A data tracking, analysis and management platform based on digital life insurance companies

Abstract

The present invention discloses a data tracking and analysis management platform based on digital life insurance companies, including a management platform, and the operation process of the management platform specifically includes the following steps: obtaining multi-source customer data; performing structured processing on the multi-source customer data to generate a customer portrait data set containing a customer tag set; generating an initial risk prediction set based on the customer tag set; real-time monitoring of customer behavior events and external environmental data, and generating dynamic adjustment signals according to preset trigger rules; performing risk reassessment to generate an updated risk prediction set and differentiated insurance strategies; generating a visual analysis report based on the customer portrait data set. The present invention has the following advantages and effects: through the dynamic fusion of multi-source data and the real-time iteration of tag weights, a closed-loop management platform is constructed, which ultimately realizes the dynamic risk assessment and significantly improves the risk response speed and customer service accuracy of life insurance companies.

Description

Technical Field

[0001] The present invention relates to the field of insurance technology, and in particular to a data tracking, analysis and management platform based on digital life insurance companies. Background Art

[0002] With the advent of the digital age, life insurance companies are increasingly relying on technology to enhance their market competitiveness, improve operational efficiency, and enhance customer service levels.

[0003] A core bottleneck facing traditional management platforms in the digital transformation of the life insurance industry lies in their static risk assessment and lack of dynamic responsiveness. Existing technologies typically rely on historical policy data and basic customer attributes. Once set, these weighting coefficients (e.g., weights for tags like health status and behavioral preferences) are permanently fixed, unable to dynamically adjust to real-time customer behavioral events (e.g., abnormal health monitoring data, delayed renewals) or sudden changes in the external environment (e.g., economic fluctuations, healthcare policy adjustments). For example, when a customer's health monitoring device continuously detects abnormal physiological indicators, traditional management platforms are unable to adjust health tag weights in real time, making it difficult to promptly increase risk levels and trigger alerts. Similarly, when faced with changes in customer renewal intentions caused by sudden economic downturns, the lack of a quantitative mechanism to link external environmental data with risk transmission pathways causes risk assessment results to lag behind actual risk evolution. This static nature makes the generation and adjustment of insurance policies heavily reliant on manual experience and periodic batch updates, making it difficult to achieve personalized, real-time, and precise risk control, exacerbating corporate risk exposure and customer churn. Therefore, a technical solution that supports the dynamic fusion of multi-source data and real-time iteration of tag weights is urgently needed to overcome the static limitations of traditional management platforms. Summary of the Invention

[0004] The purpose of the present invention is to provide a data tracking, analysis and management platform based on digital life insurance companies to solve the problems raised in the background technology.

[0005] The above technical objectives of the present invention are achieved through the following technical solutions:

[0006] A data tracking and analysis management platform for digital life insurance companies includes a management platform. The operation process of the management platform specifically includes the following steps:

[0007] S100. Acquire multi-source customer data; wherein the multi-source customer data includes basic customer information, insurance policy records, health monitoring data, consumer behavior data, and claims records;

[0008] S200: Structurally process multi-source customer data to generate a customer profile dataset; wherein the customer profile dataset includes at least one customer tag set, and the customer tag set includes a risk level tag, a behavior preference tag, a health status tag, and corresponding dynamic weight values ​​of each tag;

[0009] S300, generating an initial risk prediction set based on the customer tag set;

[0010] S400: Monitor customer behavior events and external environment data in real time, and generate dynamic adjustment signals based on preset trigger rules;

[0011] S500: Reassess the risk based on the dynamic adjustment signal and the initial risk prediction set to generate an updated risk prediction set; wherein the updated risk prediction set is used to modify the dynamic weight values ​​in the customer profile dataset and generate a differentiated insurance policy;

[0012] S600. Generate a visual analysis report based on the customer portrait dataset.

[0013] By adopting the above-mentioned technical solutions, multi-source heterogeneous data from IoT devices, mobile terminals, and enterprise databases are integrated to cover multi-dimensional information such as customer health, behavior, and economic environment, thereby building a panoramic customer view. Through a mixed real-time and batch collection mode, the synchronous processing of high-frequency health data and low-frequency economic data is taken into account to ensure data timeliness and integrity. Based on real-time monitoring of customer behavior events and the external environment, a risk reassessment mechanism is triggered to solve the problem of strategy lag caused by data update delays in traditional models. Differentiated insurance strategies combine customer risk levels, dynamic weight distribution, and external environmental influences to support personalized premium fluctuations and customized coverage, thereby improving customer satisfaction and corporate risk-return ratio. Through dynamic fusion of multi-source data and real-time iteration of label weights, a closed-loop management platform is built, ultimately realizing dynamic risk assessment and significantly improving the risk response speed and customer service accuracy of life insurance companies.

[0014] It is further configured that the structured processing of multi-source customer data in S200 includes:

[0015] S210: Associate and map the customer's basic information with the insurance policy record using the unique customer identification code to generate a basic customer profile;

[0016] S220: Classify the health monitoring data using a K-means clustering algorithm to generate health status labels, and calculate a first dynamic weight value corresponding to each health status label based on a time series analysis method; wherein the first dynamic weight value is calculated by multiplying the ratio of the mean of the health monitoring data within a preset time window to the standard deviation of the health monitoring data by a preset normalization coefficient to obtain a weight value for the health status label;

[0017] S230: Perform sentiment polarity analysis on the text information in the consumer behavior data using natural language processing technology to generate behavior preference labels, and calculate a second dynamic weight value for each behavior preference label based on a TF-IDF algorithm; wherein the second dynamic weight value is calculated by extracting a weight value of a keyword using the TF-IDF algorithm, and then multiplying the weight value by a preset behavior influence factor to generate a weight value for the behavior preference label;

[0018] S240. Match historical claims records with a preset risk event database, annotate risk level labels, and simulate the probability of occurrence and impact intensity of risk events using a Monte Carlo simulation algorithm to calculate a third dynamic weight value for each risk level label. The third dynamic weight value is calculated by multiplying the simulated probability of occurrence of the event by the impact coefficient, and then multiplying the result by a preset risk adjustment factor to generate a weight value for the risk level label.

[0019] S250. Integrate the health status label and its first dynamic weight value, the behavior preference label and its second dynamic weight value, the risk level label and its third dynamic weight value to generate a customer label set, and aggregate all customer label sets into a customer portrait data set; wherein, the customer portrait data set is stored in a distributed database and supports real-time query and batch update.

[0020] By adopting the above technical solution, the basic information of customers and insurance policy records are associated and mapped through a unique customer identification code. The management platform can achieve accurate management of customer data and avoid erroneous analysis due to information redundancy or data loss. This process generates basic customer files and provides a solid data foundation for subsequent data analysis. In the process of data processing, the K-means clustering algorithm is used to classify health monitoring data and generate health status labels. The dynamic weight value of the health status label is calculated by time series analysis, so that the customer's health status label not only reflects the customer's current health status, but also can be dynamically adjusted with changes in time and customer health data. The calculation process of the first dynamic weight value uses the ratio of the mean and standard deviation of the health monitoring data and the preset normalization coefficient to accurately assess the customer's health risk. At the same time, the sentiment polarity analysis of consumer behavior data and the basic The behavioral preference labels generated by the TF-IDF algorithm can accurately reflect customers' consumption behavior patterns. These labels and their corresponding dynamic weight values ​​not only help insurance companies better understand customer needs, but also provide support for subsequent insurance product design and risk assessment. The combination of natural language processing technology and the TF-IDF algorithm enables the platform to have a deeper understanding of customer behavior and predict their potential claims needs and health risks through their consumption behavior. By matching the risk event library of historical claims records, the platform can analyze customers' historical claims data in real time, assess their risk level, and calculate the probability and impact of risk events through Monte Carlo simulation algorithms to further refine risk predictions. This dynamic and comprehensive collection of customer labels provides insurance companies with an accurate and real-time updated customer portrait, which helps them formulate personalized and differentiated insurance strategies.

[0021] It is further configured that generating the initial risk prediction set in S300 includes:

[0022] S310: Calculate the customer's comprehensive risk score based on the first dynamic weight value, the second dynamic weight value, and the third dynamic weight value in the customer tag set;

[0023] S320, divide the risk level according to the comprehensive risk score and generate an initial risk prediction set; wherein,

[0024] S310 specifically includes the following sub-steps:

[0025] Obtaining a first dynamic weight value of the health status label, a second dynamic weight value of the behavior preference label, and a third dynamic weight value of the risk level label respectively;

[0026] Multiply each dynamic weight value by its corresponding preset risk factor, where the preset risk factors include health risk factor, behavior risk factor and risk event factor;

[0027] The sum of the results of multiplying each dynamic weight value by its corresponding preset risk factor generates the customer's comprehensive risk score;

[0028] S320 specifically includes the following sub-steps:

[0029] Preset risk level threshold table to define the comprehensive risk score intervals corresponding to low risk, medium risk, and high risk;

[0030] Match the customer's comprehensive risk score with the risk level threshold table to determine the customer's risk level;

[0031] A list of events associated with risk levels is extracted from a preset risk event library to generate an initial risk prediction set; wherein, the initial risk prediction set includes the event name, event probability, impact coefficient and priority score; wherein, the priority score is calculated based on the weighted sum of the event probability and the impact coefficient.

[0032] By adopting the above technical solution, the management platform can calculate the customer's comprehensive risk score by combining the dynamic weight values ​​of the health status label, behavioral preference label and risk level label in the customer label set. The comprehensive risk score not only takes into account the customer's health status, consumption behavior and historical claims record, but also integrates the customer's performance in different environments, such as changes in health data, changes in consumption patterns, etc.; this makes the management platform more comprehensive and accurate in risk prediction and can better reflect the customer's actual risk level; in the process of generating the initial risk prediction set, the management platform matches the customer's comprehensive risk score with low risk, medium risk and high risk through a preset risk level threshold table, thereby dividing the customer's risk level. This process can provide a basis for subsequent insurance product pricing, risk assessment and decision making; the generation of the initial risk prediction set depends not only on the customer's health status and behavioral characteristics, but also combines external environmental factors. Through this comprehensive risk prediction, insurance companies can promptly discover potential risks and make reasonable decisions, thereby achieving refined risk management.

[0033] A further setting is to perform conflict detection on the multi-dimensional label set of the same customer in S320. If the logical relationship between the health status label and the risk level label is inconsistent, a manual review process is triggered; wherein the logical relationship contradiction is defined as: the health status label indicates low risk and the risk level label indicates high risk, or the health status label indicates high risk and the risk level label is not adjusted synchronously.

[0034] By adopting the above technical solution and performing conflict detection on the multi-dimensional label sets of the same customer, the management platform can ensure that the logical relationship between the customer's health status label and the risk level label is always consistent. If the health status label indicates low risk and the risk level label indicates high risk, or the health status label indicates high risk and the risk level label is not adjusted synchronously, the management platform will automatically trigger the manual review process. This can effectively reduce the erroneous judgments caused by data contradictions, thereby improving the accuracy of risk prediction. The above conflict detection mechanism not only improves the accuracy of the data, but also ensures the reliability of the system output results through manual review. In the complex risk prediction process, the combination of automation and manual intervention can improve the robustness of the system and avoid errors that may be caused by single automated judgment.

[0035] It is further configured that generating the dynamic adjustment signal in S400 includes:

[0036] S410, setting a trigger rule library;

[0037] The trigger rules in the trigger rule library include:

[0038] If the health monitoring data exceeds the threshold range for N consecutive days, a health risk warning will be triggered;

[0039] If the economic volatility index in the external environment data exceeds the volatility threshold, it will trigger an economic risk reassessment;

[0040] If the customer fails to complete the policy renewal operation within the preset time, the churn risk mark will be triggered;

[0041] S420, monitoring customer behavior events and external environment data streams in real time through an event-driven architecture;

[0042] S430: When a behavior event is detected to match a trigger rule, a dynamic adjustment signal including an event type, a timestamp, and an impact range is generated.

[0043] By adopting the above technical solutions, customer behavioral events and external environment data streams can be monitored in real time, so that dynamic adjustment signals can be quickly generated when customer behavior changes; through the event-driven architecture, the management platform can respond to changes in customer behavior and the external environment in real time, generating dynamic adjustment signals that include event type, timestamp and scope of impact; these dynamic adjustment signals not only provide real-time data for subsequent risk reassessment, but also help insurance companies quickly adjust risk forecasts and strategies when faced with emergencies, thereby improving their response capabilities.

[0044] It is further configured that, in S500, risk reassessment is performed based on the dynamic adjustment signal and the updated risk prediction set, including:

[0045] S510: Receive a dynamic adjustment signal and parse the event type, timestamp, and impact range parameters to extract an abnormal feature vector associated with the customer tag set; wherein the abnormal feature vector includes health monitoring deviation, economic fluctuation sensitivity, and policy renewal hysteresis period;

[0046] S520: Construct a risk transmission model based on a Bayesian network, perform a joint probability distribution calculation on the abnormal feature vector and the probability of occurrence of events in the initial risk prediction set, and generate a dynamic correction coefficient;

[0047] S530, iteratively correcting the first dynamic weight value, the second dynamic weight value, and the third dynamic weight value in the customer portrait dataset based on the dynamic adjustment signal;

[0048] S540: Generate an updated risk prediction set based on the revised first dynamic weight value, the second dynamic weight value, and the third dynamic weight value, and update the customer risk level;

[0049] S550, generating differentiated insurance strategies;

[0050] S560. Write the updated risk prediction set and differentiated insurance strategies into the customer portrait dataset, overwriting the original dynamic weight values ​​and risk prediction results.

[0051] By adopting the above technical solution, the management platform can make timely adjustments when risks change by reassessing risks based on dynamic adjustment signals and updated risk prediction sets. In addition, by iteratively correcting the customer portrait data set based on dynamic adjustment signals, the customer's health status label, behavioral preference label and risk level label can be adjusted according to real-time feedback, thereby continuously optimizing the customer's risk assessment. This iterative correction process helps to continuously improve the customer portrait, ensure that the risk prediction is always consistent with the customer's actual situation, and improve the accuracy and flexibility of the prediction.

[0052] It is further configured that the process of generating the dynamic correction coefficient in S520 includes:

[0053] S520.1. Based on the event type in the dynamic adjustment signal, match a corresponding weight matrix from a preset conduction path weight matrix library. The conduction path weight matrix is ​​a three-dimensional tensor structure with dimensions including event type, abnormal feature vector, and risk event. Matrix element values ​​are conditional probability weights obtained from historical data training.

[0054] S520.2. Input the abnormal feature vector into a pre-trained random forest model, and output the probability of the risk transmission intensity of the abnormal feature to the risk event. The random forest model uses health monitoring deviation, economic fluctuation sensitivity, and policy renewal lag period as input features, and the probability of risk transmission intensity as the output target. The random forest model is trained offline using dynamic weights of historical claims records and customer label sets in the risk event database.

[0055] S520.3. Calculate the time difference between the current timestamp and the reference timestamp of the corresponding event probability in the initial risk prediction set. Substitute the time difference and a preset decay period parameter into the decay function to generate a time decay coefficient. The decay period parameter is dynamically set based on the event type defined in the trigger rule library in S410.

[0056] S520.4. Perform weighted fusion on the matching weight values ​​in the transmission path weight matrix, the risk transmission intensity probability output by the random forest model, and the time attenuation coefficient to generate a dynamic correction coefficient;

[0057] S520.5. Normalize the dynamic correction coefficient.

[0058] By employing this technical solution, we can accurately simulate the relationships between different risk factors and calculate dynamic correction coefficients, thereby incorporating more dynamic factors into the risk prediction process and improving prediction accuracy. By combining event types, abnormal feature vectors, and risk event dimensions, the platform can more accurately assess a client's risk profile in specific scenarios and make appropriate risk adjustments accordingly. This approach, based on a three-dimensional matrix and random forest model, enables efficient and accurate calculations when dealing with complex, multivariate risk predictions, providing more reliable risk assessment results.

[0059] It is further configured that the process of iteratively correcting the first dynamic weight value, the second dynamic weight value, and the third dynamic weight value in the customer portrait data set based on the dynamic adjustment signal in S530 is:

[0060] S530.1. If the dynamic adjustment signal is a health risk warning, perform the following steps:

[0061] Based on the standard deviation ratio of the health monitoring deviation and the mean of the health monitoring data calculated in S220, combined with the dynamic correction coefficient, a weighted linear interpolation method is used to update the first dynamic weight value corresponding to the health status label;

[0062] Triggering the Monte Carlo simulation recalculation of the risk level label, using the updated first dynamic weight value and dynamic correction coefficient as input parameters to regenerate the third dynamic weight value;

[0063] If the difference between the regenerated third dynamic weight value and the original value exceeds a preset threshold, a secondary correction of the cross-label association influence matrix is ​​triggered until the weight value change converges within the threshold;

[0064] S530.2. If the dynamic adjustment signal is an economic risk reassessment, perform the following steps:

[0065] Extract the economic fluctuation sensitivity, calculate the deviation coefficient between the economic fluctuation index and the preset industry benchmark, and add the product of the deviation coefficient and the dynamic correction coefficient to the risk adjustment factor corresponding to the third dynamic weight value of the risk level label to generate the revised third dynamic weight value;

[0066] S530.3. If the dynamic adjustment signal is a churn risk indicator, perform the following steps:

[0067] A negative correlation function is constructed based on the inverse of the policy renewal lag period. The second dynamic weight value of the behavioral preference label and the weighted value of the dynamic correction coefficient are substituted into the function to recalculate the behavioral risk coefficient.

[0068] Next, a revised second dynamic weight value is generated based on the updated behavioral risk coefficient.

[0069] By adopting the above technical solution, the management platform can accurately correct the dynamic weight values ​​in the customer portrait dataset based on different types of dynamic adjustment signals. Each signal type triggers a specific correction algorithm to ensure that the customer's risk assessment always conforms to their latest behavioral and environmental characteristics. For example, in the case of a health risk warning, the dynamic weight value of the health status label is updated through weighted linear interpolation to better reflect the customer's health status. Through this iterative correction process, the management platform can not only respond to changes in customer behavior in a timely manner, but also make targeted adjustments for different types of risks, improving the flexibility and accuracy of insurance policies.

[0070] Further configuration is that, in S540, an updated risk prediction set is generated based on the revised first dynamic weight value, the second dynamic weight value, and the third dynamic weight value, and the customer risk level is updated as follows:

[0071] S540.1. Re-enter the revised first dynamic weight value, second dynamic weight value, and third dynamic weight value into S310 to calculate the comprehensive risk score. Then, weighted correction is performed on the event probability using the dynamic correction coefficient to obtain a revised comprehensive risk score.

[0072] S540.2. Update the customer risk level based on the matching result between the revised comprehensive risk score and the risk level threshold table in S310;

[0073] S540.3. Extract the implicit relationship between the health status label and the risk level label through the association analysis algorithm to generate a cross-label association influence matrix;

[0074] S540.4. Generate an updated risk prediction set based on the updated customer risk level, the cross-tag association impact matrix, and the event list associated with the risk level in the risk event library in S320.

[0075] By adopting the above technical solution, combined with the cross-label association analysis algorithm and the risk event library, it is possible to discover the implicit relationship between health status labels and risk level labels through association analysis, thereby generating a cross-label association impact matrix; the cross-label association impact matrix can help the management platform more accurately consider the mutual influence between different labels when updating customer risk levels, and provide richer data support for subsequent risk prediction and decision-making; through this process, the management platform can provide each customer with a more personalized risk assessment, and at the same time, when the customer's risk changes, timely adjust the risk prediction set and insurance strategy to ensure the scientific and timely nature of the decision.

[0076] Further configuration is that the process of generating differentiated insurance policies in S550 is:

[0077] S550.1. Based on the updated customer risk level, dynamic weight distribution characteristics, and dynamic correction coefficient, match a benchmark insurance plan from the preset policy library;

[0078] S550.2. Optimize the parameters of the benchmark insurance plan through genetic algorithms to generate differentiated insurance strategies;

[0079] S550.3. Perform logical conflict detection on the optimized differentiated insurance strategy:

[0080] If it is detected that the premium floating range and the first dynamic weight value of the health status tag have an opposite trend, and the dynamic correction coefficient is lower than the preset threshold, and there is no external environmental data to support the reverse adjustment, the policy backtracking mechanism is triggered to readjust the preset risk factor weight ratio in S310;

[0081] If no logical conflict is detected, the differentiated insurance policy is confirmed as the final output policy.

[0082] By adopting the above technical solution and optimizing the benchmark insurance plan through genetic algorithms, the platform can generate personalized insurance strategies based on the specific circumstances of each customer. This process ensures that each customer's insurance plan matches their risk characteristics, thereby improving customer satisfaction and enhancing customer trust in insurance products. By performing logical conflict detection on the optimized insurance strategies, the management platform can avoid insurance strategy conflicts caused by improper parameter settings.

[0083] In summary, the present invention has the following beneficial effects:

[0084] Through the dynamic fusion of multi-source data and real-time iteration of label weights, a closed-loop management platform is built, which ultimately realizes the dynamic risk assessment and significantly improves the risk response speed and customer service accuracy of life insurance companies. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 Schematic diagram of the process of an embodiment;

[0086] Figure 2 This is the sub-step of S200 in the embodiment;

[0087] Figure 3 This is a sub-step of S300 in the embodiment;

[0088] Figure 4 This is a sub-step of S400 in the embodiment;

[0089] Figure 5 It is a sub-step of S500 in the embodiment. DETAILED DESCRIPTION

[0090] The present invention will be further described in detail below with reference to the accompanying drawings.

[0091] As attached Figures 1 to 5 As shown;

[0092] This embodiment discloses a data tracking and analysis management platform based on a digital life insurance company, including a management platform. The operation process of the management platform specifically includes the following steps:

[0093] S100. Acquire multi-source customer data; wherein the multi-source customer data includes basic customer information, insurance policy records, health monitoring data, consumer behavior data, and claims records. The multi-source customer data is collected in real time or in batches through IoT devices, mobile terminal applications, and enterprise databases;

[0094] Specifically, the platform utilizes a microservices architecture and is deployed on a cloud server cluster. It comprises a data collection layer, a computing engine layer, a decision-making layer, and an interaction layer. The data collection layer connects to smart wearable devices, mobile apps, government public databases, and enterprise ERP systems via an API gateway. The computing engine layer utilizes the Spark distributed framework for real-time stream processing and batch computing. The decision-making layer incorporates a risk model library and strategy optimization engine. The interaction layer provides a visual dashboard and a two-way push interface, specifically a WeChat mini-program for customers and a PC terminal for sales representatives.

[0095] S200: Structural processing is performed on multi-source customer data to generate a customer profile dataset; wherein the customer profile dataset includes at least one customer tag set, each customer tag set is uniquely associated with a customer, and the customer tag set includes a risk level tag, a behavior preference tag, a health status tag, and corresponding dynamic weight values.

[0096] S300: Generate an initial risk prediction set based on the customer tag set, and push the initial risk prediction set to the business decision terminal;

[0097] S400: Real-time monitoring of customer behavior events and external environmental data, and generation of dynamic adjustment signals based on pre-set triggering rules. Customer behavior events include policy renewals, abnormal changes in health data, and claim submissions, while external environmental data includes economic volatility indexes, medical policy updates, and natural disaster warnings.

[0098] S500: Reassess the risk based on the dynamic adjustment signal and the initial risk prediction set to generate an updated risk prediction set; wherein the updated risk prediction set is used to modify the dynamic weight values ​​in the customer profile dataset and generate a differentiated insurance policy;

[0099] S600. Generate a visual analysis report based on the customer portrait data set, and push differentiated insurance strategies to customer terminals and business decision terminals.

[0100] Specifically, the structured processing of multi-source customer data in S200 includes:

[0101] S210: Associate and map the customer's basic information with the insurance policy record using the unique customer identification code to generate a basic customer profile;

[0102] In this embodiment, a mapping relationship between the customer ID and the policy number is established, and the household registration address and occupation type fields are merged to generate a basic file;

[0103] S220: Classify the health monitoring data using a K-means clustering algorithm to generate health status labels, and calculate a first dynamic weight value corresponding to each health status label based on a time series analysis method; wherein the first dynamic weight value is calculated by multiplying the ratio of the mean of the health monitoring data within a preset time window to the standard deviation of the health monitoring data by a preset normalization coefficient to obtain a weight value for the health status label;

[0104] In this embodiment, K-means clustering can be performed on the customer's blood pressure data for the past three months to generate a "blood pressure fluctuation type" health label; taking the mean μ = 128 and the standard deviation σ = 8 for the 30-day window period, the first dynamic weight value = (μ / σ) × normalization coefficient. The normalization coefficient is set to 0.5, and the calculated first dynamic weight value is 8;

[0105] S230: Perform sentiment polarity analysis on the text information in the consumer behavior data using natural language processing technology to generate behavior preference labels, and calculate a second dynamic weight value for each behavior preference label based on a TF-IDF algorithm; wherein the second dynamic weight value is calculated by extracting a weight value of a keyword using the TF-IDF algorithm, and then multiplying the weight value by a preset behavior influence factor to generate a weight value for the behavior preference label;

[0106] In this example, NLP sentiment analysis was performed on the customer service conversation text in the mobile app titled "I feel my critical illness insurance coverage is insufficient." Specifically, the BERT model was used to identify "insufficient coverage" as a negative keyword. The TF-IDF score was calculated as 0.12, and multiplied by the behavioral factor of 0.8, resulting in a second dynamic weight of 0.096.

[0107] S240. Match historical claims records with a preset risk event database, annotate risk level labels, and simulate the probability of occurrence and impact intensity of risk events using a Monte Carlo simulation algorithm to calculate a third dynamic weight value for each risk level label. The third dynamic weight value is calculated by multiplying the simulated probability of occurrence of the event by the impact coefficient, and then multiplying the result by a preset risk adjustment factor to generate a weight value for the risk level label.

[0108] In this embodiment, it is detected that the customer applied for lung cancer claim in 2021, matching the "malignant tumor" entry in the risk event library; Monte Carlo simulation is performed 100,000 times, and the probability of recurrence within 5 years is 18%, the impact coefficient is 0.7, and the adjustment factor is 1.2. The third dynamic weight value is 0.151.

[0109] S250. Integrate the health status label and its first dynamic weight value, the behavior preference label and its second dynamic weight value, the risk level label and its third dynamic weight value to generate a customer label set, and aggregate all customer label sets into a customer portrait data set; wherein, the customer portrait data set is stored in a distributed database and supports real-time query and batch update.

[0110] Specifically, generating the initial risk prediction set in S300 includes:

[0111] S310: Calculate the customer's comprehensive risk score based on the first dynamic weight value, the second dynamic weight value, and the third dynamic weight value in the customer tag set;

[0112] S320. Classify risk levels according to the comprehensive risk score and generate an initial risk prediction set;

[0113] Specifically, S310 includes the following sub-steps:

[0114] Obtaining a first dynamic weight value of the health status label, a second dynamic weight value of the behavior preference label, and a third dynamic weight value of the risk level label respectively;

[0115] Multiply each dynamic weight value by its corresponding preset risk coefficient, where the preset risk coefficients include the health risk coefficient α, the behavioral risk coefficient β, and the risk event coefficient γ, which take the preset values ​​of 0.1, 5.0, and 3.0 respectively;

[0116] The sum of the results of multiplying each dynamic weight value by its corresponding preset risk factor generates the customer's comprehensive risk score;

[0117] In this embodiment, the customer's comprehensive risk score = first dynamic weight value × 0.1 + second dynamic weight value × 5.0 + third dynamic weight value × 3.0 = 1.877;

[0118] Specifically, S320 includes the following sub-steps:

[0119] Preset risk level threshold table to define the comprehensive risk score intervals corresponding to low risk, medium risk, and high risk;

[0120] The risk level threshold table is shown in Table 1;

[0121] Table 1 Preset risk level thresholds

[0122] Risk Level Comprehensive risk score range Low risk [0, 1.2) Medium risk [1.2, 2.5) High risk [2.5, +∞)

[0123] Match the customer's comprehensive risk score with the risk level threshold table to determine the customer's risk level;

[0124] In this example, the client's comprehensive risk score of 1.877 is considered medium risk;

[0125] A list of events associated with risk levels is extracted from the preset risk event library to generate an initial risk prediction set; the initial risk prediction set includes the event name, event probability, impact coefficient and priority score; the priority score is calculated based on the weighted sum of the event probability and the impact coefficient, with the specific weight ratio being 60% for the probability value and 40% for the impact coefficient.

[0126] Specifically, in S320, conflict detection is performed on the multi-dimensional label set of the same customer. If the logical relationship between the health status label and the risk level label is inconsistent, a manual review process is triggered; wherein, the logical relationship contradiction is defined as: the health status label indicates low risk and the risk level label indicates high risk, or the health status label indicates high risk and the risk level label is not adjusted synchronously.

[0127] In this example, a customer's health label shows "Normal BMI (First Dynamic Weight Value: 2.1)" but their risk label shows "High Risk (Third Dynamic Weight Value: 0.4)." The management platform detects that the first dynamic weight value is below the preset health risk safety threshold (3.0) and the third dynamic weight value is above the preset risk level warning threshold (0.3), triggering a conflict rule. A review work order is generated and sent to the underwriting department. Manual verification reveals that the customer engages in high-altitude work. After manually adding the "Occupational Risk" label, the third dynamic weight value is recalculated to 0.35.

[0128] Specifically, generating the dynamic adjustment signal in S400 includes:

[0129] S410, setting a trigger rule library;

[0130] The trigger rules in the trigger rule library include:

[0131] If the health monitoring data exceeds the threshold range for N consecutive days, a health risk warning is triggered; preferably, N is set to 3;

[0132] If the economic volatility index in the external environment data exceeds the volatility threshold, it triggers an economic risk reassessment; preferably, the volatility threshold is set at 3%;

[0133] If the customer fails to complete the policy renewal operation within the preset time, the churn risk mark will be triggered;

[0134] S420, monitoring customer behavior events and external environment data streams in real time through an event-driven architecture;

[0135] S430: When a behavior event is detected to match a trigger rule, a dynamic adjustment signal including an event type, a timestamp, and an impact range is generated.

[0136] Specifically, in S500, risk reassessment is performed based on the dynamic adjustment signal and the updated risk prediction set, including:

[0137] S510: Receive a dynamic adjustment signal and parse the event type, timestamp, and impact range parameters to extract an abnormal feature vector associated with the customer tag set; wherein the abnormal feature vector includes health monitoring deviation, economic fluctuation sensitivity, and policy renewal hysteresis period;

[0138] S520: Construct a risk transmission model based on a Bayesian network, perform a joint probability distribution calculation on the abnormal feature vector and the probability of occurrence of events in the initial risk prediction set, and generate a dynamic correction coefficient;

[0139] S530, iteratively correcting the first dynamic weight value, the second dynamic weight value, and the third dynamic weight value in the customer portrait dataset based on the dynamic adjustment signal;

[0140] S540: Generate an updated risk prediction set based on the revised first dynamic weight value, the second dynamic weight value, and the third dynamic weight value, and update the customer risk level;

[0141] S550, generating differentiated insurance strategies;

[0142] S560. Write the updated risk prediction set and differentiated insurance strategies into the customer portrait dataset, overwriting the original dynamic weight values ​​and risk prediction results.

[0143] Specifically, the process of generating the dynamic correction coefficient in S520 includes:

[0144] S520.1. Based on the event type in the dynamic adjustment signal, a corresponding weight matrix is ​​matched from a preset conduction path weight matrix library. The conduction path weight matrix is ​​a three-dimensional tensor structure, comprising event type, anomaly feature vector, and risk event dimensions. Matrix element values ​​are conditional probability weights trained using historical data. Matrix element values ​​are calculated by statistically analyzing the conditional probabilities of each event type and anomaly feature vector triggering a risk event in the historical data.

[0145] S520.2. Input the abnormal feature vector into a pre-trained random forest model, and output the probability of the risk transmission intensity of the abnormal feature to the risk event. The random forest model uses health monitoring deviation, economic fluctuation sensitivity, and policy renewal lag period as input features, and the probability of risk transmission intensity as the output target. The random forest model is trained offline using dynamic weights of historical claims records and customer label sets in the risk event database.

[0146] S520.3. Calculate the time difference between the current timestamp and the reference timestamp of the corresponding event probability in the initial risk prediction set. Substitute the time difference and a preset decay period parameter into the decay function to generate a time decay coefficient. The decay period parameter is dynamically set based on the event type defined in the trigger rule library in S410.

[0147] In this embodiment, specifically:

[0148] Health risk warning: Use the logarithmic decay function, the expression is D=log(1+Δt / T1), where D is the time decay coefficient, Δt is the time difference, and T1 is the preset cycle parameter with a value of 7;

[0149] Economic risk reassessment: A linear attenuation function is used, expressed as D=Δt / T2, where D is the time attenuation coefficient, Δt is the time difference, and T2 is the preset period parameter, with a value of 3;

[0150] Churn risk marker: An exponential decay function is used, expressed as D = exp(Δt / T3) − 1, where D is the time decay coefficient, Δt is the time difference, and T3 is the preset period parameter with a value of 15.

[0151] S520.4. Perform weighted fusion on the matching weight values ​​in the transmission path weight matrix, the risk transmission intensity probability output by the random forest model, and the time attenuation coefficient to generate a dynamic correction coefficient;

[0152] In this embodiment, the specific formula is:

[0153] C=W ij ⋅P / (1+D)

[0154] Where C represents the dynamic correction coefficient, W ij It represents the matching weight value in the conduction path weight matrix, P represents the probability of risk conduction intensity, and D represents the time attenuation coefficient.

[0155] S520.5. Normalize the dynamic correction coefficient so that it is mapped to the interval [0, 1].

[0156] Specifically, the process of iteratively correcting the first dynamic weight value, the second dynamic weight value, and the third dynamic weight value in the customer portrait data set based on the dynamic adjustment signal in S530 is as follows:

[0157] S530.1. If the dynamic adjustment signal is a health risk warning, perform the following steps:

[0158] Based on the standard deviation ratio of the health monitoring deviation and the mean of the health monitoring data calculated in S220, combined with the dynamic correction coefficient, a weighted linear interpolation method is used to update the first dynamic weight value corresponding to the health status label;

[0159] Triggering the Monte Carlo simulation recalculation of the risk level label, using the updated first dynamic weight value and dynamic correction coefficient as input parameters to regenerate the third dynamic weight value;

[0160] If the difference between the regenerated third dynamic weight value and the original value exceeds a preset threshold, which is set to 10%, a secondary correction of the cross-label correlation influence matrix is ​​triggered until the weight value change converges within the threshold;

[0161] S530.2. If the dynamic adjustment signal is an economic risk reassessment, perform the following steps:

[0162] Extract the economic fluctuation sensitivity, calculate the deviation coefficient between the economic fluctuation index and the preset industry benchmark, and add the product of the deviation coefficient and the dynamic correction coefficient to the risk adjustment factor corresponding to the third dynamic weight value of the risk level label to generate the revised third dynamic weight value;

[0163] S530.3. If the dynamic adjustment signal is a churn risk indicator, perform the following steps:

[0164] A negative correlation function is constructed based on the inverse of the policy renewal lag period. The second dynamic weight value of the behavioral preference label and the weighted value of the dynamic correction coefficient are substituted into the function to recalculate the behavioral risk coefficient.

[0165] Next, a revised second dynamic weight value is generated based on the updated behavioral risk coefficient.

[0166] Specifically, in S540, an updated risk prediction set is generated based on the revised first dynamic weight value, the second dynamic weight value, and the third dynamic weight value, and the customer risk level is updated as follows:

[0167] S540.1. Re-enter the revised first dynamic weight value, second dynamic weight value, and third dynamic weight value into S310 to calculate the comprehensive risk score. Then, weighted correction is performed on the event probability using the dynamic correction coefficient to obtain a revised comprehensive risk score.

[0168] S540.2. Update the customer risk level based on the matching result between the revised comprehensive risk score and the risk level threshold table in S310;

[0169] S540.3. Extract the implicit relationship between the health status label and the risk level label through the association analysis algorithm to generate a cross-label association influence matrix;

[0170] S540.4. Generate an updated risk prediction set based on the updated customer risk level, the cross-tag association impact matrix, and the event list associated with the risk level in the risk event library in S320.

[0171] Specifically, the process of generating a differentiated insurance policy in S550 is as follows:

[0172] S550.1. Based on the updated customer risk level, dynamic weight distribution characteristics, and dynamic correction coefficient, match a benchmark insurance plan from the preset policy library;

[0173] S550.2. Optimize the parameters of the benchmark insurance plan through a genetic algorithm to generate differentiated insurance strategies; the optimized parameters include the premium floating range, the coverage elasticity threshold, and the claim response priority.

[0174] S550.3. Perform logical conflict detection on the optimized differentiated insurance strategy:

[0175] If it is detected that the premium floating range has an opposite trend to the first dynamic weight value of the health status tag, and the dynamic correction coefficient is lower than the preset threshold, which is set to 0.5, and there is no external environmental data to support the reverse adjustment, the policy backtracking mechanism is triggered to readjust the preset risk factor weight ratio in S310;

[0176] If no logical conflict is detected, the differentiated insurance policy is confirmed as the final output policy.

[0177] In this embodiment, the preset strategy library in S550.1 matches the benchmark insurance plan "whole life insurance + critical illness supplementary insurance" for 50-year-old male customers with medium risk; the parameter optimization in S550.2 corresponds to premium-to-earnings ratio = expected claim amount / (premium * discount rate); after 100 generations of evolution, the output is "term life insurance to age 70 + cancer special insurance", reflecting the "differentiated insurance strategy" generation logic.

[0178] In S550.3, the management platform recommended a 10% premium increase for a "Health Improvement" customer (whose first dynamic weight decreased from 0.4 to 0.2). The platform detected a discrepancy between the changing trend of the first dynamic weight and the direction of the premium adjustment. A retrospective analysis revealed a new "Frequent Beneficiary Change" record in the behavior tag. The revised policy was to maintain the premium but add a beneficiary change restriction clause.

[0179] This specific embodiment is merely an explanation of the present invention and is not intended to limit the present invention. After reading this specification, those skilled in the art may make non-creative modifications to this embodiment as needed. However, as long as such modifications are within the scope of the claims of the present invention, they are protected by patent law.

Claims

1. A data tracking, analysis and management platform based on digital life insurance companies, characterized by: Including a management platform, the operation process of the management platform specifically includes the following steps: S100. Acquire multi-source customer data; wherein the multi-source customer data includes basic customer information, insurance policy records, health monitoring data, consumer behavior data, and claims records; S200: Structural processing is performed on multi-source customer data to generate a customer profile dataset; wherein the customer profile dataset includes at least one customer tag set, and the customer tag set includes a risk level tag, a behavior preference tag, a health status tag, and corresponding dynamic weight values; including: S210: Associate and map the customer's basic information with the insurance policy record using the unique customer identification code to generate a basic customer profile; S220, using a K-means clustering algorithm to classify the health monitoring data, generate health status labels, and calculate a first dynamic weight value corresponding to each health status label based on a time series analysis method; S230, performing sentiment polarity analysis on the text information in the consumer behavior data using natural language processing technology to generate behavior preference labels, and calculating a second dynamic weight value for each behavior preference label based on a TF-IDF algorithm; S240: Match historical claim records with a preset risk event database, label them with risk level tags, simulate the probability of occurrence and impact intensity of risk events using a Monte Carlo simulation algorithm, and calculate a third dynamic weight value for each risk level tag; S250, integrating the health status tag and its first dynamic weight value, the behavior preference tag and its second dynamic weight value, and the risk level tag and its third dynamic weight value to generate a customer tag set, and aggregating all customer tag sets into a customer portrait dataset; wherein the customer portrait dataset is stored in a distributed database, supporting real-time query and batch update; S300, generating an initial risk prediction set based on the customer tag set; S400: Monitor customer behavior events and external environment data in real time, and generate dynamic adjustment signals based on preset trigger rules; S500: Reassess risk based on the dynamic adjustment signal and the initial risk prediction set to generate an updated risk prediction set; wherein the updated risk prediction set is used to modify the dynamic weight values ​​in the customer profile dataset and generate a differentiated insurance policy; including: S510: Receive a dynamic adjustment signal and parse the event type, timestamp, and impact range parameters to extract an abnormal feature vector associated with the customer tag set; wherein the abnormal feature vector includes health monitoring deviation, economic fluctuation sensitivity, and policy renewal hysteresis period; S520: Construct a risk transmission model based on a Bayesian network, perform a joint probability distribution calculation on the abnormal feature vector and the probability of occurrence of events in the initial risk prediction set, and generate a dynamic correction coefficient; S530, iteratively correcting the first dynamic weight value, the second dynamic weight value, and the third dynamic weight value in the customer portrait dataset based on the dynamic adjustment signal; S540: Generate an updated risk prediction set based on the revised first dynamic weight value, the second dynamic weight value, and the third dynamic weight value, and update the customer risk level; S550, generating differentiated insurance strategies; S560: Write the updated risk prediction set and differentiated insurance strategy into the customer profile dataset, overwriting the original dynamic weight value and risk prediction result; S600. Generate a visual analysis report based on the customer portrait dataset.

2. A data tracking, analysis and management platform for digital life insurance companies according to claim 1, characterized by: The calculation process of the first dynamic weight value is: according to the ratio of the mean value of the health monitoring data in the preset time window to the standard deviation of the health monitoring data, multiplied by a preset normalization coefficient, to obtain the weight value of the health status label; The calculation process of the second dynamic weight value is as follows: extracting the weight value of the keyword through the TF-IDF algorithm, and then multiplying it by the preset behavior influence factor to generate the weight value of the behavior preference label; The calculation process of the third dynamic weight value is: multiplying the simulated event probability by the impact coefficient, and then multiplying it by a preset risk adjustment factor to generate a weight value of the risk level label.

3. A data tracking, analysis and management platform for digital life insurance companies according to claim 2, characterized by: Generating an initial risk prediction set in S300 includes: S310: Calculate the customer's comprehensive risk score based on the first dynamic weight value, the second dynamic weight value, and the third dynamic weight value in the customer tag set; S320, divide the risk level according to the comprehensive risk score and generate an initial risk prediction set; wherein, S310 specifically includes the following sub-steps: Obtaining a first dynamic weight value of the health status label, a second dynamic weight value of the behavior preference label, and a third dynamic weight value of the risk level label respectively; Multiply each dynamic weight value by its corresponding preset risk factor, where the preset risk factors include health risk factor, behavior risk factor and risk event factor; The sum of the results of multiplying each dynamic weight value by its corresponding preset risk factor generates the customer's comprehensive risk score; S320 specifically includes the following sub-steps: Preset risk level threshold table to define the comprehensive risk score intervals corresponding to low risk, medium risk, and high risk; Match the customer's comprehensive risk score with the risk level threshold table to determine the customer's risk level; A list of events associated with risk levels is extracted from a preset risk event library to generate an initial risk prediction set; wherein, the initial risk prediction set includes the event name, event probability, impact coefficient and priority score; wherein, the priority score is calculated based on the weighted sum of the event probability and the impact coefficient.

4. A data tracking, analysis and management platform for digital life insurance companies according to claim 3, characterized in that: In S320, conflict detection is performed on the multi-dimensional label set of the same customer. If the logical relationship between the health status label and the risk level label is inconsistent, a manual review process is triggered; wherein, the logical relationship contradiction is defined as: the health status label indicates low risk and the risk level label indicates high risk, or the health status label indicates high risk and the risk level label is not adjusted synchronously.

5. A data tracking, analysis and management platform for digital life insurance companies according to claim 1, characterized in that: Generating a dynamic adjustment signal in S400 includes: S410, setting a trigger rule library; The trigger rules in the trigger rule library include: If the health monitoring data exceeds the threshold range for N consecutive days, a health risk warning will be triggered; If the economic volatility index in the external environment data exceeds the volatility threshold, it will trigger an economic risk reassessment; If the customer fails to complete the policy renewal operation within the preset time, the churn risk mark will be triggered; S420, monitoring customer behavior events and external environment data streams in real time through an event-driven architecture; S430: When a behavior event is detected to match a trigger rule, a dynamic adjustment signal including an event type, a timestamp, and an impact range is generated.

6. A data tracking, analysis and management platform for digital life insurance companies according to claim 1, characterized in that: The process of generating the dynamic correction coefficient in S520 includes: S520.

1. Based on the event type in the dynamic adjustment signal, match a corresponding weight matrix from a preset conduction path weight matrix library. The conduction path weight matrix is ​​a three-dimensional tensor structure with dimensions including event type, abnormal feature vector, and risk event. Matrix element values ​​are conditional probability weights obtained from historical data training. S520.

2. Input the abnormal feature vector into a pre-trained random forest model, and output the probability of the risk transmission intensity of the abnormal feature to the risk event. The random forest model uses health monitoring deviation, economic fluctuation sensitivity, and policy renewal lag period as input features, and the probability of risk transmission intensity as the output target. The random forest model is trained offline using dynamic weights of historical claims records and customer label sets in the risk event database. S520.

3. Calculate the time difference between the current timestamp and the reference timestamp of the corresponding event probability in the initial risk prediction set. Substitute the time difference and a preset decay period parameter into the decay function to generate a time decay coefficient. The decay period parameter is dynamically set based on the event type defined in the trigger rule library in S410. S520.

4. Perform weighted fusion on the matching weight values ​​in the transmission path weight matrix, the risk transmission intensity probability output by the random forest model, and the time attenuation coefficient to generate a dynamic correction coefficient; S520.

5. Normalize the dynamic correction coefficient.

7. A data tracking, analysis and management platform for digital life insurance companies according to claim 1, characterized in that: The process of iteratively correcting the first dynamic weight value, the second dynamic weight value, and the third dynamic weight value in the customer portrait data set based on the dynamic adjustment signal in S530 is as follows: S530.

1. If the dynamic adjustment signal is a health risk warning, perform the following steps: Based on the standard deviation ratio of the health monitoring deviation and the mean of the health monitoring data calculated in S220, combined with the dynamic correction coefficient, a weighted linear interpolation method is used to update the first dynamic weight value corresponding to the health status label; Triggering the Monte Carlo simulation recalculation of the risk level label, using the updated first dynamic weight value and dynamic correction coefficient as input parameters to regenerate the third dynamic weight value; If the difference between the regenerated third dynamic weight value and the original value exceeds a preset threshold, a secondary correction of the cross-label association influence matrix is ​​triggered until the weight value change converges within the threshold; S530.

2. If the dynamic adjustment signal is an economic risk reassessment, perform the following steps: Extract the economic fluctuation sensitivity, calculate the deviation coefficient between the economic fluctuation index and the preset industry benchmark, and add the product of the deviation coefficient and the dynamic correction coefficient to the risk adjustment factor corresponding to the third dynamic weight value of the risk level label to generate the revised third dynamic weight value; S530.

3. If the dynamic adjustment signal is a churn risk indicator, perform the following steps: A negative correlation function is constructed based on the inverse of the policy renewal lag period. The second dynamic weight value of the behavioral preference label and the weighted value of the dynamic correction coefficient are substituted into the function to recalculate the behavioral risk coefficient. Next, a revised second dynamic weight value is generated based on the updated behavioral risk coefficient.

8. A data tracking, analysis and management platform for digital life insurance companies according to claim 1, characterized in that: In S540, an updated risk prediction set is generated based on the revised first dynamic weight value, the second dynamic weight value, and the third dynamic weight value, and the customer risk level is updated as follows: S540.

1. Re-enter the revised first dynamic weight value, second dynamic weight value, and third dynamic weight value into S310 to calculate the comprehensive risk score. Then, weighted correction is performed on the event probability using the dynamic correction coefficient to obtain a revised comprehensive risk score. S540.

2. Update the customer risk level based on the matching result between the revised comprehensive risk score and the risk level threshold table in S310; S540.

3. Extract the implicit relationship between the health status label and the risk level label through the association analysis algorithm to generate a cross-label association influence matrix; S540.

4. Generate an updated risk prediction set based on the updated customer risk level, the cross-tag association impact matrix, and the event list associated with the risk level in the risk event library in S320.

9. A data tracking, analysis and management platform for digital life insurance companies according to claim 1, characterized in that: The process of generating differentiated insurance policies in S550 is as follows: S550.

1. Based on the updated customer risk level, dynamic weight distribution characteristics, and dynamic correction coefficient, match a benchmark insurance plan from the preset policy library; S550.

2. Optimize the parameters of the benchmark insurance plan through genetic algorithms to generate differentiated insurance strategies; S550.

3. Perform logical conflict detection on the optimized differentiated insurance strategy: If it is detected that the premium floating range and the first dynamic weight value of the health status tag have an opposite trend, and the dynamic correction coefficient is lower than the preset threshold, and there is no external environmental data to support the reverse adjustment, the policy backtracking mechanism is triggered to readjust the preset risk factor weight ratio in S310; If no logical conflict is detected, the differentiated insurance policy is confirmed as the final output policy.

Images