Insurance price dynamic matching method and system in combination with user behavior track

Abstract

The invention discloses an insurance price dynamic matching method and system combined with a user behavior track, and relates to the technical field of data processing, and the method comprises the steps: collecting a user association data set, and constructing a user dynamic portrait database; performing rule analysis and parameter extraction on the insurance terms and the rate table to obtain an insurance rate key parameter set, and generating a personalized rate analyzer in combination with user historical data; performing insurance price matching on the user dynamic portrait database based on a personalized rate analyzer, and outputting a basic insurance pricing scheme; and fusing the market change data and the emergency data by adopting a Bayesian network to carry out risk quantitative evaluation, determining a risk probability quantitative parameter, carrying out dynamic strategy regulation and control on the basic insurance pricing scheme based on the risk probability quantitative parameter, and determining a target insurance pricing scheme. The technical problem that in the prior art, insurance pricing lacks a personalized and real-time dynamic adjustment mechanism is solved, and the technical effect of improving insurance pricing accuracy and dynamic matching capacity is achieved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method and system for dynamically matching insurance prices based on user behavior patterns. Background Technology

[0002] Traditional insurance pricing methods primarily rely on static statistical models and pre-set rate tables, typically basing risk assessments and premium calculations on limited static factors such as age, gender, and occupation. This approach struggles to reflect the dynamic risk characteristics of individual users in terms of actual behavior, health changes, and market fluctuations, leading to significant discrepancies in pricing accuracy and fairness. Furthermore, the lack of means to collect and analyze real-time user data makes it difficult for insurance products to respond promptly to changes in user risk status, hindering the implementation of personalized and dynamic insurance pricing strategies. Summary of the Invention

[0003] This application provides a method and system for dynamically matching insurance prices based on user behavior patterns, which is used to address the technical problem that insurance pricing in the prior art lacks a personalized and real-time dynamic adjustment mechanism.

[0004] In view of the above problems, this application provides a method and system for dynamic matching of insurance prices based on user behavior patterns.

[0005] The first aspect of this application provides a method for dynamically matching insurance prices based on user behavior patterns, the method comprising: A user-related dataset is collected, including behavioral trajectory data, historical purchase records, and personal health information. This dataset is aggregated and analyzed using a distributed computing framework to construct a dynamic user profile database. A logical reasoning engine is used to parse rules and extract parameters from insurance terms and rate tables, resulting in a set of key insurance rate parameters. This set of key parameters is then combined with historical user data to generate a personalized rate analyzer. Based on this personalized rate analyzer, insurance prices are matched to the dynamic user profile database to output a basic insurance pricing scheme. Finally, a Bayesian network is used to fuse market change data and event data for risk quantification assessment, determining risk probability quantification parameters. Based on these risk probability quantification parameters, dynamic strategy adjustments are made to the basic insurance pricing scheme to determine a target insurance pricing scheme.

[0006] A second aspect of this application provides a dynamic insurance price matching system that incorporates user behavior patterns, the system comprising: The data acquisition and analysis module collects user-related datasets, including behavioral trajectory data, historical purchase records, and personal health information. It aggregates and analyzes this data using a distributed computing framework to construct a dynamic user profile database. The reasoning module uses a logical reasoning engine to parse rules and extract parameters from insurance terms and rate tables, obtaining a set of key insurance rate parameters. It then combines historical user data with these parameters to generate a personalized rate analyzer. The matching module matches insurance prices to the dynamic user profile database based on the personalized rate analyzer, outputting a basic insurance pricing scheme. The scheme determination module uses a Bayesian network to fuse market change data and event data for risk quantification assessment, determining risk probability quantification parameters. Based on these parameters, it dynamically adjusts the basic insurance pricing scheme to determine a target insurance pricing scheme.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application collects a user-related dataset, including behavioral trajectory data, historical purchase records, and personal health information. It aggregates and analyzes this user-related data using a distributed computing framework to construct a dynamic user profile database. A logical reasoning engine is used to parse rules and extract parameters from insurance terms and rate tables, obtaining a set of key insurance rate parameters. This set of key parameters is then combined with historical user data to generate a personalized rate analyzer. Based on this personalized rate analyzer, insurance prices are matched to the dynamic user profile database to output a basic insurance pricing scheme. A Bayesian network is used to fuse market change data and event data for risk quantification assessment, determining risk probability quantification parameters. Based on these risk probability quantification parameters, the basic insurance pricing scheme is dynamically adjusted to determine a target insurance pricing scheme. This invention addresses the technical problem of the lack of personalized and real-time dynamic adjustment mechanisms in existing insurance pricing. By collecting user behavioral trajectories, historical records, and health information to construct dynamic profiles, and combining logical reasoning and Bayesian networks to achieve personalized rate analysis and risk quantification assessment, it improves the accuracy and dynamic matching capabilities of insurance pricing. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1A schematic diagram of the method for dynamically matching insurance prices based on user behavior patterns provided in this application embodiment; Figure 2 A schematic diagram of the structure of an insurance price dynamic matching system that combines user behavior trajectories, provided in an embodiment of this application.

[0010] Figure labeling: Data acquisition and analysis module 11, reasoning module 12, matching module 13, scheme determination module 14. Detailed Implementation

[0011] This application provides a method and system for dynamic matching of insurance prices by combining user behavior trajectories. It addresses the technical problem of the lack of personalized and real-time dynamic adjustment mechanisms in existing insurance pricing. By collecting user behavior trajectories, historical records, and health information to construct dynamic profiles, and combining logical reasoning and Bayesian networks to achieve personalized rate analysis and risk quantification assessment, the technical effect of improving the accuracy of insurance pricing and dynamic matching capabilities is achieved.

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0013] It should be noted that any variation of the terms "comprising" and "having" is intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0014] Example 1, as Figure 1 As shown, this application provides a method for dynamically matching insurance prices based on user behavior patterns, the method comprising: Step S100: Collect user-related datasets, which include behavioral trajectory data, historical purchase records, and personal health information. Aggregate and analyze the user-related data through a distributed computing framework to construct a dynamic user profile database.

[0015] In this embodiment, multi-source data of the target user is first retrieved from a pre-established historical database to form a user-related dataset. This dataset includes behavioral trajectory data, historical purchase records, and personal health information. Different types of insurance correspond to different behavioral trajectory data. For example, for auto insurance, the behavioral trajectory consists of time-series data collected via IoT devices (such as wearable devices and vehicle sensors), including user location, activity frequency, and driving behavior. Historical purchase records include the user's past auto insurance purchase records (such as policy information, coverage items, sum insured, and insurance application date) and claims history (such as claim reporting time, compensation amount, accident type, and liability allocation). Personal health information includes the user's electronic health record, recording data such as past medical examination reports.

[0016] Subsequently, user-related data is aggregated and analyzed using a distributed computing framework. This process begins by constructing a three-layer data structure based on the deployed distributed computing framework: a raw storage data layer, a real-time query data layer, and a multi-source analysis data layer. The collected user-related datasets (including behavioral trajectory data, historical purchase records, and personal health information) are then stored in the raw storage data layer, and a unified timestamp is added to the data for time-series identification, thereby generating a user-related sequence dataset with time-series attributes. A fixed time segmentation window strategy is set, and user data within the corresponding time period is retrieved through the real-time query data layer to generate a time-window user dataset with sliding characteristics. Finally, the sliding time-window user dataset is progressively aggregated and its features extracted based on the multi-source analysis data layer, fusing multi-dimensional user information to ultimately construct a dynamic user profile database.

[0017] Furthermore, in the method provided in the application embodiments, the construction of the user dynamic profile database further includes: Based on the distributed computing framework, a raw storage data layer, a real-time query data layer, and a multi-source analysis data layer are obtained. The user-related dataset is stored in the raw storage data layer for time-series identification to obtain a user-related sequence dataset. A preset time segmentation window is used, and the real-time query data layer queries and calls the user-related sequence dataset according to the time segmentation window to obtain a sliding time window user dataset. Based on the multi-source analysis data layer, the sliding time window user dataset is aggregated and analyzed sequentially to construct a user dynamic profile database.

[0018] In this embodiment, a three-layer data architecture is first constructed based on a distributed computing framework (such as Apache Hadoop): a raw storage data layer, a real-time query data layer, and a multi-source analysis data layer. The raw storage data layer carries unprocessed raw user data, possessing high fault tolerance and large-scale parallel processing capabilities, and forms the underlying foundation of the entire data processing flow. The real-time query data layer supports fast data retrieval based on time windows, meeting the real-time requirements of subsequent dynamic analysis. The multi-source analysis data layer provides multi-dimensional data feature fusion and computation capabilities.

[0019] The collected user-related dataset (including behavioral trajectory data, historical purchase records, and personal health information) is then written into the original storage data layer, and each data record is uniformly identified by time sequence. That is, by adding timestamps and unique user identifiers, the data is processed into time sequence, thereby constructing a user-related sequence dataset with time sequence relationships.

[0020] To extract user behavior characteristics at different time periods, a preset time segmentation window is set, such as a daily, weekly, or monthly periodic unit. Data in the user-related sequence dataset is called through the real-time query data layer, and user activity information within the target time period is obtained by time segmentation, forming a continuous sliding time window user dataset.

[0021] Finally, based on the multi-source analysis data layer, the sliding time window user dataset is aggregated and analyzed sequentially. In this process, firstly, based on the actual needs of insurance price analysis, profile feature dimensions for characterizing user traits are designed and determined. Then, feature extraction is performed item by item on the sliding time window user dataset around these profile feature dimensions, generating a user profile feature set containing multi-dimensional attributes. Based on this, the profile feature dimensions are classified and analyzed, corresponding profile label generation rules are formulated, and these rules are applied to perform labeling and aggregation analysis on the extracted user profile feature set, ultimately completing the construction of the dynamic user profile database.

[0022] Furthermore, in the method provided in the application embodiments, the step of sequentially aggregating and analyzing the sliding time window user dataset based on the multi-source analysis data layer to construct a user dynamic profile database further includes: The multi-source analysis data layer constructs profile feature dimensions based on insurance price analysis requirements; it extracts features from the sliding time window user dataset sequentially according to the profile feature dimensions to obtain a multi-dimensional user profile feature set; it performs label classification and parsing on the profile feature dimensions based on the insurance price analysis requirements to determine profile label generation rules; and it performs label aggregation analysis on the multi-dimensional user profile feature set based on the profile label generation rules to construct the user dynamic profile database.

[0023] In this embodiment, the first step, based on the needs of insurance price analysis, is to filter out feature factors related to insurance prices from historical user data. This step uses Pearson correlation coefficient analysis to calculate the linear relationship between user attribute variables and historical premium data. By calculating the correlation coefficient between each candidate variable (such as driving behavior indicators, transaction records, claims data, etc.) and the insurance price, variables with absolute correlation coefficient values ​​higher than a set threshold (such as 0.6) are retained as the profile feature dimensions required to construct the profile. Examples include the number of rapid accelerations, the proportion of nighttime driving, and historical claims rates.

[0024] Next, features are extracted from the user dataset within the sliding time window according to the profile feature dimensions. In this process, a moving average method is used to process the time-series data. The user's daily behavioral data within the sliding time window (e.g., 7 days) is summed and averaged to obtain feature values ​​with time smoothness. For example, the moving average method is applied to the daily number of emergency braking events to obtain the 7-day average number of emergency braking events; the 7-day average nighttime driving ratio is calculated based on nighttime driving duration; and the 7-day number of claims is calculated based on claim records, thus constructing a multi-dimensional user profile feature set reflecting the user's periodic behavioral status.

[0025] Subsequently, based on the needs of insurance price analysis, the profile feature dimensions were categorized and analyzed using labels. To ensure the standardization and feasibility of label generation, an equidistant interval division method was adopted. Each continuous feature was divided into several fixed-width intervals according to its numerical range, and a semantically clear label name was assigned to each interval. For example, the frequency of rapid acceleration was divided into three equidistant intervals: 0-3, 4-6, and more than 7 times, labeled as low-risk driving, medium-risk driving, and high-risk driving, respectively. The proportion of nighttime driving was divided into 0%-30%, 31%-60%, and 61%-100%, labeled as less nighttime driving, moderate nighttime driving, and frequent nighttime driving, respectively. Finally, a profile label generation rule based on equidistant numerical division was formed.

[0026] Finally, based on the user profile tag generation rules, a tag-based aggregation analysis is performed on the multi-dimensional user profile feature set. A conditional matching method is used to iterate through each feature value in the multi-dimensional user profile feature set, identifying its corresponding tag according to the interval range set in the tag generation rules. For example, User A's 7-day average frequency of rapid acceleration is 5 times, matching the tag "medium-risk driving"; the proportion of night driving is 72%, matching the tag "frequent night driving"; and the number of claims in 7 days is 2, matching "medium-frequency claims user". All tags are aggregated with the user's unique identifier to form a complete tag set for the user within the current sliding time window, and this tag set is stored in a structured manner. This completes the construction of the user dynamic profile database.

[0027] Step S200: Use the logic reasoning engine to perform rule parsing and parameter extraction on the insurance terms and rate table to obtain a set of key parameters for insurance rates. Combine the user's historical data with the set of key parameters for insurance rates to generate a personalized rate analyzer.

[0028] In this embodiment, a logic reasoning engine is first used to parse rules and extract parameters from the insurance terms and rate tables. Specifically, the insurance terms are segmented into chapters to form a set of key insurance chapter content. Based on a unified field naming rule, fields are extracted and standardized from this content set and the rate table to generate a set of insurance terms field sets and a set of insurance rate field sets. Subsequently, the logic reasoning engine performs conditional logic parsing on these field sets to construct an insurance price logic library expressing the pricing relationship of insurance products. Rule parameter extraction and consistency checks are then performed on this logic library to extract structured parameters related to pricing, forming a set of key parameters for insurance rates.

[0029] After obtaining the key parameter set for insurance rates, it is combined with user historical data. First, the original user data undergoes user historical data standardization to ensure consistent data structure and computability. Then, the key parameter set for insurance rates is divided into static and dynamic insurance rate parameters according to attributes, and calculation logic analysis is performed based on this, constructing a rate parameter-calculation logic mapping table. Finally, available user historical data is combined with this mapping table to generate a personalized rate analyzer tailored to individual user characteristics.

[0030] Furthermore, in the method provided in the application embodiments, obtaining the key parameter set of insurance premium rates further includes: The insurance terms are segmented and extracted to obtain a set of key insurance chapters. Field naming rules are constructed, and based on these rules, fields are extracted and their names standardized for the key insurance chapters and the rate table to obtain a set of insurance terms field sets and a set of insurance rate field sets. A logic reasoning engine is used to parse the insurance terms field sets and insurance rate field sets using conditional logic to construct an insurance price logic library. Rule parameters are extracted and consistency checks are performed on the insurance price logic library to obtain a set of key insurance rate parameters.

[0031] In this embodiment, the insurance clause text is first processed in a structured manner. Using a text segmentation and recognition method, combined with paragraph numbering, keyword distribution, and heading hierarchy, the original insurance clause is automatically split, extracting parts with clear semantic boundaries, such as insurance liability clauses, insurance period clauses, exclusion clauses, and applicable rate clauses, thereby forming a set of key insurance chapters.

[0032] After obtaining the key insurance chapter content set, a unified field naming rule was established to standardize the expression of the same concepts in different clauses. This rule, based on common insurance industry terminology and rate model requirements, explicitly maps various expressions appearing in natural language descriptions (such as "date of birth," "insured's age," and "age group") to the standard field name "age." Subsequently, according to the field naming rule, field identification was performed on the text in the key insurance chapter content set and the structured information in the insurance rate table. Fields in the text were extracted using keyword extraction and semantic rule matching, while fields in the rate table were extracted using header parsing and row / column positioning. This ultimately yielded the insurance clause field set and the insurance rate field set, respectively.

[0033] Subsequently, a logic reasoning engine is used to perform conditional logic parsing on the insurance clause field set and the insurance premium rate field set. During this process, the logic reasoning engine uses rule recognition methods to perform pattern matching and logical relationship extraction on logical statements such as "if...then..." structures and "applies when...is satisfied..." appearing in the insurance clause field set and the insurance premium rate field set. For example, from the description "the insured's age is over 60 years old and an additional senior citizen fee is required," the condition variable "age," the threshold "60," and the result "additional fee" are identified and converted into a computable conditional structure. By analyzing all fields involved in the calculation rules, a complete logical system covering pricing conditions, scope of application, parameter values, and the impact of results is established, forming an insurance pricing logic library containing rule statements, condition judgments, and parameter calls.

[0034] Finally, after the insurance pricing logic library is built, all variables involved in the insurance pricing calculation are extracted, and rule parameters are extracted and consistency checks are performed. During parameter extraction, numerical variables, categorical variables, and their value ranges appearing in all logical structures are identified, and these parameters are uniformly verified to ensure consistency across different rules. For example, whether the same fields use the same units, whether boundary values ​​overlap, and whether expressions are repetitive. Through deduplication, conflict checking, and boundary handling, the structural and semantic consistency of all parameters is ensured. Ultimately, a set of parameters processed through logical analysis and standard verification is obtained, namely the key parameter set for insurance rates.

[0035] Furthermore, in the method provided in the application embodiments, the generation of the personalized rate analyzer further includes: The user historical data is standardized to obtain available user historical data; the key parameter set of insurance rates is classified to obtain static insurance rate parameters and dynamic insurance rate parameters; calculation logic analysis is performed based on the static insurance rate parameters and dynamic insurance rate parameters to construct a rate parameter-calculation logic mapping table; and a personalized rate analyzer is generated by combining the available user historical data and the rate parameter-calculation logic mapping table.

[0036] In this embodiment, user historical data is first standardized. By employing data standardization methods, including field alignment, format conversion, and unit unification, data with inconsistent formats, naming conventions, and units is cleaned and mapped to a standard. Specifically, a field mapping table is used to unify the naming of synonymous fields such as "date of birth" and "insured's age"; different time formats such as "YYYY / MM / DD" and "MM-DD-YYYY" are converted to a unified format using time format conversion methods; and units such as "kilometer," "kilometer," or "km" are normalized. The result is usable user historical data with a consistent structure, unified fields, and standardized data format.

[0037] The key parameters for insurance rates were then categorized using a parameter source classification method. Based on the definition and acquisition path of the parameters, they were divided into static and dynamic insurance rate parameters. Static insurance rate parameters refer to those directly stipulated in the insurance terms and conditions and obtainable directly by reading fixed values ​​or rule tables, such as age-based rate segments, gender-based surcharges, and occupational category surcharges. Dynamic insurance rate parameters rely on historical user data for statistical calculations and reflect the dynamic changes in user behavioral risk. For example, the "No-Claim Discount (NCD)" is obtained through a backtracking method of annual claims records. This involves performing time-series analysis on the user's policy claims records over the past few years, calculating the number of consecutive years without claims, and then mapping the NCD to an industry standard table (e.g., 95% to 60% discount for 1-5 years) to derive the NCD. Risk adjustment factors are obtained by integrating multiple behavioral indicators, such as "nighttime driving ratio," "frequency of rapid acceleration," "number of accidents in the past three years," and "annual claims rate." For these parameters, the original indicators are first normalized, then combined with preset weighting factors, and finally the user's comprehensive risk score is obtained by weighted summation. By comparing with preset risk level thresholds, the score is mapped to a discrete risk adjustment coefficient.

[0038] Subsequently, a calculation logic analysis was performed based on static and dynamic insurance rate parameters to construct a rate parameter-calculation logic mapping table. In this process, for each static insurance rate parameter, a corresponding static rate mapping path was established, such as mapping the "age" field to the "age-segmented rate table." For each dynamic insurance rate parameter, a corresponding behavioral statistical path was established, such as the "No-Claims Discount Coefficient (NCD)" which is obtained by identifying the timestamps of claims records and calculating the number of consecutive years without claims, and then matching the coefficient with the "NCD Level Discount Table." The "Risk Adjustment Coefficient" is calculated by combining multiple variables to determine the risk score, and then the adjustment coefficient is extracted by referring to the risk pricing table. All mapping paths and calculation rules are ultimately unified into a rate parameter-calculation logic mapping table.

[0039] Finally, a personalized rate analyzer is generated by combining available user historical data and a rate parameter-calculation logic mapping table. This process begins by invoking and executing rules based on the rate parameter-calculation logic mapping table to map user behavior characteristics to rate factors, generating user rate sample data. Subsequently, a deep neural network structure is used to divide the user rate sample data into training and testing sets. Through layer-by-layer training of the network model, the nonlinear mapping relationship between user characteristics and insurance prices is learned, thus constructing an initial rate analyzer with basic predictive capabilities. Based on this, the initial model undergoes systematic performance verification and hyperparameter tuning, including adjusting parameters such as learning rate, number of layers, and number of nodes. Optimization is achieved through evaluation metrics such as loss function, accuracy, and generalization ability, ultimately generating a personalized rate analyzer that can be used for dynamic insurance price prediction and personalized matching.

[0040] Furthermore, in the method provided in the application embodiments, the step of generating a personalized rate analyzer by combining the available user historical data and the rate parameter-calculation logic mapping table further includes: Based on the rate parameter-calculation logic mapping table, the available user historical data is calculated and identified to obtain user rate sample data; the user rate sample data is proportionally divided and analyzed using a deep neural network structure to construct an initial rate analyzer; the initial rate analyzer is then subjected to performance verification and hyperparameter tuning to generate the personalized rate analyzer.

[0041] In this embodiment, the available user historical data is first calculated and identified based on the rate parameter-calculation logic mapping table. This process uses a field matching method to extract fields such as age, occupation category, number of historical claims, and nighttime driving ratio from the available user historical data according to the field names and calculation paths defined in the mapping table, and then sequentially calls the corresponding rate rules to perform calculations to generate user rate sample data.

[0042] Subsequently, a deep neural network structure is used to model and train the user premium rate sample data. This step first employs a fixed-ratio partitioning method, dividing the user premium rate sample data into training and testing sets at, for example, an 80%:20% ratio, to ensure the independence of model training and evaluation. Next, a multilayer perceptron (MLP) structure is used as the network architecture, with multiple input features from the user premium rate sample data (such as age segmentation coefficients, NCD coefficients, risk adjustment coefficients, etc.) as input layer nodes, configured with several hidden layers and activation functions such as ReLU, and finally outputting the predicted premium as the output layer node. During training, the error backpropagation algorithm combined with the mean squared error (MSE) loss function is used to optimize the weight parameters through gradient descent, gradually approximating the true premium value, thereby constructing an initial premium rate analyzer with preliminary predictive capabilities.

[0043] The initial rate analyzer is then validated and its hyperparameters fine-tuned. In the re-validation phase, cross-validation methods (such as K-fold cross-validation) are used to repeatedly evaluate the model's performance on both the training and test sets. Key metrics such as mean squared error (MSE) and mean absolute error (MAE) are analyzed to assess the model's fitting ability and generalization performance. Based on this, a grid search method is used to systematically adjust the model's hyperparameters, including but not limited to learning rate, batch size, number of network layers, number of neurons per layer, and activation function type. Through continuous iterative experimentation and performance comparison, the optimal hyperparameter combination is selected, and the model structure and parameters are adjusted and solidified, ultimately generating a personalized rate analyzer that can stably output accurate prediction results.

[0044] Step S300: Based on the personalized rate analyzer, perform insurance price matching on the user dynamic profile database and output a basic insurance pricing scheme.

[0045] In this embodiment, the multi-dimensional user profile feature set in the user dynamic profile database is first invoked and calculated based on the personalized rate analyzer. Using a feature field mapping method, key tag fields related to pricing in the profile are extracted, such as age group tags, occupational category tags, risk adjustment coefficients, and no-claims bonus coefficients (NCD), and these are input factors into the personalized rate analyzer for premium prediction calculation.

[0046] During the prediction process, the personalized rate analyzer, based on a pre-trained deep neural network model and combined with an existing rate parameter-calculation logic mapping table, performs nonlinear mapping and combination calculations on the aforementioned input fields, outputting the predicted premium value for the user. In the calculation process, both static insurance rate parameters (such as age-segmented rates and gender surcharges) and dynamic insurance rate parameters (such as risk adjustment coefficients and NCD level discounts) are used as input factors in the prediction logic.

[0047] Finally, the above prediction results are output as a preliminary quote based on the user's current risk status, thus forming a basic insurance pricing scheme.

[0048] Step S400: Use a Bayesian network to fuse market change data and emergency event data to conduct a risk quantification assessment, determine the risk probability quantification parameters, and dynamically adjust the basic insurance pricing scheme based on the risk probability quantification parameters to determine the target insurance pricing scheme.

[0049] In this embodiment, market change data and emergency event data are first collected, and a Bayesian network is used to perform risk cascading analysis on this data to construct a hierarchical risk node network representing the causal relationship of risks. Subsequently, by combining historical insurance claims data and expert knowledge, conditional probability tables are constructed and risk propagation training is performed on this network to form an insurance risk probability propagation network for risk reasoning. Based on this insurance risk probability propagation network, and after inputting current market change data and emergency event data, a quantitative assessment of external environmental risks is completed, and quantitative risk probability parameters for pricing regulation are output.

[0050] After obtaining the risk probability quantification parameters, the preset insurance pricing control strategy library is called to perform strategy matching analysis on the risk probability quantification parameters, trigger the corresponding target pricing control strategy, and dynamically adjust the basic insurance pricing scheme based on the strategy, and finally generate a target insurance pricing scheme that conforms to the current risk status.

[0051] Furthermore, in the method provided in the application embodiments, determining the risk probability quantification parameter further includes: A Bayesian network is used to perform risk cascade analysis on the market change data and emergency event data to generate a hierarchical risk node network; historical insurance claims data is obtained, and based on expert knowledge and the historical insurance claims data, a conditional probability table is constructed and risk propagation training is performed on the hierarchical risk node network to obtain an insurance risk probability propagation network; based on the insurance risk probability propagation network, the risk of the market change data and emergency event data is quantitatively assessed to determine the risk probability quantification parameters.

[0052] In this embodiment, a Bayesian network is first used to perform a risk cascade analysis on market change data and emergency event data, generating a hierarchical risk node network for risk modeling. Market change data includes macroeconomic indicators (such as GDP growth rate and interest rate changes) and industry rate changes; emergency event data includes high-impact factors such as natural disasters, public health events, major accidents, and sudden policy changes. Specifically, firstly, insurance pricing influencing factors are extracted from the market change data and emergency event data to identify variables that directly affect the risk structure, constructing root nodes for insurance pricing as the first layer of input nodes. Secondly, dependency analysis is performed on these root nodes to identify their associated derived variables, forming a set of intermediate nodes for insurance pricing. Then, the output variables of the intermediate nodes are used as the set of leaf nodes for insurance pricing at the end of the network. Finally, a Bayesian network structure learning method is used to perform structural cascade modeling on the root nodes, intermediate nodes, and leaf nodes, generating a hierarchical risk node network representing multi-level causal relationships.

[0053] Subsequently, historical insurance claims data was acquired, including information such as payout amount, accident type, liability allocation, accident occurrence time, region, and insured type. First, data cleaning and structuring methods were used to standardize fields and handle missing information in the raw claims data. Then, based on this claims data and an existing hierarchical risk node network, a conditional probability table (CPT) for each risk node was constructed using the maximum likelihood estimation method. The joint distribution characteristics from historical data were then used to fit the conditional dependencies between nodes.

[0054] During the construction of the conditional probability table, expert knowledge-driven rules provided by domain experts are incorporated to set prior conditions for certain marginal events or sparse data intervals to compensate for insufficient data or improve estimation robustness. For example, the rule "If the region is a typhoon-prone area, the probability of flooding accidents increases significantly" is set to calibrate the relationship structure between risk nodes. Based on the above, a risk propagation training method is adopted, utilizing the structural dependencies and CPT parameters of each node in the Bayesian network to train the entire hierarchical risk node network. Through an iterative learning process, the conditional probabilities between each node are adjusted to establish an insurance risk probability propagation network.

[0055] Finally, based on the constructed insurance risk probability propagation network, risk inference is performed on real-time acquired market change data and emergency event data. Using observed data as input nodes, the state probability of target risk nodes (such as "surge in auto insurance payout rates" or "increased critical illness claim rates") is calculated using posterior probability inference methods. To improve inference efficiency, for nodes that are difficult to solve directly in the posterior calculation, variational inference methods are used to approximate the probability distribution, obtaining their probability value distribution under current conditions. By summarizing, normalizing, and standardizing the posterior probabilities of all target risk nodes, the final output is a quantitative parameter of risk probability used for adjusting insurance pricing strategies.

[0056] Furthermore, in the method provided in the application embodiments, the generation of the hierarchical risk node network further includes: Factors influencing insurance pricing are extracted from the market change data and emergency event data to construct the root node of insurance pricing influence. Dependency variables are associated based on the root node to obtain the intermediate node set of insurance pricing. The resulting variables of the intermediate node set are used as the leaf node set of insurance pricing. A Bayesian network is used to perform a structural cascade analysis on the root node, intermediate node set, and leaf node set of insurance pricing influence to generate the hierarchical risk node network.

[0057] In this embodiment, market change data and emergency event data are first processed to extract factors influencing insurance pricing, thereby constructing the root node of insurance pricing influence. Specifically, market change data includes macroeconomic indicators (such as GDP growth rate and interest rates), industry operating data (such as the average loss ratio of the auto insurance market and the floating range of rates published by regulators), and policy change signals (such as updates to regulatory policies and adjustments to tax incentives). Emergency event data includes natural disaster data (such as the event level and affected areas collected through disaster early warning systems for earthquakes, typhoons, and floods), public safety event data (such as early warnings of frequent traffic accidents and mass incidents), and information on public health emergencies (such as early warnings of infectious disease outbreaks). Principal component analysis (PCA) is used to reduce the dimensionality and classify the above variables, extracting the dominant change factors. For example, the analysis can extract principal component variables such as "macroeconomic uncertainty index," "natural disaster frequency factor," and "industry pricing fluctuation trend" as the root node of insurance pricing influence.

[0058] Next, the relationship between the aforementioned root nodes affecting insurance pricing and other relevant variables is analyzed to identify intermediate variables influenced by these input variables. Specifically, the Pearson correlation coefficient method is used to calculate the correlation between each pair of variables, statistically assessing their linear association. For example, the analysis found a significant positive correlation between "regulatory policy adjustments" and "changes in policy subscription rates" and "auto insurance business growth rate," therefore these two variables are included in the insurance pricing intermediate node set. Through this process, the insurance pricing intermediate node set is obtained.

[0059] Subsequently, variables directly affecting the insurance price output from the intermediate node set are filtered and designated as insurance pricing leaf nodes. This process is conducted by technical experts based on experience and known pricing logic. Specifically, by reviewing historical pricing calculation rules and considering the meaning of each intermediate node's output, it is determined whether it will be directly used for pricing decisions. For example, if an intermediate node, "increased claims costs," directly leads to a price increase, then that node is added to the insurance pricing leaf node set. Through this process, the insurance pricing leaf node set is obtained.

[0060] Finally, a Bayesian network structure learning method was employed to model the root node, intermediate node set, and leaf node set of insurance pricing. By analyzing the causal dependencies between these nodes, a complete hierarchical structure was constructed. During the modeling process, a structure scoring algorithm (such as BIC scoring) combined with a greedy search strategy was used to automatically learn the possible directed connections between nodes, identifying the parent-child relationships of each node, thus forming a layer-by-layer propagation path from the root node to the leaf node. Ultimately, a hierarchical risk node network composed of multiple variable nodes was obtained.

[0061] Furthermore, in the method provided in the application embodiments, the step of determining the target insurance pricing scheme further includes: An insurance pricing control strategy library is constructed. Based on the insurance pricing control strategy library, strategy matching analysis is performed on the risk probability quantification parameters to trigger a target pricing control strategy. Based on the target pricing control strategy, the basic insurance pricing scheme is dynamically controlled to determine the target insurance pricing scheme.

[0062] In this embodiment, an insurance pricing control strategy library is first constructed. This library consists of a set of operable control rules set by technical experts based on historical claims patterns, user behavior characteristics, and experience with market risk events. Each rule includes two parts: strategy triggering conditions and pricing adjustment actions. For example, a strategy set for user behavior is "when a user makes more than two claims in the past 12 months, cancel all premium discounts"; another strategy set for market risk events is "when the probability of a major natural disaster in a certain region exceeds 85%, add a temporary risk rate of 10% to vehicles insured in that region".

[0063] Subsequently, based on the risk probability quantification parameters obtained in the previous steps—that is, the risk probability values ​​obtained after quantifying and evaluating market change data and emergency event data through Bayesian networks (e.g., the probability of a typhoon impact in a certain region is 90%, and the probability of an abnormal payout rate for a certain customer group is 80%)—strategy matching analysis is performed. Specifically, a condition matching algorithm is used to compare the current risk probability quantification parameters with the trigger conditions of each strategy in the insurance pricing control strategy library. For example, if a strategy sets the trigger condition to "the probability of an abnormal payout rate is greater than 75%", and the actual parameter is 80%, then the strategy is considered to have met the trigger condition, and it is identified and activated as a target pricing control strategy.

[0064] After identifying the target pricing adjustment strategy, the existing basic insurance pricing scheme is dynamically adjusted based on this strategy. For example, if the basic insurance pricing scheme includes an original discount factor NCD of 0.8, and the target pricing adjustment strategy requires "setting NCD to 1," the NCD parameter in the pricing scheme will be replaced, and the final premium will be recalculated. The entire adjustment process includes steps such as parameter modification, pricing factor recalculation, and rule execution, ensuring that changes in risk are promptly mapped into the actual pricing results. Finally, the adjusted pricing scheme, i.e., the target insurance pricing scheme, is output.

[0065] In summary, the embodiments of this application have at least the following technical effects: This application collects a user-related dataset, including behavioral trajectory data, historical purchase records, and personal health information. It aggregates and analyzes this user-related data using a distributed computing framework to construct a dynamic user profile database. A logical reasoning engine is used to parse rules and extract parameters from insurance terms and rate tables, obtaining a set of key insurance rate parameters. This set of key parameters is then combined with historical user data to generate a personalized rate analyzer. Based on this personalized rate analyzer, insurance prices are matched to the dynamic user profile database to output a basic insurance pricing scheme. A Bayesian network is used to fuse market change data and event data for risk quantification assessment, determining risk probability quantification parameters. Based on these risk probability quantification parameters, the basic insurance pricing scheme is dynamically adjusted to determine a target insurance pricing scheme. This invention addresses the technical problem of the lack of personalized and real-time dynamic adjustment mechanisms in existing insurance pricing. By collecting user behavioral trajectories, historical records, and health information to construct dynamic profiles, and combining logical reasoning and Bayesian networks to achieve personalized rate analysis and risk quantification assessment, it improves the accuracy and dynamic matching capabilities of insurance pricing.

[0066] Example 2, based on the same inventive concept as the method for dynamically matching insurance prices by incorporating user behavior trajectories in the foregoing examples, such as... Figure 2 As shown, this application provides a dynamic insurance price matching system that combines user behavior trajectories. The system and method embodiments in this application are based on the same inventive concept. The system includes: The data acquisition and analysis module 11 is used to collect user-related datasets, which include behavioral trajectory data, historical purchase records, and personal health information. It aggregates and analyzes these user-related data using a distributed computing framework to construct a dynamic user profile database. The reasoning module 12 uses a logical reasoning engine to parse rules and extract parameters from insurance terms and rate tables, obtaining a set of key insurance rate parameters. It then combines historical user data with this set of key insurance rate parameters to generate a personalized rate analyzer. The matching module 13 uses the personalized rate analyzer to match insurance prices to the dynamic user profile database, outputting a basic insurance pricing scheme. The scheme determination module 14 uses a Bayesian network to fuse market change data and event data for risk quantification assessment, determining risk probability quantification parameters. Based on these risk probability quantification parameters, it dynamically adjusts the basic insurance pricing scheme to determine a target insurance pricing scheme.

[0067] Furthermore, the system is also used to implement the following functions: Based on the distributed computing framework, a raw storage data layer, a real-time query data layer, and a multi-source analysis data layer are obtained. The user-related dataset is stored in the raw storage data layer for time-series identification to obtain a user-related sequence dataset. A preset time segmentation window is used, and the real-time query data layer queries and calls the user-related sequence dataset according to the time segmentation window to obtain a sliding time window user dataset. Based on the multi-source analysis data layer, the sliding time window user dataset is aggregated and analyzed sequentially to construct a user dynamic profile database.

[0068] Furthermore, the system is also used to implement the following functions: The multi-source analysis data layer constructs profile feature dimensions based on insurance price analysis requirements; it extracts features from the sliding time window user dataset sequentially according to the profile feature dimensions to obtain a multi-dimensional user profile feature set; it performs label classification and parsing on the profile feature dimensions based on the insurance price analysis requirements to determine profile label generation rules; and it performs label aggregation analysis on the multi-dimensional user profile feature set based on the profile label generation rules to construct the user dynamic profile database.

[0069] Furthermore, the system is also used to implement the following functions: The insurance terms are segmented and extracted to obtain a set of key insurance chapters. Field naming rules are constructed, and based on these rules, fields are extracted and their names standardized for the key insurance chapters and the rate table to obtain a set of insurance terms field sets and a set of insurance rate field sets. A logic reasoning engine is used to parse the insurance terms field sets and insurance rate field sets using conditional logic to construct an insurance price logic library. Rule parameters are extracted and consistency checks are performed on the insurance price logic library to obtain a set of key insurance rate parameters.

[0070] Furthermore, the system is also used to implement the following functions: The user historical data is standardized to obtain available user historical data; the key parameter set of insurance rates is classified to obtain static insurance rate parameters and dynamic insurance rate parameters; calculation logic analysis is performed based on the static insurance rate parameters and dynamic insurance rate parameters to construct a rate parameter-calculation logic mapping table; and a personalized rate analyzer is generated by combining the available user historical data and the rate parameter-calculation logic mapping table.

[0071] Furthermore, the system is also used to implement the following functions: Based on the rate parameter-calculation logic mapping table, the available user historical data is calculated and identified to obtain user rate sample data; the user rate sample data is proportionally divided and analyzed using a deep neural network structure to construct an initial rate analyzer; the initial rate analyzer is then subjected to performance verification and hyperparameter tuning to generate the personalized rate analyzer.

[0072] Furthermore, the system is also used to implement the following functions: A Bayesian network is used to perform risk cascade analysis on the market change data and emergency event data to generate a hierarchical risk node network; historical insurance claims data is obtained, and based on expert knowledge and the historical insurance claims data, a conditional probability table is constructed and risk propagation training is performed on the hierarchical risk node network to obtain an insurance risk probability propagation network; based on the insurance risk probability propagation network, the risk of the market change data and emergency event data is quantitatively assessed to determine the risk probability quantification parameters.

[0073] Furthermore, the system is also used to implement the following functions: Factors influencing insurance pricing are extracted from the market change data and emergency event data to construct the root node of insurance pricing influence. Dependency variables are associated based on the root node to obtain the intermediate node set of insurance pricing. The resulting variables of the intermediate node set are used as the leaf node set of insurance pricing. A Bayesian network is used to perform a structural cascade analysis on the root node, intermediate node set, and leaf node set of insurance pricing influence to generate the hierarchical risk node network.

[0074] Furthermore, the system is also used to implement the following functions: An insurance pricing control strategy library is constructed. Based on the insurance pricing control strategy library, strategy matching analysis is performed on the risk probability quantification parameters to trigger a target pricing control strategy. Based on the target pricing control strategy, the basic insurance pricing scheme is dynamically controlled to determine the target insurance pricing scheme.

[0075] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0076] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0077] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for dynamically matching insurance prices based on user behavior patterns, characterized in that: The method includes: Collect user-related datasets, which include behavioral trajectory data, historical purchase records, and personal health information. Aggregate and analyze the user-related data through a distributed computing framework to construct a dynamic user profile database. The logic reasoning engine is used to parse the insurance terms and rate tables by rules and extract parameters to obtain a set of key parameters for insurance rates. Combined with the user's historical data and the set of key parameters for insurance rates, a personalized rate analyzer is generated. Based on the personalized rate analyzer, insurance prices are matched to the user dynamic profile database to output a basic insurance pricing scheme. A Bayesian network is used to fuse market change data and emergency event data to conduct a quantitative risk assessment, determine the quantitative parameters of risk probability, and dynamically adjust the basic insurance pricing scheme based on the quantitative parameters of risk probability to determine the target insurance pricing scheme.

2. The method for dynamically matching insurance prices based on user behavior trajectories as described in claim 1, characterized in that, The construction of the user dynamic profile database includes: Based on the distributed computing framework, the raw storage data layer, the real-time query data layer, and the multi-source analysis data layer are obtained; The user-associated dataset is stored in the original storage data layer for time-series identification to obtain the user-associated sequence dataset. A preset time segmentation window is used to query and call the user-related sequence dataset according to the time segmentation window through the real-time query data layer to obtain the sliding time window user dataset; Based on the multi-source analysis data layer, the user dataset of the sliding time window is aggregated and analyzed sequentially to construct a user dynamic profile database.

3. The method for dynamically matching insurance prices based on user behavior trajectories as described in claim 2, characterized in that, The process of sequentially aggregating and analyzing the sliding time window user dataset based on the multi-source analysis data layer to construct a user dynamic profile database includes: The multi-source analysis data layer constructs profile feature dimensions based on insurance price analysis needs; The user dataset in the sliding time window is subjected to feature extraction according to the portrait feature dimensions in sequence to obtain a multi-dimensional user portrait feature set. Based on the insurance price analysis requirements, the profile feature dimensions are classified and parsed to determine the profile tag generation rules. Based on the aforementioned profile tag generation rules, the multidimensional user profile feature set is subjected to tag-based aggregation analysis to construct the user dynamic profile database.

4. The method for dynamically matching insurance prices based on user behavior trajectories as described in claim 1, characterized in that, The obtained set of key parameters for insurance premium rates includes: The insurance terms are segmented and extracted to obtain a set of key insurance chapters. Construct field naming rules, and based on the field naming rules, extract and standardize the naming of fields in the key insurance chapter content set and the rate table to obtain the insurance clause field set and the insurance rate field set; The insurance clause field set and insurance premium rate field set are parsed using a logic reasoning engine to construct an insurance price logic library. The insurance price logic library is subjected to rule parameter extraction and consistency checks to obtain a set of key parameters for insurance rates.

5. The method for dynamically matching insurance prices based on user behavior trajectories as described in claim 1, characterized in that, The personalized rate analyzer includes: The user historical data is standardized to obtain usable user historical data; The key parameters of the insurance premium rate are classified to obtain static insurance premium rate parameters and dynamic insurance premium rate parameters; Based on the static and dynamic insurance premium rate parameters, a calculation logic analysis is performed to construct a rate parameter-calculation logic mapping table. By combining the available user historical data and the rate parameter-calculation logic mapping table, a personalized rate analyzer is generated.

6. The method for dynamically matching insurance prices based on user behavior trajectories as described in claim 5, characterized in that, The process of generating a personalized rate analyzer by combining the available user historical data and the rate parameter-calculation logic mapping table includes: Based on the rate parameter-calculation logic mapping table, the available user historical data is calculated and identified to obtain user rate sample data; A deep neural network structure is used to proportionally divide and analyze the user rate sample data for training, and an initial rate analyzer is constructed. The initial rate analyzer is then subjected to performance verification and hyperparameter tuning to generate the personalized rate analyzer.

7. The method for dynamically matching insurance prices based on user behavior trajectories as described in claim 1, characterized in that, The parameters for determining the probability quantification of risk include: A Bayesian network is used to perform risk cascade analysis on the market change data and emergency event data to generate a hierarchical risk node network. Historical insurance claims data is obtained, and based on expert knowledge and the historical insurance claims data, a conditional probability table is constructed and risk propagation training is performed on the hierarchical risk node network to obtain an insurance risk probability propagation network. Based on the insurance risk probability propagation network, the market change data and emergency event data are used to perform risk quantification assessment to determine the risk probability quantification parameters.

8. The method for dynamically matching insurance prices based on user behavior trajectories as described in claim 7, characterized in that, The generation of the hierarchical risk node network includes: The market change data and emergency event data are used to extract factors affecting insurance pricing, and the root node affecting insurance pricing is constructed. Based on the root node that influences insurance pricing, dependency variables are associated to obtain the set of intermediate nodes for insurance pricing; The result variables of the intermediate node set of insurance pricing are used as the leaf node set of insurance pricing; A Bayesian network is used to perform structural cascade analysis on the root node affecting insurance pricing, the set of intermediate nodes for insurance pricing, and the set of leaf nodes for insurance pricing, thereby generating the hierarchical risk node network.

9. The method for dynamically matching insurance prices based on user behavior trajectories as described in claim 1, characterized in that, The determination of the target insurance pricing scheme includes: Construct an insurance pricing control strategy library, and perform strategy matching analysis on the risk probability quantification parameters based on the insurance pricing control strategy library to trigger the target pricing control strategy; Based on the target pricing control strategy, the basic insurance pricing scheme is dynamically adjusted to determine the target insurance pricing scheme.

10. An insurance price dynamic matching system that combines user behavior patterns, characterized in that, The system is used to execute the dynamic insurance price matching method combining user behavior trajectories as described in any one of claims 1-9, the system comprising: The data collection and analysis module is used to collect user-related datasets, which include behavioral trajectory data, historical purchase records, and personal health information. The user-related data is aggregated and analyzed through a distributed computing framework to construct a dynamic user profile database. The reasoning module is used to perform rule parsing and parameter extraction on insurance terms and rate tables using a logic reasoning engine to obtain a set of key parameters for insurance rates. Combined with user historical data and the set of key parameters for insurance rates, a personalized rate analyzer is generated. The matching module is used to match insurance prices to the user dynamic profile database based on the personalized rate analyzer and output a basic insurance pricing scheme. The scheme determination module is used to perform risk quantification assessment by fusing market change data and emergency event data using Bayesian networks, determine risk probability quantification parameters, and dynamically adjust the basic insurance pricing scheme based on the risk probability quantification parameters to determine the target insurance pricing scheme.

Images