement the group insurance product recommendation algorithm based on big data as described in claim 1, characterized in that, include: Feature Fusion and Standardization Module: Integrates heterogeneous data sources such as product library, customer ledger and salesperson profile, constructs cross-domain feature matrix, converts geographical location into raster coordinates through spatial encoding, eliminates dimensional differences through standardization methods, encrypts sensitive fields, and outputs a set of feature vectors with unified spatiotemporal granularity. Spatiotemporal Adaptability Calculation Module: Comprehensively evaluates the coverage and matching degree between customer's geographical location and product service scope, the compatibility between occupational risk category and insurance terms, and the degree of fit between policy period and time sensitivity, and establishes a three-dimensional spatial-attribute-time customer product adaptability quantitative model; Group Preference Propagation Module: Based on customer feature similarity calculation and behavioral pattern association analysis, a customer relationship graph is constructed. Based on graph neural network, a preference propagation equation is designed to spread explicit purchase behavior data along the network topology in multiple hops, mine potential group selection preferences, and form an implicit demand association mapping between customer nodes. Dynamic demand forecasting module: Integrates historical renewal cycle patterns and real-time market event influencing factors to build a time series forecasting model and analyze the fluctuation characteristics of customers' potential insurance demand at different time points; Multi-objective optimization ranking module: Constructs a dual-objective optimization function that takes into account both corporate profit goals and customer utility value, sets differentiated strategy parameters according to user role type, and generates a recommended product ranking sequence that satisfies the balance of interests of all parties; Feedback Adaptive Module: By collecting click-through rate and conversion rate data of recommendation results in real time, an online learning mechanism is established to dynamically adjust model parameters and achieve continuous iterative optimization of recommendation strategies.

3. A computer device, characterized in that, The computer program stores a data-based group insurance product recommendation algorithm as described in claim 1, which can be loaded by a processor and executed.

Images