Insurance strategy generation method and device based on risk tensor field, equipment and medium

Abstract

The invention relates to the technical field of data analysis, and particularly discloses an insurance strategy generation method and device based on a risk tensor field, equipment and a medium. According to the method, when insurance combination strategy optimization is triggered, the multi-dimensional risk time sequence data of the user is obtained, the current risk tensor field is constructed according to the multi-dimensional risk time sequence data, the multi-dimensional overall risk portrait is obtained, the limitation of isolation of data of all dimensions is broken through, user risk changes are captured in real time, and the user experience is improved. The timeliness of the risk tensor field is improved, gradient optimization is performed on the insurance combination according to the current risk tensor field, and the timeliness and accuracy of the insurance combination are improved. The method is applied to financial insurance product planning or recommendation services, user risk information can be comprehensively analyzed, insurance combinations meeting user risk guarantee requirements can be quickly generated, and user experience is improved.

Description

Method, apparatus, equipment, and medium for generating insurance strategies based on risk tensor fields Technical Field

[0001] This application relates to the field of data analysis technology, and in particular to a method, apparatus, device and medium for generating insurance strategies based on risk tensor fields. Background Technology

[0002] With the acceleration of digital transformation, intelligent recommendation technology has become a core support for insurance product planning and sales. Most existing intelligent insurance recommendation technologies use rule engines, combining static rules such as manually preset age ranges, premium budgets, and medical history to match and recommend insurance products. However, they cannot cope with dynamic changes in user risk characteristics, resulting in low accuracy and timeliness of recommendations. Traditional machine learning models such as logistic regression and gradient boosting decision trees are also widely used in insurance recommendations, but once the models are trained, the recommendation strategies are fixed and cannot respond in real time to dynamic adjustments in user risk dimensions. Therefore, improving the timeliness and accuracy of insurance strategy recommendations has become an urgent problem to be solved. Summary of the Invention

[0003] This application provides a method, apparatus, device, and medium for generating insurance strategies based on risk tensor fields, in order to improve the timeliness and accuracy of insurance strategy recommendations.

[0004] In a first aspect, this application provides an insurance strategy generation method based on a risk tensor field. The method includes: when triggering insurance portfolio strategy optimization, acquiring multi-dimensional risk time-series data corresponding to the target user; constructing a current risk tensor field based on the multi-dimensional risk time-series data; analyzing historical insurance portfolios and the current risk tensor field to determine the risk gap loss value of the historical insurance portfolios; and performing gradient optimization on the historical insurance portfolios based on the risk gap loss value to obtain a target insurance portfolio strategy.

[0005] Secondly, this application also provides an insurance strategy generation device based on a risk tensor field. The device includes: a risk time series data acquisition module, used to acquire multi-dimensional risk time series data corresponding to a target user when insurance portfolio strategy optimization is triggered; a risk tensor field acquisition module, used to construct a current risk tensor field based on the multi-dimensional risk time series data; and an insurance portfolio strategy acquisition module, used to perform gradient optimization on the historical insurance portfolio based on the risk gap loss value to obtain a target insurance portfolio strategy.

[0006] Thirdly, this application also provides a computer device, the computer device including a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and, when executing the computer program, implement the insurance strategy generation method based on the risk tensor field as described above.

[0007] Fourthly, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to implement the above-described method for generating insurance strategies based on risk tensor fields.

[0008] This application discloses a method, apparatus, device, and medium for generating insurance strategies based on a risk tensor field. When insurance portfolio strategy optimization is triggered, multi-dimensional risk time-series data corresponding to the target user is acquired; a current risk tensor field is constructed based on the multi-dimensional risk time-series data; historical insurance portfolios and the current risk tensor field are analyzed to determine the risk gap loss value of the historical insurance portfolios; and gradient optimization is performed on the historical insurance portfolios based on the risk gap loss value to obtain the target insurance portfolio strategy. This application acquires the user's multi-dimensional risk time-series data when insurance portfolio strategy optimization is triggered, constructs a current risk tensor field based on this data, obtains a multi-dimensional overall risk profile, breaks the limitations of isolated data across dimensions, and achieves real-time capture of user risk changes, improving the timeliness of the risk tensor field. Gradient optimization of the insurance portfolio based on the current risk tensor field further improves the timeliness and accuracy of the insurance portfolio. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 is a first schematic flowchart of an insurance strategy generation method based on a risk tensor field provided by an embodiment of this application; Figure 2 is a second schematic flowchart of an insurance strategy generation method based on a risk tensor field provided by an embodiment of this application; Figure 3 is a third schematic flowchart of an insurance strategy generation method based on a risk tensor field provided by an embodiment of this application; Figure 4 is a schematic block diagram of an insurance strategy generation device based on a risk tensor field provided by an embodiment of this application; Figure 5 is a schematic block diagram of the structure of a computer device provided by an embodiment of this application. Detailed Implementation

[0011] 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 some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0012] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0013] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0014] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0015] This application provides an insurance strategy generation method, apparatus, device, and medium based on a risk tensor field. The risk tensor field-based insurance strategy generation method can be applied to a server. When insurance portfolio strategy optimization is triggered, it acquires multi-dimensional risk time-series data of the user, constructs a current risk tensor field based on this data, and obtains a multi-dimensional overall risk profile. This overcomes the limitations of isolated data across dimensions and enables real-time capture of user risk changes, improving the timeliness of the risk tensor field. Gradient optimization of the insurance portfolio based on the current risk tensor field further enhances the timeliness and accuracy of the insurance portfolio. The server can be a standalone server or a server cluster.

[0016] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0017] Please refer to Figure 1, which is a schematic flowchart of an insurance strategy generation method based on a risk tensor field provided by an embodiment of this application. This risk tensor field-based insurance strategy generation method can be applied to a server to acquire multi-dimensional risk time-series data of users when insurance portfolio strategy optimization is triggered. Based on this multi-dimensional risk time-series data, a current risk tensor field is constructed to obtain a multi-dimensional overall risk profile, breaking the limitations of isolated data across dimensions. Furthermore, it achieves real-time capture of user risk changes, improving the timeliness of the risk tensor field. Based on the current risk tensor field, gradient optimization of the insurance portfolio is performed, improving the timeliness and accuracy of the insurance portfolio.

[0018] As shown in Figure 1, the insurance strategy generation method based on risk tensor field specifically includes steps S101 to S104.

[0019] S101. When the insurance portfolio strategy optimization is triggered, multi-dimensional risk time series data corresponding to the target user is obtained; in one embodiment, when the triggering conditions of the insurance portfolio strategy optimization process are met (such as changes in user risk data, external event disturbances, or risk gap thresholds), the insurance portfolio strategy optimization is triggered.

[0020] In one embodiment, multi-dimensional risk time-series data includes, but is not limited to: health risk time-series data: sleep quality, exercise frequency, changes in physical examination indicators, etc. of target users; vehicle risk time-series data: driving behavior (number of times of emergency braking, frequency of speeding), accident records, vehicle age, etc.; family responsibility risk time-series data: changes in family structure (birth of children, support for the elderly), mortgage / car loan liabilities, etc.; asset risk time-series data: income fluctuations, changes in investment portfolios, increases or decreases in asset size, etc.; external event time-series data: outbreaks of epidemics, extreme weather, policy adjustments, etc.

[0021] Collect continuous, timestamped data (such as time-series data updated daily / weekly) from the aforementioned dimensions to ensure data is linked to the time dimension. Specifically, this can be uploaded by users; alternatively, with user authorization, real-time monitoring of time-series records from corresponding devices, systems, or platforms across various dimensions, such as user health management apps, smart wearables, connected vehicle devices, asset management systems, and weather platforms, can be conducted.

[0022] Furthermore, before obtaining the multi-dimensional risk time-series data corresponding to the target user when triggering insurance portfolio strategy optimization, the method further includes: monitoring the multi-dimensional risk data of the target user; when an update to the multi-dimensional risk data is detected, determining the risk gap loss value corresponding to the current multi-dimensional risk data and the historical insurance portfolio based on a preset risk gap loss function; and determining to trigger insurance portfolio optimization when the risk gap loss value is greater than a preset threshold.

[0023] In one embodiment, risk data for each dimension can be uploaded by the user; alternatively, with the user's authorization, real-time monitoring can be performed on the time-series data recorded by the corresponding devices, systems, or platforms for each dimension, such as the user's health management app, smart wearable devices, connected vehicle devices, asset management systems, and weather platforms.

[0024] For example, health risk time-series data can be time-series data such as sleep quality, exercise frequency, and physiological indicators (blood pressure / blood sugar) collected through health management apps and smart wearable devices (wristbands / watches); vehicle risk time-series data can be dynamic data such as driving behavior (sudden braking / speeding), accident records, and vehicle age obtained through vehicle networking systems or car insurance platforms; family liability risk time-series data can be obtained by tracking updates to family information filled in by users (such as the birth of children and support for the elderly) and changes in mortgage / car loan liabilities in financial systems; asset risk time-series data can be obtained by connecting to users' financial data or asset management systems to monitor income fluctuations, investment portfolio adjustments, and increases or decreases in asset size; external event risk time-series data can be obtained in real time from third-party data sources (meteorological platforms, policy announcements, epidemic prevention and control systems) to obtain disturbance information such as extreme weather, policy adjustments, and public health events.

[0025] In one embodiment, the monitoring frequency for each dimension of data can be different. For example, a real-time collection strategy can be used for high-frequency data such as health indicators and driving behavior, while a periodic polling strategy (such as daily / weekly) can be used for low-frequency data such as family structure and asset size. The monitoring frequency can be set according to actual needs.

[0026] In one embodiment, the monitoring results are analyzed to determine if an update has occurred. For numerical data, the change is assessed to determine if it exceeds a preset threshold (e.g., blood pressure ±10%). For categorical data, status changes are detected (e.g., family structure changes from "married" to "with children"). For external events, new events are monitored (e.g., regional outbreaks).

[0027] In one embodiment, the risk gap loss function measures the degree of underinsurance or overinsurance in the historical insurance portfolio under the current risk tensor field. A larger loss value indicates a larger protection gap or more severe resource waste. Specifically, based on the weighted calculation of the mapping relationship between each risk vector in the risk field and the current strategy, for each risk dimension, it assesses whether the sum insured / coverage of the corresponding insurance type matches the risk vector of each risk dimension (e.g., if health risk increases, is the critical illness insurance sum insured sufficient). The total loss value is obtained by weighted summation, reflecting the overall protection gap. The weight coefficients of each dimension can be set by the user according to actual needs. In one embodiment, the preset threshold can be set by the user according to business needs (protection gap tolerance, compliance requirements) or customer preferences.

[0028] In one embodiment, when a data update is detected, the current risk gap loss value is calculated and compared with a threshold. If the loss value is less than or equal to the threshold, the combination is well matched and no optimization is needed; if the loss value is greater than the threshold, there is a significant gap and optimization needs to be triggered.

[0029] S102. Construct the current risk tensor field based on the multi-dimensional risk time series data; in one embodiment, the current risk tensor field refers to the current risk field state of the target user, which is a multi-dimensional tensor space model constructed based on the multi-dimensional risk time series data.

[0030] Specifically, the latest multi-dimensional risk time series data is input into the risk field mapping function, and the function fuses the vectors of each dimension to obtain the element values ​​in the tensor space, thereby generating the current risk tensor field.

[0031] S103. Analyze the historical insurance portfolio and the current risk tensor field to determine the risk gap loss value of the historical insurance portfolio; in one embodiment, the historical insurance portfolio refers to the target user's current insurance strategy, including parameters such as the sum insured of each type of insurance, the configuration ratio, and the premium budget.

[0032] In one embodiment, the risk gap loss value is calculated using the current risk tensor field and historical insurance portfolios. The risk gap loss value measures the degree to which the historical insurance portfolio is under-insured or over-insured under the current risk tensor field.

[0033] Specifically, risk vectors for each dimension of the user's current risk are extracted from the current risk tensor field. The strategy parameters of historical insurance portfolios are converted into quantitative indicators that match the dimensions of the current risk tensor field. For example, critical illness insurance covering 100 diseases is converted into a disease coverage score, corresponding to the disease risk in the current risk tensor field. Based on a preset risk gap loss function, the risk vectors for each dimension and the quantitative indicators of historical insurance portfolios are calculated to obtain the risk gap loss value for each risk dimension. Finally, the risk gap loss values ​​for each risk dimension are weighted and calculated to obtain the risk gap loss value of the historical insurance portfolio under the current risk tensor field.

[0034] S104. Based on the risk gap loss value, perform gradient optimization on the historical insurance portfolio to obtain the target insurance portfolio strategy.

[0035] In one embodiment, the strategy parameters are adjusted using a gradient descent algorithm to obtain an optimized target insurance portfolio strategy, such as adjusting the sum insured, adding new insurance products, or optimizing cross-insurance product configurations.

[0036] Furthermore, the step of performing gradient optimization on the historical insurance portfolio based on the risk gap loss value to obtain the target insurance portfolio strategy includes: determining the adjustment direction parameter and magnitude parameter of the historical insurance portfolio based on the gradient descent algorithm and the risk gap loss value; adjusting the historical insurance portfolio based on the direction parameter and magnitude parameter until the risk gap loss value is less than a preset threshold, thereby obtaining the target insurance portfolio strategy.

[0037] In one embodiment, a strategy self-evolution mechanism is employed to automatically optimize the insurance portfolio based on the gradient of the risk field. The optimization formula is as follows:

[0038] in, For historical insurance portfolios, For the risk gap loss function, The learning rate is used to control the step size and avoid oscillations or slow convergence.

[0039] In one embodiment, based on the current risk field Historical insurance portfolio (Including parameters such as sum assured, insurance type configuration ratio, and premium allocation), preset learning rate η, to calculate historical insurance portfolio strategies. Corresponding risk gap loss value Calculate the loss function for gradient of each parameter Update the policy parameters in the opposite direction of the gradient, control the step size with the learning rate η, and generate a new policy. .

[0040] In one embodiment, the risk gap loss function Its gradient is used to measure the degree to which historical insurance portfolios under-cover the current risk tensor field. The calculation logic is as follows: Gradient definition: ,in, , i=1,2,...,n, are the parameters of the insurance portfolio strategy (e.g. For life insurance coverage, (Medical insurance percentage); if a certain parameter When the coverage gap decreases, the loss L decreases (i.e., the coverage gap shrinks). Parameters need to be increased (e.g., if the critical illness insurance coverage is insufficient, increasing the coverage can reduce losses); if a certain parameter When the coverage increases, the loss L increases (i.e., overinsurance or underinsurance worsens), then The parameters need to be reduced (e.g., if life insurance coverage is excessive, reducing the coverage amount can reduce losses); if No parameter adjustments are required.

[0041] In one embodiment, the adjustment magnitude parameter is determined by both the learning rate η and the absolute value of the gradient, i.e., magnitude = The larger the absolute value of the gradient, the greater the adjustment range (reflecting the urgency of the risk gap).

[0042] In one embodiment, gradient descent-based parameter tuning is a iterative optimization process. Specifically, using historical insurance portfolios... Starting from the gradient, calculate the gradient. L determines the direction and extent of the adjustment. Based on the direction and extent of the adjustment, updates the sum assured (e.g., increasing the sum assured for critical illness insurance), adds / removes insurance products (e.g., adding children's education insurance), and optimizes the allocation ratio (e.g., increasing the proportion of medical insurance) to generate a new strategy. The loss value of the new strategy is recalculated and executed cyclically until the termination condition is met. At this point, the insurance portfolio is used as the target insurance portfolio strategy. .

[0043] In one embodiment, the termination condition can be that the current risk gap loss value is less than a preset loss threshold; it can also be that the number of iterations reaches the upper limit to avoid infinite loops; or it can be that the absolute value of the gradient is less than the minimum value, at which point the parameters are close to the optimal solution and the adjustment benefit is negligible.

[0044] In one embodiment, when the iteration terminates, the current insurance portfolio strategy is the target insurance portfolio strategy.

[0045] In the above embodiments, when insurance portfolio strategy optimization is triggered, multi-dimensional risk time-series data of the user is acquired. Based on this data, a current risk tensor field is constructed to obtain a multi-dimensional overall risk profile, breaking the limitations of isolated data across dimensions. Furthermore, real-time capture of user risk changes is achieved, improving the timeliness of the risk tensor field. Gradient optimization of the insurance portfolio is then performed based on the current risk tensor field, enhancing both the timeliness and accuracy of the insurance portfolio. Secondly, by differentiating the risk gap loss function using the gradient descent algorithm, the adjustment direction of historical insurance portfolios is quantitatively determined. This addresses the shortcomings of existing technologies that rely on manual rules or static models, leading to inaccurate adjustments and improving the accuracy of strategy adjustments.

[0046] Furthermore, after performing gradient optimization on the historical insurance portfolio based on the risk gap loss value to obtain the target insurance portfolio strategy, the method further includes: analyzing the multi-dimensional risk time series data to determine risk factors; determining the risk propagation path corresponding to the risk factors based on the current risk tensor field; and generating a strategy optimization logic description based on the risk factors, the risk propagation path, and the target insurance portfolio strategy for user review.

[0047] In one embodiment, for each risk dimension, key features strongly correlated with the insurance strategy are extracted from time-series data. This strong correlation can be determined by learning from historical insurance optimization events. For example, if blood pressure leads to a significant number of events requiring insurance portfolio optimization, it indicates a strong correlation between blood pressure and the insurance strategy.

[0048] In one embodiment, a risk factor refers to a risk variable that has a significant impact on insurance strategy optimization. The time-series rate of change of features (e.g., a 12% week-on-week decrease in sleep duration) is calculated using a sliding window to identify long-term trends (e.g., a decrease in exercise frequency over three consecutive months). Abrupt points (e.g., a sudden increase in blood pressure exceeding a preset threshold, or discrete changes in family structure from "married" to "having children") are captured based on pre-set thresholds, and the risk characteristics of these captured abrupt changes are used as risk factors. In another embodiment, the gradient contribution to the risk gap loss function can also be greater than or equal to a preset gradient contribution threshold (e.g., 10%).

[0049] In one embodiment, the risk propagation path is the path by which a risk factor influences other risk dimensions or insurance types within the current risk tensor field. Specifically, the risk dimension to which the risk factor belongs (e.g., decreased sleep in the health dimension) is determined. Then, other dimensions or insurance types with a correlation strength greater than or equal to a preset correlation strength threshold are found through the core tensor in the current risk tensor field (e.g., the correlation strength between the health risk dimension and life insurance is greater than or equal to a preset correlation strength threshold). This generates a path from the dimension to which the risk factor belongs to, to the associated dimension, and then to the insurance type, or a path from the dimension to which the risk factor belongs to, such as decreased sleep leading to increased health risk leading to increased life insurance coverage demand.

[0050] In one embodiment, the strategy optimization logic description includes a triggering factor description, which clarifies the risk factors that trigger optimization and details of their changes; a risk propagation path, which demonstrates the propagation chain of factors in the risk field; and a strategy optimization causal chain, which explains the specific optimization actions and logic.

[0051] For example, the triggering factor is explained as follows: Your sleep duration decreases by 12% for 7 consecutive days, triggering a health risk factor. The risk propagation path is explained as follows: The health risk factor propagates through the risk field to the life insurance dimension, leading to a widening of the life insurance coverage gap. The causal link for strategy optimization is explained as follows: Therefore, your life insurance coverage needs to be increased from 500,000 to 650,000 to cover the coverage gap caused by the increased health risk.

[0052] In one embodiment, the strategy optimization logic can be presented in natural language, or in the form of a visual flowchart, a compliance audit version, or other similar means.

[0053] In the above embodiments, risk factors, propagation paths, and target strategies are combined to generate a structured explanation of optimization logic, thereby reducing regulatory risks and improving user experience.

[0054] Please refer to Figure 2, which is a schematic flowchart of an insurance strategy generation method based on a risk tensor field provided by an embodiment of this application. This risk tensor field-based insurance strategy generation method can be applied to a server to transform user dynamic risk into a computable high-dimensional space state. It uniformly expresses the user's overall risk through the tensor space, preserving risk propagation relationships. This solves the shortcomings of existing technologies, such as fragmented multi-insurance product structures and the inability to integrate multi-source heterogeneous data, thereby improving the accuracy of insurance portfolio strategies generated based on the risk tensor field.

[0055] As shown in Figure 2, the insurance strategy generation method based on risk tensor field specifically includes steps S201 to S202.

[0056] S201. The multi-dimensional risk time series data is vectorized to generate risk vectors corresponding to each dimension of the risk time series data; in one embodiment, the latest acquired multi-dimensional risk time series data is input into the risk field mapping function f to obtain the risk tensor field at the current moment. .

[0057] In a specific embodiment, the unstructured / semi-structured time-series risk data is first vectorized to convert it into a numerical vector.

[0058] Specifically, key risk features are extracted from the time-series data for each risk dimension. For example, for the health dimension: key features such as average daily sleep duration, percentage of deep sleep, number of weekly exercise sessions, blood pressure, and blood sugar fluctuations are extracted from daily sleep / exercise / physical examination time-series data; for the vehicle dimension: features such as the number of emergency brakings per hour, percentage of nighttime driving, speeding frequency, and number of accident record updates are extracted from driving trajectory time-series data; for the family responsibility dimension: features such as changes in the number of children, number of elderly people supported, changes in mortgage balance, and the proportion of family income structure are extracted from family information update time-series data; for the asset dimension: features such as monthly income fluctuation, changes in stock investment proportion, and real estate value updates are extracted from financial data time-series data; for the external event dimension: quantitative features such as regional epidemic infection rate, extreme weather level, and policy adjustment coefficient are extracted from third-party time-series event data.

[0059] The extracted features are standardized in terms of dimensions to eliminate the influence of numerical differences. This includes using Min-Max (deviation standardization) to normalize numerical features (such as sleep duration and blood pressure) to the [0,1] interval, or using Z-score (standard score) standardization (mean 0, variance 1); using one-hot encoding or label encoding to convert categorical / discrete features (such as changes in family structure) into numerical features (such as single to value 0, married to value 1, and having children to value 2); and aligning all features according to the timestamp t to form a time-stamped feature sequence.

[0060] Standardized features in the same dimension are assembled into a time-stamped risk vector in a fixed order, with each vector element corresponding to the value of a feature at time t.

[0061] In one embodiment, a dynamic risk vector is generated for each dimension based on the features extracted from each dimension. The length of each vector is equal to the number of features extracted in that dimension, and it is dynamically updated over time t. For example, a health risk vector... [Sleep duration (t), percentage of deep sleep (t), number of exercise sessions per week (t), blood pressure (t), blood sugar (t),...].

[0062] S202. Map each of the risk vectors to a multidimensional tensor space to obtain the current risk tensor field.

[0063] Furthermore, the step of mapping each of the risk vectors to a multidimensional tensor space to obtain the current risk tensor field includes: performing a tensor product operation on each of the risk vectors to obtain a high-dimensional tensor; decomposing the high-dimensional tensor to obtain a core tensor and a factor matrix; and generating the current risk tensor field based on the core tensor and the factor matrix.

[0064] In one embodiment, the risk vectors of each dimension are combined into an initial high-dimensional tensor through tensor product operations. The tensor shape is: [number of health features] Vehicle feature number Family characteristic number Asset characteristics [Number of external event features], where, This represents the tensor product. The eigenvalues ​​of each risk vector are filled into the corresponding positions in the tensor; the eigenvalues ​​are the standardized values ​​of that eigenvalue.

[0065] In one embodiment, a multidimensional tensor model (such as Tucker decomposition, a high-order tensor decomposition method) is used to achieve the mapping. Specifically, the high-dimensional tensor is decomposed into the Tucker product of the core tensor (representing the potential correlation strength between risks) and the factor matrix (representing the weight distribution of features in each dimension), thus obtaining the risk tensor field at the current moment. Among them, the core tensor captures the risk propagation relationship, and the factor matrix ensures that the feature weights are consistent with the business logic (e.g., the weight of "blood pressure value" is higher than that of "sleep duration").

[0066] In the above embodiments, the user's dynamic risk is transformed into a computable high-dimensional space state. The user's overall risk is uniformly expressed through tensor space, and the risk propagation relationship is preserved. This solves the defects of existing technologies such as the fragmentation of multiple insurance types and the inability to integrate multi-source heterogeneous data, thereby improving the accuracy of insurance portfolio strategies generated based on the risk tensor field.

[0067] Please refer to Figure 3, which is a schematic flowchart of an insurance strategy generation method based on a risk tensor field provided by an embodiment of this application. This risk tensor field-based insurance strategy generation method can be applied to a server to predict risks based on multi-dimensional risk time-series data, obtain future risk change trends for users in advance, realize forward-looking risk assessment in intelligent insurance planning, dynamically adjust insurance portfolios to cope with future risks, and improve the comprehensiveness and accuracy of insurance portfolio strategies.

[0068] As shown in Figure 3, the insurance strategy generation method based on risk tensor field specifically includes steps S301 to S303.

[0069] S301. Perform risk prediction based on the multi-dimensional risk time series data to obtain predicted risk time series data; S302. Combine the multi-dimensional risk time series data and the predicted risk time series data, and construct the current risk tensor field based on the combined risk time series data; S303. Optimize the historical insurance portfolio based on the current risk tensor field to obtain the target insurance portfolio strategy.

[0070] In one embodiment, taking into account the characteristics of multi-dimensional risk time series data, a dimensional differentiated prediction approach is adopted, that is, different prediction schemes are used for different dimensions. For example, for the health risk dimension: for continuous time series data (such as sleep duration and blood pressure fluctuations), an LSTM (Long Short-Term Memory) model is used to capture long-term trends and nonlinear associations (such as the lagged effect of sleep deprivation on blood pressure); for the vehicle risk dimension: for driving behavior time series data (such as the frequency of sudden braking and the proportion of nighttime driving), a Prophet / ARIMA (Autoregressive Integrated Moving Average Model) model is used to fit periodic trends (such as differences in driving habits on weekdays and weekends); for the family responsibility dimension: for discrete event-type data (such as the birth of children and the settlement of mortgages), an event-driven probability model is used, combining user age, marital status, and other characteristics to predict the probability of event occurrence; for the asset dimension: for financial time series data (such as income fluctuations and investment returns), a GARCH (Autoregressive Conditional Heteroskedasticity) model is used. The model (autoregressive conditional heteroscedasticity model) captures volatility and peak-and-fat-tail characteristics; for external event dimensions, it connects to third-party professional prediction interfaces (such as extreme weather predictions from meteorological platforms and epidemic trend predictions from disease control centers) to directly obtain quantified future disturbance data.

[0071] Based on historical multi-dimensional risk time series data, we adopt dimensional differentiated prediction to obtain the predicted risk time series data of each dimension within the future time window T (such as sleep quality prediction for the next 7 days, driving risk trend for the next 7 days, and asset volatility prediction for the next 7 days).

[0072] In one embodiment, the current real risk time series data (historical data up to time t) and the predicted risk time series data are concatenated to form a complete "historical + future" time series, ensuring that the timestamps are continuous. For example: health dimension: concatenating the real sleep data of the past 30 days + the predicted sleep data of the next 7 days to form a complete 37-day series; external event dimension: concatenating the current epidemic data + the predicted epidemic trend data of the next 7 days to form a series that includes future perturbations.

[0073] In one embodiment, based on the concatenated complete time series, the multidimensional risk tensor field mapping is re-executed to obtain the current risk tensor field. Specifically, features are re-extracted from the concatenated sequence, the risk vectors of each dimension are updated according to the extracted features, the risk vectors of each dimension are decomposed using Tucker, the core tensor and factor matrix are updated, and the current risk tensor field is generated.

[0074] In one embodiment, the risk gap loss function and gradient parameters corresponding to the current risk tensor field are calculated to determine the adjustment direction and magnitude of each strategy parameter. The strategy parameters are then updated in the opposite direction of the gradient: adjusting the coverage amount (e.g., increasing critical illness insurance from 500,000 to 650,000), adding / removing insurance types (e.g., adding children's education insurance), and optimizing the allocation ratio (e.g., increasing the proportion of medical insurance). This process is repeated until the risk gap loss function value is less than a preset threshold (the coverage gap is acceptable) or the number of iterations reaches the target, thus obtaining the target insurance portfolio strategy.

[0075] In the above embodiments, risk prediction is performed based on multi-dimensional risk time series data, and the future risk change trend of users is obtained in advance. This realizes the forward-looking assessment of risks in intelligent insurance planning, dynamically adjusts the insurance portfolio to cope with future risks, and improves the comprehensiveness and accuracy of the insurance portfolio strategy.

[0076] Please refer to Figure 4, which is a schematic block diagram of an insurance strategy generation device based on a risk tensor field according to an embodiment of this application. This risk tensor field-based insurance strategy generation device is used to execute the aforementioned risk tensor field-based insurance strategy generation method. The risk tensor field-based insurance strategy generation device can be configured on a server.

[0077] As shown in Figure 4, the insurance strategy generation device 400 based on the risk tensor field includes: a risk time series data acquisition module 401, used to acquire multi-dimensional risk time series data corresponding to the target user when insurance portfolio strategy optimization is triggered; a risk tensor field acquisition module 402, used to construct the current risk tensor field based on the multi-dimensional risk time series data; a gap loss value acquisition module 403, used to analyze the historical insurance portfolio and the current risk tensor field to determine the risk gap loss value of the historical insurance portfolio; and an insurance portfolio strategy acquisition module 404, used to perform gradient optimization on the historical insurance portfolio based on the risk gap loss value to obtain the target insurance portfolio strategy.

[0078] Furthermore, the risk tensor field acquisition module 402 includes: a risk vector generation unit, used to vectorize the multi-dimensional risk time series data to generate risk vectors corresponding to the risk time series data of each dimension; and a risk tensor field acquisition unit, used to map each of the risk vectors to a multi-dimensional tensor space to obtain the current risk tensor field.

[0079] Furthermore, the risk tensor field acquisition unit includes: a high-dimensional tensor acquisition subunit, used to perform tensor product operation on each of the risk vectors to obtain a high-dimensional tensor; a high-dimensional tensor decomposition subunit, used to decompose the high-dimensional tensor to obtain a core tensor and a factor matrix; and a risk tensor field acquisition subunit, used to generate the current risk tensor field based on the core tensor and the factor matrix.

[0080] Furthermore, the insurance portfolio strategy acquisition module 404 includes: an adjustment parameter determination unit, used to determine the adjustment direction parameter and magnitude parameter of the historical insurance portfolio based on the gradient descent algorithm and the risk gap loss value; and a target strategy acquisition unit, used to adjust the historical insurance portfolio based on the direction parameter and magnitude parameter until the risk gap loss value is less than a preset threshold, and then obtain the target insurance portfolio strategy.

[0081] Furthermore, the insurance strategy generation device 400 based on the risk tensor field also includes a trigger monitoring module, which includes: a multi-dimensional risk data monitoring unit for monitoring the multi-dimensional risk data of the target user; a risk gap loss value determination unit for determining the risk gap loss value corresponding to the current multi-dimensional risk data and the historical insurance portfolio based on a preset risk gap loss function when an update to the multi-dimensional risk data is detected; and an insurance portfolio optimization trigger unit for determining to trigger insurance portfolio optimization when the risk gap loss value is greater than a preset threshold.

[0082] Furthermore, the insurance strategy generation device 400 based on the risk tensor field further includes: a risk time series data prediction module, used to perform risk prediction based on the multi-dimensional risk time series data to obtain predicted risk time series data; a risk tensor field acquisition module, used to combine the multi-dimensional risk time series data and the predicted risk time series data, and construct a current risk tensor field based on the combined risk time series data; and an insurance portfolio strategy acquisition module, used to optimize the historical insurance portfolio based on the current risk tensor field to obtain a target insurance portfolio strategy.

[0083] Furthermore, the insurance strategy generation device 400 based on the risk tensor field also includes an optimization logic description generation module. This module comprises: a risk factor determination unit, used to analyze the multi-dimensional risk time-series data to determine risk factors; a risk propagation path determination unit, used to determine the risk propagation path corresponding to the risk factor based on the current risk tensor field; and a logic description generation unit, used to generate a strategy optimization logic description based on the risk factor, the risk propagation path, and the target insurance portfolio strategy, for user viewing.

[0084] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the above-described apparatus and modules can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0085] The aforementioned device can be implemented as a computer program that can run on the computer device shown in Figure 5.

[0086] Please refer to Figure 5, which is a schematic block diagram of a computer device provided in an embodiment of this application. This computer device may be a server.

[0087] Referring to Figure 5, the computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0088] Non-volatile storage media can store operating systems and computer programs. These computer programs include program instructions that, when executed, cause the processor to perform any insurance policy generation method based on a risk tensor field.

[0089] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0090] Internal memory provides an environment for the execution of computer programs stored in non-volatile storage media. When the computer program is executed by the processor, it enables the processor to execute any insurance strategy generation method based on the risk tensor field.

[0091] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that the structure shown in Figure 5 is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0092] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0093] In one embodiment, the processor is configured to run a computer program stored in a memory to perform the following steps: when insurance portfolio strategy optimization is triggered, acquire multi-dimensional risk time-series data corresponding to the target user; construct a current risk tensor field based on the multi-dimensional risk time-series data; analyze historical insurance portfolios and the current risk tensor field to determine the risk gap loss value of the historical insurance portfolios; and perform gradient optimization on the historical insurance portfolios based on the risk gap loss value to obtain the target insurance portfolio strategy.

[0094] In one embodiment, when the processor constructs the current risk tensor field based on the multi-dimensional risk time series data, it is configured to: vectorize the multi-dimensional risk time series data to generate risk vectors corresponding to each dimension of risk time series data; and map each risk vector to a multi-dimensional tensor space to obtain the current risk tensor field.

[0095] In one embodiment, when the processor maps each of the risk vectors to a multidimensional tensor space to obtain the current risk tensor field, it performs the following: performs a tensor product operation on each of the risk vectors to obtain a high-dimensional tensor; decomposes the high-dimensional tensor to obtain a core tensor and a factor matrix; and generates the current risk tensor field based on the core tensor and the factor matrix.

[0096] In one embodiment, when the processor performs gradient optimization on the historical insurance portfolio based on the risk gap loss value to obtain a target insurance portfolio strategy, it is configured to: determine the adjustment direction parameter and magnitude parameter of the historical insurance portfolio based on the gradient descent algorithm and the risk gap loss value; adjust the historical insurance portfolio based on the direction parameter and magnitude parameter until the risk gap loss value is less than a preset threshold, and obtain the target insurance portfolio strategy.

[0097] In one embodiment, before the processor acquires the multi-dimensional risk time-series data corresponding to the target user when triggering insurance portfolio strategy optimization, it is further configured to: monitor the multi-dimensional risk data of the target user; when an update to the multi-dimensional risk data is detected, determine the risk gap loss value corresponding to the current multi-dimensional risk data and the historical insurance portfolio based on a preset risk gap loss function; and determine to trigger insurance portfolio optimization when the risk gap loss value is greater than a preset threshold.

[0098] In one embodiment, after acquiring multi-dimensional risk time-series data corresponding to the target user when triggering insurance portfolio strategy optimization, the processor is further configured to: perform risk prediction based on the multi-dimensional risk time-series data to obtain predicted risk time-series data; combine the multi-dimensional risk time-series data and the predicted risk time-series data, and construct a current risk tensor field based on the combined risk time-series data; optimize the historical insurance portfolio based on the current risk tensor field to obtain the target insurance portfolio strategy.

[0099] In one embodiment, after the processor performs gradient optimization on the historical insurance portfolio based on the risk gap loss value to obtain the target insurance portfolio strategy, it is further configured to: analyze the multi-dimensional risk time series data to determine risk factors; determine the risk propagation path corresponding to the risk factors based on the current risk tensor field; and generate a strategy optimization logic description based on the risk factors, the risk propagation path, and the target insurance portfolio strategy for user review.

[0100] The embodiments of this application also provide a computer-readable storage medium storing a computer program, the computer program including program instructions, and the processor executing the program instructions to implement any of the insurance strategy generation methods based on risk tensor fields provided in the embodiments of this application.

[0101] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0102] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for generating insurance strategies based on risk tensor fields, characterized in that, include: When insurance portfolio strategy optimization is triggered, multi-dimensional risk time series data corresponding to the target user is obtained; Construct the current risk tensor field based on the aforementioned multi-dimensional risk time series data; Analyze the historical insurance portfolio and the current risk tensor field to determine the risk gap loss value of the historical insurance portfolio; Based on the risk gap loss value, the historical insurance portfolio is subjected to gradient optimization to obtain the target insurance portfolio strategy.

2. The insurance strategy generation method based on risk tensor field according to claim 1, characterized in that, The step of constructing the current risk tensor field based on the multi-dimensional risk time series data includes: vectorizing the multi-dimensional risk time series data to generate risk vectors corresponding to each dimension of risk time series data; and mapping each risk vector to a multi-dimensional tensor space to obtain the current risk tensor field.

3. The insurance strategy generation method based on risk tensor field according to claim 2, characterized in that, The step of mapping each of the risk vectors to a multidimensional tensor space to obtain the current risk tensor field includes: performing a tensor product operation on each of the risk vectors to obtain a high-dimensional tensor; decomposing the high-dimensional tensor to obtain a core tensor and a factor matrix; and generating the current risk tensor field based on the core tensor and the factor matrix.

4. The insurance strategy generation method based on risk tensor field according to claim 1, characterized in that, The step of performing gradient optimization on the historical insurance portfolio based on the risk gap loss value to obtain the target insurance portfolio strategy includes: determining the adjustment direction parameter and magnitude parameter of the historical insurance portfolio based on the gradient descent algorithm and the risk gap loss value; adjusting the historical insurance portfolio based on the direction parameter and magnitude parameter until the risk gap loss value is less than a preset threshold, thereby obtaining the target insurance portfolio strategy.

5. The insurance strategy generation method based on risk tensor field according to claim 1, characterized in that, Before obtaining the multi-dimensional risk time-series data corresponding to the target user when triggering insurance portfolio strategy optimization, the method further includes: monitoring the multi-dimensional risk data of the target user; when an update to the multi-dimensional risk data is detected, determining the risk gap loss value corresponding to the current multi-dimensional risk data and the historical insurance portfolio based on a preset risk gap loss function; and determining to trigger insurance portfolio optimization when the risk gap loss value is greater than a preset threshold.

6. The insurance strategy generation method based on risk tensor field according to claim 1, characterized in that, When triggering insurance portfolio strategy optimization, after obtaining multi-dimensional risk time-series data corresponding to the target user, the process further includes: performing risk prediction based on the multi-dimensional risk time-series data to obtain predicted risk time-series data; combining the multi-dimensional risk time-series data and the predicted risk time-series data, and constructing a current risk tensor field based on the combined risk time-series data; and optimizing the historical insurance portfolio based on the current risk tensor field to obtain the target insurance portfolio strategy.

7. The insurance strategy generation method based on risk tensor field according to any one of claims 1 to 6, characterized in that, After performing gradient optimization on the historical insurance portfolio based on the risk gap loss value to obtain the target insurance portfolio strategy, the method further includes: analyzing the multi-dimensional risk time series data to determine risk factors; determining the risk propagation path corresponding to the risk factors based on the current risk tensor field; and generating a strategy optimization logic description based on the risk factors, the risk propagation path, and the target insurance portfolio strategy for user review.

8. An insurance strategy generation device based on a risk tensor field, characterized in that, include: The risk time series data acquisition module is used to acquire multi-dimensional risk time series data corresponding to the target user when the insurance portfolio strategy optimization is triggered. The risk tensor field acquisition module is used to construct the current risk tensor field based on the multi-dimensional risk time series data. The gap loss value acquisition module is used to analyze the historical insurance portfolio and the current risk tensor field to determine the risk gap loss value of the historical insurance portfolio. The insurance portfolio strategy acquisition module is used to perform gradient optimization on the historical insurance portfolio based on the risk gap loss value to obtain the target insurance portfolio strategy.

9. A computer device, characterized in that, The computer device includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and, when executing the computer program, implement the insurance strategy generation method based on the risk tensor field as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to implement the insurance strategy generation method based on a risk tensor field as described in any one of claims 1 to 7.

Images