Risk early warning method and device, computer equipment

Abstract

The embodiment of the application provides a risk early warning method and device, and a computer device, and belongs to the technical field of artificial intelligence. The method comprises the following steps: performing insurance risk assessment on a target insured object by using a target risk assessment model and an object portrait to obtain insurance risk assessment data; performing underwriting processing on the target insured object according to the insurance risk assessment data and the object portrait to obtain underwriting information; obtaining real-time insurance correlation data of the target insured object and a risk category of the real-time insurance correlation data according to the underwriting information; screening a selected risk prediction model from candidate risk prediction models according to the risk category; performing claim risk prediction by using the selected risk prediction model and the real-time insurance correlation data to obtain claim risk prediction data; and performing a target early warning operation on the target insured object according to the claim risk prediction data. The application can be applied to a business system that needs a large amount of data, such as financial technology, to save underwriting manpower, reduce claim risk and improve claim processing efficiency.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and is applied to the field of financial technology, particularly to a risk warning method and device, and computer equipment. Background Technology

[0002] Currently, the insurance industry can automate underwriting, claims risk assessment, and claims processing using artificial intelligence. However, the underwriting process still requires manual analysis of the insurable risk profile of the applicant before finalizing the underwriting using an underwriting model. Furthermore, after underwriting, traditional insurance risk warnings rely heavily on historical insurance data analysis following the initiation of a claim, lacking real-time monitoring and forward-looking prediction capabilities, resulting in a significant lag in risk warnings. For example, in auto insurance claims, the traditional damage assessment process requires manual inspection, which is time-consuming and hinders rapid response and intervention. Similarly, in disaster insurance, due to a lack of real-time data integration and analysis capabilities, insurance companies often only assess losses after a disaster occurs, failing to achieve pre-disaster warnings and risk reduction. Therefore, improving risk warning capabilities, reducing claims risk, and increasing claims efficiency have become pressing technical challenges. Summary of the Invention

[0003] The main objective of this application is to propose a risk warning method, apparatus, and computer equipment, which aims to improve the realization of risk warning, reduce claims risk, and improve claims efficiency.

[0004] To achieve the above objectives, a first aspect of this application proposes a risk warning method, the method comprising: Obtain a profile of the target insured individual; The insurance risk assessment of the target insured object is carried out by the preset target risk assessment model and the target profile to obtain insurance risk assessment data; Based on the insurance risk assessment data and the target profile, the underwriting process is carried out on the target insured object to obtain underwriting information; Based on the underwriting information, obtain the real-time insurance-related data of the target insured and the risk category of the real-time insurance-related data; Select a risk prediction model from the preset candidate risk prediction models based on the risk category; Claim risk prediction data is obtained by using the selected risk prediction model and the real-time insurance correlation data to predict claim risk. Based on the claims risk prediction data, a target early warning operation is performed on the target insured object.

[0005] In some embodiments, before performing an insurance risk assessment on the target insured object using a preset target risk assessment model and the object profile to obtain insurance risk assessment data, the method further includes: Obtain the raw training data; wherein, the raw training data includes historical profiles of the sample objects; The distribution difference of the sample objects in the original training data is evaluated to obtain distribution difference evaluation data; wherein, the distribution difference evaluation data characterizes the degree of distribution difference of the sample objects; Based on the distribution difference assessment data, the original training data is subjected to sample balancing to obtain the target training data; The target risk assessment model is obtained by training the preset original risk assessment model based on the target training data and preset training constraint parameters.

[0006] In some embodiments, the step of performing sample balancing on the original training data based on the distribution difference evaluation data to obtain the target training data includes: Based on the distribution difference assessment data, key difference features are extracted from the historical profile; wherein, the key difference features are object features whose distribution difference assessment data exceeds a preset threshold. Based on the distribution difference assessment data, a selected correction index is selected from the preset candidate correction indexes; Based on the key difference features and the selected correction index, training data simulation is performed to obtain candidate training data; The candidate training data and the original training data are mixed to obtain mixed training data; The authenticity of the mixed training data is evaluated to obtain authenticity evaluation data; The candidate training data is optimized based on the authenticity evaluation data to obtain the selected training data; The selected training data and the original training data are concatenated to obtain the target training data.

[0007] In some embodiments, obtaining the object profile of the target insured object includes: Obtain multi-source data of the target insured object; wherein, the multi-source data includes historical insurance data, insurance behavior data, insurance-related data, and exchange model parameters; The preset original portrait construction model is updated according to the exchange model parameters to obtain the target portrait construction model; The target insured object is profiled by constructing a profile using the target profile model, the historical insurance data, the insurance behavior data, and the insurance association data, thus obtaining the object profile.

[0008] In some embodiments, the underwriting process for the target insured object based on the insured risk assessment data and the object profile to obtain underwriting information includes: Based on the insurance risk assessment data, the key assessment features and their impact values ​​are extracted from the object profile. The key evaluation features and the evaluation impact values ​​are visualized to obtain an evaluation view; Underwriting is performed on the target insured object based on the assessment view.

[0009] In some embodiments, after performing underwriting processing on the target insured object based on the insurance risk assessment data and the object profile, the method further includes: Obtain the real-time insurance-related data of the target insured individual and the risk category of the real-time insurance-related data; Select a risk prediction model from the preset candidate risk prediction models based on the risk category; Claim risk prediction data is obtained by using the selected risk prediction model and the real-time insurance correlation data to predict claim risk. Based on the claims risk prediction data, a target early warning operation is performed on the target insured object.

[0010] In some embodiments, performing a target early warning operation on the target insured based on the claims risk prediction data includes: The aforementioned claims risk prediction data is processed to obtain an early warning level; The target warning operation is selected from the preset candidate warning operations based on the warning level; Perform the target early warning operation on the target insured object.

[0011] In some embodiments, after performing underwriting processing on the target insured object based on the insured risk assessment data and the object profile to obtain underwriting information, the method further includes: Receive the claim request from the target insured; Based on the claim request, obtain the insurance case characteristics of the target insured; Based on the characteristics of the insurance cases and the preset historical claims information, a claims level is obtained by classifying the claims into different levels. The selected claim processing mode is selected from the preset candidate claim processing modes according to the claim level; wherein the selected claim processing mode includes at least one of the following: manual processing mode and intelligent processing mode; The claims of the target insured are processed according to the selected claims processing mode.

[0012] In some embodiments, the characteristics of the insurance case include: insurance category, insured resource value, claim resource value, insured risk assessment data, and object characteristics; The process of classifying claims based on the characteristics of the insurance cases and pre-defined historical claims information to obtain claim levels includes: The case complexity is assessed based on the insurance category, the insured resource value, the compensation resource value, the insured risk assessment data, and the object characteristics to obtain case complexity assessment data; The difficulty of case processing is assessed based on the insurance category, the insured resource value, the compensation resource value, the insured risk assessment data, the object characteristics, and the historical claims information, resulting in case processing difficulty assessment data. The claim level is determined by classifying cases based on the case complexity assessment data and the case processing difficulty assessment data.

[0013] To achieve the above objectives, a second aspect of this application provides a risk warning device, the device comprising: The profile acquisition module is used to acquire the profile of the target insured object; The insurance risk assessment module is used to assess the insurance risk of the target insured object through a preset target risk assessment model and the object profile, and obtain insurance risk assessment data. The underwriting module is used to perform underwriting processing on the target insured object based on the insured risk assessment data and the object profile, and obtain underwriting information; The data acquisition module is used to acquire real-time insurance-related data of the target insured object and the risk category of the real-time insurance-related data based on the underwriting information; The model filtering module is used to select a risk prediction model from a preset pool of candidate risk prediction models based on the risk category. The claims risk prediction module is used to predict claims risk using the selected risk prediction model and the real-time insurance-related data, and to obtain claims risk prediction data. The risk warning module is used to perform target warning operations on the target insured object based on the claim risk prediction data.

[0014] To achieve the above objectives, a third aspect of the present application provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method described in the first aspect.

[0015] The risk warning method, apparatus, computer equipment, and storage medium proposed in this application, when assessing the risk of a target insured object, combine a target risk assessment model and an object profile to obtain insured risk assessment data. Then, based on the insured risk assessment data and the object profile, the underwriting of the target insured object is completed to obtain underwriting information, achieving automated underwriting and saving manpower. Next, a selected risk prediction model is selected from candidate risk prediction models based on the risk category. By selecting the risk prediction model and real-time insurance correlation data, the claim risk of the target insured object can be accurately predicted to obtain claim risk prediction data. Then, based on the claim risk prediction data, corresponding target warning operations are performed, providing early warning of claim risk. This not only reduces the actual claim risk level but also facilitates rapid loss assessment in subsequent claims, improving claims efficiency. Attached Figure Description

[0016] Figure 1 This is a flowchart of the risk warning method provided in the embodiments of this application; Figure 2 This is a flowchart of a risk warning method provided in another embodiment of this application; Figure 3 yes Figure 2 The flowchart of step S203 in the process; Figure 4 yes Figure 1 The flowchart of step S101 in the text; Figure 5 yes Figure 1 The flowchart of step S103 in the process; Figure 6 This is a schematic diagram of the assessment view in the risk warning method provided in the embodiments of this application; Figure 7 This is a schematic diagram of a risk warning method provided in another embodiment of this application; Figure 8 yes Figure 7 The flowchart of step S704 in the process; Figure 9 This is a schematic diagram of the target warning operation in the risk warning method provided in the embodiments of this application; Figure 10 This is a flowchart of a risk warning method provided in another embodiment of this application; Figure 11 This is a schematic diagram of manual claims processing in the risk warning method provided in this application embodiment; Figure 12 yes Figure 10 The flowchart of step S1003 in the process; Figure 13 This is an overall flowchart of the risk warning method provided in the embodiments of this application; Figure 14This is a schematic diagram of the risk warning device provided in the embodiments of this application; Figure 15 This is a schematic diagram of the hardware structure of the computer device provided in the embodiments of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0018] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0020] First, let's analyze some of the terms used in this application: Artificial intelligence (AI) is a new branch of computer science that studies, develops, and applies theories, methods, technologies, and systems to simulate, extend, and expand human intelligence. It aims to understand the essence of intelligence and produce intelligent machines that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can simulate the information processes of human consciousness and thought. Furthermore, AI utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to achieve optimal results.

[0021] Multi-source data refers to a collection of data from multiple different sources, such as different databases, sensors, systems, text, images, etc. It differs from multivariate data (multivariate data from a single source) in that its core characteristic lies in the integration and fusion of data across sources and modalities.

[0022] Generative Adversarial Networks (GANs) are generative models that learn to generate high-quality data through a game-like interaction between a generator and a discriminator. The basic idea of ​​GANs is that the generator attempts to produce data as realistic as possible, while the discriminator tries to distinguish between real and generated data. GANs are widely used in image generation, image inpainting, and unsupervised learning, offering advantages such as modeling data distribution and not requiring Markov chains, but they also suffer from problems like model instability.

[0023] Local Interpretable Model-agnostic Explanations (LIME) is a powerful "post-hoc explanation" tool that provides an easily understandable explanation for a single prediction of any machine learning model by simulating the behavior of a complex black-box model near a single prediction point using a locally simple "surrogate" model.

[0024] Shapley Additive Explanations (SHAP) is a method for interpreting the output of machine learning models by assigning an importance value to each feature for a specific prediction. It is based on Shapley values, a concept in cooperative game theory used to measure each participant's contribution to the collective outcome.

[0025] Streaming computing is a technology that processes large-scale data streams in real time, enabling analysis and computation the instant data arrives without waiting for all the data to arrive. This approach is particularly suitable for low-latency, high-throughput scenarios, such as real-time monitoring, financial transactions, and the Internet of Things (IoT).

[0026] Recurrent Neural Network (RNN) algorithms are a type of neural network used to process sequential data. Unlike feedforward neural networks, RNNs can capture the temporal dependencies in sequential data, making them very effective for processing time-series data such as text, speech, and video.

[0027] Deep learning is an important branch of artificial intelligence. It simulates the hierarchical information processing of the human brain through multi-layered neural networks, automatically learning feature representations from data. It is widely used in processing complex, high-dimensional data such as images, text, and speech. Its core lies in its end-to-end learning capability, which can directly extract features from raw data and complete tasks.

[0028] Graph Neural Networks (GNNs) are deep learning models specifically designed for processing graph-structured data. A graph consists of vertices and edges; nodes represent entities, and edges represent relationships between entities. Unlike traditional neural networks, GNNs can directly learn from the graph's topology and the features of its nodes and edges, making them ideal for processing non-Euclidean data such as social networks, molecular structures, and knowledge graphs.

[0029] Currently, the insurance industry faces the following main problems when applying AI technology for underwriting problem analysis and risk warning: Algorithmic bias and discrimination risks: Before underwriting, a risk assessment model is used to assess the risk of insured users, and the training of this model is highly dependent on the training data. If there are differences in the distribution of sample users in the training data, resulting in inherent biases and discrimination, the trained risk assessment model will amplify these biases, affecting risk assessment and the fairness of underwriting. For example, a health insurance model might increase premiums or refuse coverage for a specific group because certain diseases are prevalent in that group. This not only exacerbates unfairness in underwriting but could also trigger a crisis of trust in the insurance industry among consumers.

[0030] Based on this, embodiments of this application provide a risk warning method, apparatus, and computer equipment. The aim is to conduct risk assessment on a target insured object by combining a target risk assessment model and an object profile to obtain insured risk assessment data. Then, based on the insured risk assessment data and the object profile, underwriting of the target insured object is completed to obtain underwriting information, achieving automated underwriting and saving manpower. Next, a selected risk prediction model is selected from candidate risk prediction models based on the risk category. By selecting the risk prediction model and real-time insurance correlation data, the claim risk of the target insured object can be accurately predicted to obtain claim risk prediction data. Then, corresponding target warning operations are performed based on the claim risk prediction data, providing early warning of claim risks. This not only reduces the actual claim risk level but also facilitates rapid loss assessment in subsequent claims, improving claims efficiency.

[0031] The risk warning method, apparatus, and computer equipment provided in this application are specifically described through the following embodiments. First, the risk warning method in this application is described.

[0032] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0033] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0034] The risk warning method provided in this application relates to the field of artificial intelligence technology and is applied to fintech scenarios. The risk warning method provided in this application can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the risk warning method, but is not limited to the above forms.

[0035] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer computer devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0036] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0037] Figure 1 This is an optional flowchart of the risk warning method provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S107.

[0038] Step S101: Obtain the object profile of the target insured object; Step S102: Conduct an insurance risk assessment on the target insured object using a preset target risk assessment model and object profile to obtain insurance risk assessment data; Step S103: Based on the risk assessment data and the target profile, conduct underwriting processing on the target insured to obtain underwriting information; Step S104: Obtain the real-time insurance-related data and risk category of the target insured object based on the underwriting information; Step S105: Select a risk prediction model from the preset candidate risk prediction models according to the risk category; Step S106: By selecting a risk prediction model and real-time insurance-related data, claim risk prediction data is obtained. Step S107: Perform target early warning operation on the target insured object based on the claims risk prediction data.

[0039] Steps S101 to S107, as illustrated in this embodiment, involve combining a target risk assessment model and a target profile to assess the insured's risk and obtain insured risk assessment data. Underwriting for the target insured is then completed based on this data and the target profile. This achieves accurate and automated underwriting, saving manpower. Simultaneously, a selected risk prediction model is chosen from candidate risk prediction models based on the risk category. By selecting the risk prediction model and using real-time insurance correlation data, the claim risk of the target insured can be accurately predicted, yielding claim risk prediction data. Corresponding early warning operations are then performed based on this data, providing advance warning of claim risks. This not only reduces the actual claim risk level but also facilitates rapid loss assessment and improves claims efficiency.

[0040] In some embodiments, prior to step S102, the risk warning method may also include, but is not limited to, steps S201 to S204: Step S201: Obtain the original training data; wherein, the original training data includes the historical profiles of the sample objects; Step S202: Evaluate the distribution differences of the sample objects in the original training data to obtain distribution difference evaluation data; wherein, the distribution difference evaluation data characterizes the degree of distribution difference of the sample objects. Step S203: Perform sample balancing on the original training data based on the distribution difference assessment data to obtain the target training data; Step S204: Train the preset original risk assessment model based on the target training data and preset training constraint parameters to obtain the target risk assessment model.

[0041] In step S201 of some embodiments, the original training data can be obtained from a development data source or from a historical database on the insurance business platform, and is not limited to these. The original training data includes historical profiles of multiple sample objects, and the historical profiles are constructed based on underwriting-related historical object data. It should be noted that the sample objects can be users, vehicles, fields, buildings, etc., and the distribution of sample objects differs for different categories of sample objects, and the measurement of distribution differences also differs. If the sample object is a user, the historical object data includes: basic information of the historical object, historical medical records, historical insurance records, etc. If the sample object is a vehicle, the historical object data includes historical maintenance records, historical insurance records, and historical driving data, etc. Therefore, the historical object data collected is different for different categories of sample objects, but the constructed historical profiles are all related to insurance risk assessment, so the historical profiles can accurately and comprehensively characterize the risk situation of the target object in terms of insurance.

[0042] In step S202 of some embodiments, the distribution difference assessment is used to evaluate the distribution difference of sample objects in the original training data. Specifically, this embodiment extracts the portrait features of each sample object in the original training data and obtains distribution difference assessment data by statistically analyzing the degree of difference of sample objects in different portrait features. For example, this embodiment uses the Demographic Parity Difference algorithm to detect the distribution difference of portrait features of sample objects in the original training data to obtain distribution difference assessment data. If the sample object is a user, and the original risk assessment model is used to assess the risk of the user's health insurance, the original training data is the user's historical health profile. If the sample size of the rare disease patient group (such as group A) is only 1 / 10 of that of other groups, and the distribution difference assessment data is determined to be the distribution difference value of the original training data exceeding a preset threshold, it will lead to insufficient confidence of the original risk assessment model in assessing the user risk of the rare disease patient group.

[0043] In step S203 of some embodiments, sample balancing involves adjusting sample objects in the original training data whose distribution differences exceed a preset threshold, making the distribution of sample objects in the original training data more balanced. The balanced target training data can train an accurate target risk assessment model and also allows the target risk assessment model to assess different groups more fairly during risk assessment.

[0044] In step S204 of some embodiments, training constraint parameters are used to constrain the training of the original risk assessment model. These training constraint parameters are also called distribution equilibrium loss functions. The selected training data generated by the GAN model is used as regularization items through the distribution equilibrium loss function, forcing the original risk assessment model to learn an unbiased decision boundary. For example, in health insurance risk assessment, the risk assessment for group A is relatively high. Adding training constraint parameters during the training of the original risk assessment model improves the consistency between the original risk assessment model and the risk assessment of the mainstream group.

[0045] It should be noted that the training constraint parameters include the group equality constraint parameter and the opportunity equality constraint parameter. The group equality constraint parameter requires that the difference in insurance risk assessment among different groups does not exceed a preset difference threshold; the opportunity equality constraint parameter ensures that the true underwriting rate of high-risk groups is consistent with the underwriting rate of the mainstream group.

[0046] In some embodiments, the trained target risk assessment model also tracks its risk assessment deviation data in real time, and the risk assessment deviation data characterizes the fairness of the target risk assessment model in assessing insurance risks.

[0047] For example, if the original risk assessment model is applied to a health insurance scenario, F1 disease is highly prevalent among young people, but the amount of data is insufficient. This causes the original risk assessment model to classify all users with F1 disease as high-risk, resulting in a high rejection rate for F1 disease users. By using a GAN model to generate synthetic medical records (e.g., diagnosis codes, treatment costs, etc.) for young patients, the original training data can be optimized to obtain target training data. If the original risk assessment model has a mean risk assessment of 0.8 for users with F1 disease, while the mean risk assessment of other groups is 0.5, the target risk assessment model trained with the optimized target training data has a mean risk assessment of 0.65 for users with F1 disease. Therefore, the accuracy of the target risk assessment model in assessing the risk of young people is improved by 18%, and the difference in rejection rate is reduced to 40%.

[0048] If the original risk assessment model is applied to a car insurance scenario, and the original training data consists of historical driving data, and nighttime driving data leads the original risk assessment model to evaluate nighttime accidents as high-risk, then a cGAN model is used to generate nighttime driving data to augment the original training data, resulting in target training data. Therefore, the target risk assessment model trained with the target training data can improve the accuracy of nighttime accident risk assessment from 0.62 to 0.79. Thus, when there is excessive variation in the sample distribution of the original training data, sample balancing is performed to obtain target training data, and the target risk assessment model trained with the target training data improves the accuracy of risk assessment.

[0049] In steps S201 to S204 of this embodiment, distribution difference assessment data is obtained by acquiring the original training data and evaluating the degree of distribution difference among the sample objects in the original training data. Based on the distribution difference assessment data, the original training data is balanced to obtain target training data. The original risk assessment model is then trained using the target training data and training constraint parameters to obtain the target risk assessment model. Therefore, before training the original risk assessment model, the original training data with an unbalanced sample distribution is adjusted to the target training data with a balanced sample distribution, correcting the sample distribution bias in the original training data. The target risk assessment model trained using the target training data provides more accurate risk assessment and improves the fairness of risk assessment for different objects. Therefore, during the underwriting process, insurance risk assessment can be accurately completed, reducing bias in insurance risk assessment and realizing a shift from "historical bias replication" to "dynamic fair decision-making," ensuring the accuracy of underwriting while avoiding the risk of underwriting discrimination.

[0050] Please see Figure 3 In some embodiments, step S203 may include, but is not limited to, steps S301 to S307: Step S301: Extract key features of the differences from the historical profile based on the distribution difference assessment data; wherein, the key features of the differences are the object features that exceed the preset threshold in the distribution difference assessment data. Step S302: Select the correction index from the preset candidate correction indexes based on the distribution difference assessment data; Step S303: Simulate training data based on key difference features and selected correction indicators to obtain candidate training data; Step S304: Mix the candidate training data and the original training data to obtain mixed training data; Step S305: Perform a realism assessment on the mixed training data to obtain realism assessment data; Step S306: Optimize the candidate training data based on the authenticity evaluation data to obtain the selected training data; Step S307: The selected training data and the original training data are concatenated to obtain the target training data.

[0051] In step S301 of some embodiments, the key difference feature is the key feature that causes the distribution difference between sample objects, and it is also the object feature in the distribution difference assessment data where the distribution difference value exceeds a preset threshold. For example, this embodiment uses technologies such as SHAP / LIME to identify the key difference features that cause the sample distribution difference in the original training data. If the sample object is a user, the key difference feature can be the age distribution, frequency of medical visits, etc. of the rare disease group (sample objects of group A).

[0052] In step S302 of some embodiments, the candidate correction index is an index that indicates the adjustment of the risk assessment of the sample objects in the original training data by the original risk assessment model, and indirectly indicates the adjustment of the sample objects in the original training data. Different candidate correction indices correspond to different distribution difference assessment data. For example, if the selected correction index is to increase the risk assessment probability of group A to 0.8 based on the distribution difference assessment data, then...

[0053] In step S303 of some embodiments, training data simulation is performed by combining key difference features and selected correction indicators to generate simulated profiles of sample objects that produce distributional differences, thus obtaining candidate training data. This embodiment uses a generative model (GAN model) to generate candidate training data. Key difference features (such as age stratification and region of group A) and selected correction indicators are input into the generative model, which outputs candidate training data simulating the real data distribution of sample objects, such as medical records and insurance policy features of group A sample objects.

[0054] In steps S304 to S305 of some embodiments, the generated candidate training data may not accurately represent the profile of the sample object. Therefore, it is necessary to mix the candidate training data and the original training data into mixed training data, and then perform a realism evaluation on the mixed training data to obtain realism evaluation data. It should be noted that the realism evaluation data represents the authenticity of the data, and the realism of the mixed training data is judged by a discriminator.

[0055] In step S306 of some embodiments, the authenticity evaluation data is fed back to the GAN model, which can optimize the data generation strategy and obtain selected training data by optimizing the candidate training data.

[0056] In step S307 of some embodiments, the optimized selected training data and the original training data are concatenated to form target training data, which achieves the balance of sample objects.

[0057] In some embodiments, candidate training data for minority classes are generated on demand based on the differential distribution assessment data. For example, if the sample size of group A is less than 5%, additional candidate training data is generated using the GAN model to increase the sample size of group A to 20%. It should be noted that fairness constraints are embedded during the generation process (e.g., the group equality difference in rare disease risk prediction ≤ 0.05). The generated candidate training data and the original training data are merged to form mixed training data. Cross-validation is used to ensure the quality of the generated candidate training data. The candidate training data is then optimized to obtain selected training data, and the feature distribution of the selected training data has a divergence of < 0.1 compared to the original training data. Finally, the selected training data and the original training data are merged into target training data, constructing target training data with a more balanced sample distribution.

[0058] In steps S301 to S307 of this embodiment, key difference features are extracted from the original training data based on the distribution difference assessment data, and then a correction index is selected from the candidate correction indexes based on the distribution difference assessment data. The key difference features and the selected correction index are combined to generate candidate training data, achieving sample balance in the original training data. Further, the candidate training data and the original training data are mixed to form mixed training data, and the authenticity of the mixed training data is judged to obtain authenticity assessment data. Based on the authenticity assessment data, the candidate training data is optimized into selected training data, and finally, the selected training data and the original training data are combined to form the target training data. Therefore, in the process of balancing the original training data samples, training data of sample objects that produce distribution bias is simulated, and in order to improve the authenticity of the generated training data, the authenticity of the data is further evaluated, constructing selected training data with strong authenticity that can balance the problem of imbalanced sample object distribution in the original training data.

[0059] In step S101 of some embodiments, after the target insured individual selects an insurance category on the insurance business platform, underwriting of the target insured individual is required in advance to determine whether the target insured individual is eligible for insurance. Therefore, it is necessary to obtain an object profile of the target insured individual, and the object profile includes key object characteristics for insurance risk assessment. It should be noted that the data for constructing the object profile can come not only from the insurance business platform but also from external platforms, thereby improving the accuracy of insurance risk assessment.

[0060] Currently, customer information, business data, and financial data accumulated within insurance business platforms are typically scattered across different systems, forming data silos. Meanwhile, significant barriers exist to external data acquisition. Constrained by factors such as data security management, departments like agriculture, meteorology, and disaster prevention are reluctant to share data or open access ports to institutions. This data fragmentation makes it difficult to build complete client profiles; profiles can only be constructed using partially publicly available data, impacting underwriting accuracy.

[0061] Please see Figure 4 In some embodiments, step S101 may include, but is not limited to, steps S401 to S403: Step S401: Obtain multi-source data of the target insured; wherein, the multi-source data includes historical insurance data, insurance behavior data, insurance-related data, and exchange model parameters; Step S402: Update the preset original portrait construction model according to the exchange model parameters to obtain the target portrait construction model; Step S403: A profile of the target insured object is constructed by using the target profile construction model, historical insurance data, insurance behavior data, and insurance-related data to obtain the object profile.

[0062] In step S401 of some embodiments, the multi-source data refers to data from multiple data sources, including internal and external data sources. The multi-source data includes historical insurance data, insurance behavior data, insurance-related data, and exchange model parameters. Historical insurance data and insurance behavior data are collected through internal data sources, and include historical policy information and historical claims records. Insurance-related data is collected through external data sources, and is collected according to the category of the target insured object. Specifically, if the target insured object is a vehicle, the insurance-related data includes driving behavior data, vehicle status data, and real-time traffic data. Driving behavior data includes: number of emergency brakings, high-speed driving duration, and nighttime driving frequency. Vehicle status data includes OBD fault codes and annual inspection records. Real-time traffic data includes traffic congestion index and accident-prone road sections. Specifically, driving behavior data and vehicle status data are collected in real time through an in-vehicle terminal or mobile terminal, and real-time traffic data uploaded by traffic management departments is also collected to obtain the vehicle's insurance-related data. If the target insured object is agricultural products, the associated data includes meteorological data, soil sensor data, and satellite remote sensing images. Meteorological data includes precipitation and temperature anomalies, soil sensor data includes soil temperature and pH, and satellite remote sensing images include crop growth information and disaster coverage. Meteorological and soil sensor data are obtained through the meteorological bureau's open-source API interface and field IoT devices. If the target insured object is a user who needs to purchase health insurance, the associated data includes historical medical records, wearable device data, and genetic testing data. Historical medical records include outpatient / inpatient medical history, and wearable device data includes heart rate and step count collected by wearable devices. It should be noted that historical medical records and genetic testing data are provided in conjunction with the medical institution where the target insured object resides.

[0063] In step S402 of some embodiments, since some data involves privacy and cannot be directly provided through external data sources, this embodiment employs federated learning technology to achieve multi-party joint modeling without directly sharing data. Therefore, each data source node trains an original profile construction model locally. By exchanging model parameters, the model parameters sent by the external data source are defined as the exchanged model data. The original profile construction model is adjusted into a target profile construction model based on the exchanged model parameters. Thus, the optimized target profile construction model incorporates data from the external data source without directly acquiring data from it, reducing the risk of data leakage. For example, in cross-institutional anti-fraud identification scenarios, multiple insurance companies can jointly train a target profile construction model based on federated learning, improving the accuracy of profile construction and reducing the risk of data leakage.

[0064] In step S403 of some embodiments, historical insurance data, insurance behavior data, and insurance-related data are input into the target profile construction model to construct a profile, thereby obtaining an object profile of the target insured. Therefore, the construction of the object profile not only includes static basic information but also integrates dynamic behavioral data, real-time related data, etc., to construct a more comprehensive profile representing the target insured's insurance risk.

[0065] In steps S401 to S403 of this embodiment, when constructing the target insurance profile, not only are historical insurance data and insurance behavior data collected from internal data sources, but also insurance-related data and exchange model parameters from external data sources are collected. The exchange model parameters optimize the local original profile construction model to obtain the target profile construction model, ensuring the accuracy of profile construction and reducing the risk of data leakage. Finally, a comprehensive object profile of the target insurance applicant is constructed using the target profile construction model, multi-source historical insurance data, insurance behavior data, and insurance-related data.

[0066] In step S102 of some embodiments, the target insured object is assessed for insurance risk using a target risk assessment model, mainly involving a multi-dimensional risk assessment. It should be noted that the assessment method of the target risk assessment model differs for different categories of target insured objects.

[0067] For example, if the target insured object is a target vehicle, a target vehicle profile is generated by combining the target vehicle's historical insurance data, insurance behavior data, driving behavior data, vehicle usage data, and real-time road condition data. Then, the profile features are extracted, including historical insurance features, insurance behavior features, driving features, vehicle features, and road condition features. By assigning weights to each feature, the risk assessment data is obtained by using a target risk assessment model to assess the historical insurance features, insurance behavior features, driving features, vehicle features, and road condition features.

[0068] In step S203 of some embodiments, the insurance risk assessment data directly affects the underwriting of the target insured. If the insurance risk assessment data is higher than or equal to a set assessment threshold, it indicates that the target insured cannot purchase the selected insurance. If the insurance risk assessment data is lower than the set assessment threshold, it indicates that the target insured can purchase the selected insurance.

[0069] It should be noted that traditional underwriting lacks transparency and explainability, making it difficult for insurance companies to justify their underwriting results and exposing them to compliance and reputational risks. To address this, this embodiment provides an assessment view after underwriting to explain the factors influencing the underwriting results, making underwriting decisions more transparent.

[0070] Please see Figure 5In some embodiments, step S103 may include, but is not limited to, steps S501 to S503: Step S501: Extract the key assessment features and their impact values ​​from the target profile based on the insurance risk assessment data. Step S502: Visualize the key features and impact values ​​to obtain the evaluation view; Step S503: Perform underwriting processing on the target insured object according to the assessment view to obtain underwriting information.

[0071] In step S501 of some embodiments, the key features being evaluated are the object features that influence the risk assessment data of the target insurance profile, and the evaluation impact value characterizes the degree of influence of the key underwriting features on the insurance risk assessment. Specifically, the evaluation impact value is obtained by quantifying the impact of each object feature in the object profile on the insurance risk assessment using SHAP technology. For example, in health insurance risk assessment, the key features being evaluated are "genetic risk score" and "abnormal frequency of medical visits"; in auto insurance risk assessment, the key features being evaluated are "nighttime driving duration" and "frequency of emergency braking".

[0072] In step S502 of some embodiments, the visualization process mainly uses a visualization template to generate an evaluation view from the key evaluation features and evaluation impact values. The evaluation view also displays the insurance risk assessment data and the key evaluation features of the assessment process, and uses color depth to represent the degree of contribution of different key evaluation features to the insurance risk assessment. For example... Figure 6 As shown, Figure 6 The diagram illustrates the assessment view. In the risk assessment of auto insurance, the key assessment features are identified as "nighttime driving duration" and "frequency of emergency braking." Since the assessment impact value of "nighttime driving duration" is higher than that of "frequent emergency braking," the "nighttime driving duration" is colored darker, indicating that it is a high-risk factor. The diagram also presents the hierarchical judgment logic of the target risk assessment model in flowchart form. For example, "obtain the number of emergency braking instances > if the number of emergency braking instances is greater than 5 times / hour → trigger the high-risk label").

[0073] It should be noted that the assessment view can also be adjusted in response to user actions, including parameter adjustment simulation and case comparison modes. Parameter adjustment simulation allows users to manually modify feature values ​​in the assessment view (such as "reduce nighttime driving time"), updating the insurance risk assessment data in real time. Case comparison mode displays side-by-side the different decision outcomes of similar customers (e.g., the same age, policy type) due to a single feature difference.

[0074] In step S503 of some embodiments, the underwriting of the target insured is completed according to the evaluation view, so as to show the target insured or underwriter the key evaluation features and evaluation impact values ​​in the underwriting process, thereby enhancing the transparency and credibility of the underwriting process.

[0075] In steps S501 to S503 of this embodiment, during the underwriting process, the key assessment features and their impact values ​​for the risk assessment of the insured are extracted, and an assessment view is generated by combining the key assessment features and their impact values. This view clearly shows the key influencing factors and their weights to underwriters and target insured persons, thereby enhancing the transparency and credibility of the underwriting process.

[0076] After underwriting is completed, traditional insurance risk warnings mainly rely on the analysis of historical insurance data after a claim is initiated, lacking real-time monitoring and forward-looking prediction capabilities, resulting in a significant lag in risk warnings. For example, in the auto insurance claims process, the traditional loss assessment process requires manual inspection, which is time-consuming and cannot provide rapid response and intervention. Similarly, in disaster insurance, due to the lack of real-time data integration and analysis capabilities, insurance companies often only conduct loss assessments after a disaster occurs, failing to achieve pre-disaster warnings and risk reduction. To address this, this embodiment collects real-time insurance-related data of the target insured after underwriting and uses this data for risk warnings, realizing a shift from a "post-disaster remediation" to a "pre-disaster warning" model.

[0077] Please see Figure 7 In some embodiments, after step S103, the risk warning method may also include, but is not limited to, steps S701 to S704: Step S701: Obtain the real-time insurance-related data and risk category of the real-time insurance-related data for the target insured object; Step S702: Select a risk prediction model from the preset candidate risk prediction models according to the risk category; Step S703: By selecting a risk prediction model and real-time insurance-related data, claim risk prediction data is obtained. Step S704: Perform a target early warning operation on the target insured object based on the claims risk prediction data.

[0078] In step S701 of some embodiments, the real-time insurance-related data is multimodal data related to the insurance risks of the target insured object, and is collected in real time through various channels such as APIs, IoT devices, and third-party data interfaces. For example, in disaster insurance scenarios, real-time insurance-related data is obtained by integrating satellite remote sensing data, weather forecast data, geological sensor data, etc., to construct a comprehensive natural disaster monitoring network. In auto insurance scenarios, driving behavior data is collected in real time through in-vehicle devices or mobile devices as real-time insurance-related data, and the driving behavior data includes ten dimensions such as mileage, number of emergency brakings, and high-speed driving duration.

[0079] In step S702 of some embodiments, candidate risk prediction models adapted to different risk categories are set. Specifically, if the risk category is a time-series risk, such as climate change-related disaster risk, the corresponding candidate risk prediction model is a recurrent neural network, such as LSTM or GRU neural networks. If the risk category is an image-based risk, such as vehicle damage assessment, the corresponding candidate risk prediction model is a deep learning model such as CNN. If the risk category is a complex multidimensional risk, the corresponding candidate risk prediction model uses a graph neural network. Therefore, selecting the appropriate risk prediction model from the candidate risk prediction models for different risk categories is the core step in building an efficient and accurate risk early warning system. Risk prediction is not based on a single model, but rather achieves intelligent matching of the selected risk prediction model through a four-step strategy of risk type identification, modality matching, algorithm adaptability evaluation, and dynamic ensemble optimization.

[0080] It should be noted that risk categories include: temporal risks, image-based risks, and complex multidimensional risks. Temporal risks are those with strong time dependence, whose future state is highly dependent on historical sequences. Typical scenarios include typhoon path prediction due to climate change, early warning of customer health indicator trends, and financial market volatility. The characteristics of temporal risks are continuity, trends, periodicity, and suddenness. Image-based risks are risks primarily carried by visual information, requiring feature extraction from spatial structures. Typical scenarios include vehicle damage assessment and crop pest and disease identification in agricultural insurance. Their characteristics include high-dimensional pixel data and a balance between local texture and global structure. Complex multidimensional risks involve the design of risks that are interconnected and propagated among multiple entities, exhibiting networked and non-linear propagation characteristics. Typical scenarios include supply chain disruption risks and insurance fraud risk identification. Their characteristics include graph-structured data, inter-node dependencies, and risk cascading effects.

[0081] In step S703 of some embodiments, real-time insurance association features are first extracted from the real-time insurance association data, and the real-time insurance association features corresponding to different risk categories are different. Specifically, for time-series risks, the extracted real-time insurance association features are time-series features; for image-related risks, the extracted real-time insurance association features are image features; and for complex multidimensional risks, the extracted real-time insurance association features are graph structure features. The real-time insurance association features are input into a selected risk prediction model to predict claims risk, and the claims risk prediction data represents the probability of claim risk prediction for the target insured object.

[0082] Specifically, for time-series risks, recurrent neural network models such as LSTM and GRU are used, which can capture long-term dependencies and solve the gradient vanishing problem of traditional RNNs. These models are suitable for nonlinear and non-stationary time series, such as sudden changes in meteorological data or gradual deterioration of customer health indicators. For image-related risks, CNN models are used. CNN models, through local receptive fields and weight sharing, can capture spatial hierarchical features in images, improving the accuracy of image-related risk predictions. For complex multidimensional risks, GNN models are used. GNN models aggregate neighbor node information through message passing mechanisms and learn the embedding identifiers of nodes and graphs, thus improving the prediction accuracy of complex multidimensional risks.

[0083] In step S704 of some embodiments, a target early warning operation is determined based on the claims risk prediction data, and different target early warning operations are set according to the height of the claims risk prediction probability.

[0084] In steps S701 to S704 of this embodiment, multiple machine learning algorithms are integrated to select a risk prediction model from candidate risk prediction models based on the risk category. By selecting the risk prediction model and real-time insurance-related data, the claim risk of the target insured can be accurately predicted to obtain claim risk prediction data. Then, based on the claim risk prediction data, corresponding target early warning operations are performed to provide early warning of claim risks. This not only reduces the actual claim risk level but also facilitates the rapid completion of loss assessment in subsequent claims.

[0085] Please see Figure 8 In some embodiments, step S704 includes, but is not limited to, steps S801 to S803: Step S801: Perform early warning classification processing on the claims risk prediction data to obtain the early warning level; Step S802: Select the target warning operation from the preset candidate warning operations according to the warning level; Step S803: Perform target early warning operation on the target insured object.

[0086] In step S801 of some embodiments, the claims risk prediction data includes claims risk prediction probability and risk impact data. The claims risk prediction probability, risk impact data and a preset early warning classification table are combined to determine the early warning level.

[0087] In step S802 of some embodiments, a higher warning level indicates a higher risk of claims for the target insured and a more severe impact of the risk; conversely, a lower warning level indicates a lower risk of claims for the target insured and a less severe impact of the risk. Therefore, determining different target warning actions for different warning levels not only allows claims personnel to understand the current claims risk but also allows policyholders to understand the risk of their insured target, enabling them to take preventative measures and reduce the impact of the risk.

[0088] In step S803 of some embodiments, a target early warning operation is performed on the target insured object, specifically by sending corresponding early warning information to the insured user of the target insured object through multiple communication channels. This embodiment divides the early warning levels into Level 1, Level 2, and Level 3. Corresponding early warning information is sent to different early warning levels through various communication channels such as SMS, APP, and email, using different colored early warning labels. For example, Level 1 uses a yellow early warning label, Level 2 uses an orange early warning label, and Level 3 uses a red early warning label.

[0089] In steps S801 to S803 of this embodiment, after classifying the claim risk prediction data into early warning levels, corresponding early warning operations are performed on the target insured individuals based on the early warning level. This accurately provides early warning prompts to the target insured individuals, enabling them to take precautions against risks in advance and reduce the damage caused by risks.

[0090] Please refer to Figure 9 If the target insured individuals are those purchasing disaster risk insurance, data can be collected in advance on users potentially affected by a typhoon before its landfall. Based on current typhoon data, the system can predict current claim risk. If the warning level is determined to be Level 2, a message can be sent to affected users stating, "Your area will be affected by a typhoon, and the risk level is Level 2. Please take precautions. xxxxx." Therefore, providing advance risk prediction and warnings to the target insured individuals helps them take preventative measures, reducing the impact of disasters and lowering claim risk.

[0091] Insurance operations are often over-automated, lacking humanized adjustment mechanisms. Relying solely on algorithms to handle special cases can lead to unreasonable outcomes. For example, in the claims process, AI systems may reject reasonable claims requests due to their inability to identify special circumstances, resulting in a decreased user experience on the insurance platform. Simultaneously, the ineffective integration of human resources leads to low efficiency in front-end and back-end collaboration. While AI-assisted insurance agents may improve work efficiency, it cannot replace the emotional connection and personalized service provided by human interaction. To address this, this embodiment establishes an intelligent task distribution mechanism that automatically assigns the claims processing mode—whether it's human or AI customer service—based on the case characteristics in the claims request, improving claims processing efficiency and enhancing the user experience during the claims process.

[0092] Please see Figure 10 In some embodiments, after step S103, the risk warning method may also include, but is not limited to, steps S1001 to S1005: Step S1001: Receive the claim request from the target insured; Step S1002: Obtain the insurance case characteristics of the target insured based on the claim request; Step S1003: Classify the claims according to the characteristics of the insurance case and the preset historical claims information to obtain the claims level; Step S1004: Select a claim processing mode from the preset candidate claim processing modes according to the claim level; wherein, the selected claim processing mode includes at least one of the following: manual processing mode and intelligent processing mode. Step S1005: Process the claims of the target insured according to the selected claims processing mode.

[0093] In step S1001 of some embodiments, the claim request is a request initiated by the target insured on the insurance business platform for the target case, and the claim request includes claim case information and claim basis information. The claim case information is the insurance policy of the type of insurance purchased by the target insured, and the claim basis information is the documents required for the claim. For example, in the case of a car insurance claim, the claim case information is the car insurance policy purchased by the user, and the claim basis information is the images of the vehicle to be compensated and the repair records. In the case of a health insurance claim, the claim case information is the car insurance policy purchased by the user, and the claim basis information is the medical records and medical diagnosis records.

[0094] In step S1002 of some embodiments, the insurance case characteristics include: insurance type, insured resource value, compensation resource value, insured risk assessment data, and object characteristics, etc.

[0095] In step S1003 of some embodiments, historical claims information includes historical claims case characteristics and historical claims processing characteristics. The historical claims information and insurance case characteristics are combined to classify the cases into levels, mainly into the target cases of the target insured.

[0096] In step S1004 of some embodiments, a mapping relationship between each claim level and candidate claim processing modes is set, so the selected claim processing mode can be determined directly based on the claim level. Specifically, the claim levels in this embodiment include: simple level, medium level, and high level. The selected claim processing mode for the simple level is the intelligent processing mode, the selected claim processing modes for the medium level are the manual processing mode and the intelligent processing mode, and the selected claim processing mode for the high level is the manual processing mode.

[0097] Please refer to Figure 11 If the target case is a car insurance injury case, the claim level is determined to be medium, and the claim processing mode is selected as a combination of manual processing mode and intelligent processing mode. The intelligent customer service first conducts a preliminary review of the target case to generate a claim list, and then the human reviews the claim list and the target case. This not only improves the efficiency of claim processing, but also ensures accurate completion of the claim processing.

[0098] In step S1005 of some embodiments, if the selected claims processing mode includes manual processing mode, the interface where the human customer service is located needs to be called to process the claims. Alternatively, if the selected claims processing mode is intelligent processing mode, but the intelligent customer service cannot process the target case, a manual intervention process will be automatically triggered, directly switching the intelligent processing mode to the manual processing mode, thereby improving claims processing efficiency.

[0099] In steps S1001 to S1005 of this embodiment, after the target insured initiates a claim request, human and / or intelligent customer service will be selected to process the claim based on the claim level of the claim case, realizing a human-machine collaborative claim processing mode. This not only improves the efficiency of claim processing and reduces labor costs, but also improves service quality and user satisfaction.

[0100] It's worth noting that intelligent customer service empowers insurance business processing, primarily by reducing costs by over 30% in claims processing and customer service. Life insurance policy underwriting rates reach 93%, and instant claims coverage reaches 56%. Vehicle damage assessment speed is increased by 4000 times, and anti-fraud systems significantly reduce losses. Therefore, by retaining human intervention while maintaining intelligent customer service, we not only reduce insurance business processing costs but also provide users with convenient services, enhancing the competitiveness of insurance products.

[0101] In some embodiments, please refer to Figure 12Step S1003 may include, but is not limited to, steps S1201 to S1203: Step S1201: Based on the insurance category, insured resource value, claim resource value, insured risk assessment data, and object characteristics, conduct a case complexity assessment to obtain case complexity assessment data; Step S1202: Based on the insurance category, insured resource value, claim resource value, insured risk assessment data, target characteristics and historical claims information, assess the difficulty of case processing to obtain case processing difficulty assessment data; Step S1203: Classify the claims based on the case complexity assessment data and the case processing difficulty assessment data to obtain the claim level.

[0102] In step S1201 of some embodiments, the insured resource value is also called the insured amount, and the compensation resource value is also called the compensation amount. The case complexity assessment is performed together with the insurance category, the insured amount, the compensation amount, the insured risk assessment data, and the object characteristics. The case complexity assessment is completed by the model to output the case processing difficulty assessment data.

[0103] In step S1202 of some embodiments, the difficulty of handling a case is assessed by combining the insurance category, the insured amount, the compensation amount, the insured risk assessment data, and the characteristics of the object, so as to determine the difficulty of handling the target case as the case handling difficulty assessment data.

[0104] In step S1203 of some embodiments, the claim level of the target case is determined by combining the case processing difficulty assessment data and the case complexity assessment data, which can accurately classify the claim level of the case.

[0105] For example, if the insurance type is auto insurance, and the incident is a single-vehicle accident, minor scratches, no injuries, occurred during off-peak hours, the payout is ≤5000, and the risk assessment data indicates low risk, then the claim level is determined to be simple. If the insurance type is auto insurance, and the incident involves water damage ("water bomb"), major collision, total vehicle loss ("bare vehicle"), or third-party liability disputes, and the payout is ≥10000, and the risk assessment data indicates high risk, then the claim level is determined to be high. If the insurance type is health / accident insurance, the payout is ≤5000, and the incident involves minor accidents, outpatient treatment, and complete documentation, then the claim level is determined to be moderate. If the insurance type is health / accident insurance, the payout is ≥10000, and the incident involves death, serious illness, disability, hospitalization, or third-party liability, then the claim level is determined to be high. Therefore, claims can be accurately categorized into different levels based on insurance type, target characteristics, and claim amount. Different claims processing models can be implemented according to different claim levels, improving the accuracy and efficiency of claims and enhancing the user experience.

[0106] In steps S1201 to S1203 of this embodiment, the complexity and processing difficulty of the target case are determined by combining the insurance category, the insured amount, the compensation amount, the insured risk assessment data, and the characteristics of the object. Then, the claim level is determined based on the complexity and processing difficulty of the case, so that the claim processing mode can be accurately selected according to the claim level, thereby improving the claim processing efficiency and the user experience during the claim processing.

[0107] Please refer to Figure 13 This application discloses the overall process of insurance business, from underwriting to risk warning to claims processing.

[0108] Step S1301: Connect multiple data sources, including external and internal data sources. Collect historical insurance data and insurance behavior data through internal data sources, and collect insurance-related data and exchange model parameters through external data sources. Integrate historical insurance data, insurance behavior data, insurance-related data and exchange model parameters into multi-source data. Step S1302: Clean and standardize the multi-source data, and then handle missing values ​​and unify the format of the standardized multi-source data. Step S1303: Extract data features from multi-source data and construct an object profile of the target insured object based on the data features; Step S1304: Collect the original risk assessment model and its original training data in advance; Step S1305: If the distribution difference of sample objects in the original training data is detected to be greater than a preset threshold, then the original training data is simulated to generate a new portrait, and the new portrait is added to the original training data to form target training data with balanced sample objects. Step S1306: Train the original risk assessment model using the target training data and training constraint parameters to obtain the target risk assessment model; Step S1307: Conduct an insurance risk assessment on the target insured object using the target risk assessment model and object profile to obtain insurance risk assessment data; Step S1308: Extract the key assessment features and assessment impact values ​​of the target insured object from the object profile through the insurance risk assessment data, visualize the key assessment features and assessment impact values ​​to form an assessment view, and underwrite the target insured object according to the assessment view.

[0109] Step S1309: Real-time insurance-related data and risk categories of the target insured object are collected in real time, and a selected risk prediction model is determined based on the risk categories. Claim risk prediction data is obtained by performing claim risk prediction through the selected risk prediction model and real-time insurance-related data. Step S310: Determine the warning level based on the claims risk prediction data and the preset warning threshold, and send the corresponding warning information according to the warning level; wherein, the preset warning thresholds are different for different warning levels; Step S1311: When the target insured initiates a claim request, first determine the complexity and processing difficulty of the target case in the claim request, and select the manual processing mode and / or intelligent processing mode according to the complexity and processing difficulty.

[0110] This embodiment primarily collects multi-source data to predict insurance risks, improving the accuracy of such predictions. It also utilizes multi-source data for risk warnings, enhancing their precision and timeliness. Furthermore, it provides visualized risk assessments, offering clear and reasonable explanations for underwriting, thus increasing customer trust. Finally, a human-machine collaborative decision-making framework optimizes resource allocation through reasonable division of labor, achieving the goal of cost reduction and efficiency improvement.

[0111] Please see Figure 14 This application also provides a risk warning device that can implement the above-mentioned risk warning method. The device includes: The profile acquisition module 1401 is used to acquire the profile of the target insured object. The insurance risk assessment module 1402 is used to conduct insurance risk assessment on the target insured object through a preset target risk assessment model and object profile, and obtain insurance risk assessment data. The underwriting module 1403 is used to perform underwriting processing on the target insured object based on the insured risk assessment data and the object profile, and obtain underwriting information. Data acquisition module 1404 is used to acquire real-time insurance-related data and risk categories of real-time insurance-related data of the target insured object based on underwriting information; The model screening module 1405 is used to select a risk prediction model from the preset candidate risk prediction models according to the risk category. The claims risk prediction module 1406 is used to predict claims risk by selecting a risk prediction model and real-time insurance-related data, and to obtain claims risk prediction data. The risk warning module 1407 is used to perform target warning operations on the target insured object based on the claim risk prediction data.

[0112] The specific implementation method of this risk warning device is basically the same as the specific implementation method of the above-mentioned risk warning method, and will not be described again here.

[0113] This application also provides a computer device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned risk warning method. This computer device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0114] Please see Figure 15 , Figure 15 The hardware structure of a computer device according to another embodiment is illustrated. The computer device includes: The processor 1501 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 1502 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 1502 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1502 and is called and executed by the processor 1501 using the risk warning method of the embodiments of this application. The input / output interface 1503 is used to implement information input and output; The communication interface 1504 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 1505 transmits information between various components of the device (e.g., processor 1501, memory 1502, input / output interface 1503, and communication interface 1504); The processor 1501, memory 1502, input / output interface 1503 and communication interface 1504 are connected to each other within the device via bus 1505.

[0115] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned risk warning method.

[0116] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0117] The risk warning method, apparatus, and computer equipment provided in this application, when assessing the risk of a target insured object, combine a target risk assessment model and an object profile to obtain insured risk assessment data. Then, based on the insured risk assessment data and the object profile, the underwriting of the target insured object is completed to obtain underwriting information, achieving automated underwriting and saving manpower. Next, a selected risk prediction model is selected from candidate risk prediction models based on the risk category. By selecting the risk prediction model and real-time insurance correlation data, the claim risk of the target insured object can be accurately predicted to obtain claim risk prediction data. Then, based on the claim risk prediction data, corresponding target warning operations are performed, providing early warning of claim risk. This not only reduces the actual claim risk level but also facilitates rapid loss assessment in subsequent claims, improving claims efficiency.

[0118] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0119] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0120] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0121] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0122] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0123] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0124] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0125] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0126] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0127] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0128] It should be noted that any software tools or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0129] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A risk warning method, characterized in that, The method includes: Obtain a profile of the target insured individual; The insurance risk assessment of the target insured object is carried out by using a preset target risk assessment model and the object profile to obtain insurance risk assessment data; Based on the insurance risk assessment data and the target profile, the underwriting process is carried out on the target insured object to obtain underwriting information; Based on the underwriting information, obtain the real-time insurance-related data of the target insured and the risk category of the real-time insurance-related data; Select a risk prediction model from the preset candidate risk prediction models based on the risk category; Claim risk prediction data is obtained by using the selected risk prediction model and the real-time insurance correlation data to predict claim risk. Based on the claims risk prediction data, a target early warning operation is performed on the target insured object.

2. The method according to claim 1, characterized in that, Before performing an insurance risk assessment on the target insured object using a preset target risk assessment model and the object profile to obtain insurance risk assessment data, the method further includes: Obtain the raw training data; wherein, the raw training data includes historical profiles of the sample objects; The distribution difference of the sample objects in the original training data is evaluated to obtain distribution difference evaluation data; wherein, the distribution difference evaluation data characterizes the degree of distribution difference of the sample objects; Based on the distribution difference assessment data, the original training data is subjected to sample balancing to obtain the target training data; The target risk assessment model is obtained by training the preset original risk assessment model based on the target training data and preset training constraint parameters.

3. The method according to claim 2, characterized in that, The step of performing sample balancing on the original training data based on the distribution difference assessment data to obtain the target training data includes: Based on the distribution difference assessment data, key difference features are extracted from the historical profile; wherein, the key difference features are object features whose distribution difference assessment data exceeds a preset threshold. Based on the distribution difference assessment data, a selected correction index is selected from the preset candidate correction indexes; Based on the key difference features and the selected correction index, training data simulation is performed to obtain candidate training data; The candidate training data and the original training data are mixed to obtain mixed training data; The authenticity of the mixed training data is evaluated to obtain authenticity evaluation data; The candidate training data is optimized based on the authenticity evaluation data to obtain the selected training data; The selected training data and the original training data are concatenated to obtain the target training data.

4. The method according to claim 2, characterized in that, The process of obtaining the object profile of the target insured individual includes: Obtain multi-source data of the target insured object; wherein, the multi-source data includes historical insurance data, insurance behavior data, insurance-related data, and exchange model parameters; The preset original portrait construction model is updated according to the exchange model parameters to obtain the target portrait construction model; The target insured object is profiled by constructing a profile using the target profile model, the historical insurance data, the insurance behavior data, and the insurance association data, thus obtaining the object profile.

5. The method according to any one of claims 1 to 4, characterized in that, The underwriting process for the target insured object based on the insured risk assessment data and the object profile, resulting in underwriting information, includes: Based on the insurance risk assessment data, the key assessment features and their impact values ​​are extracted from the object profile. The key evaluation features and the evaluation impact values ​​are visualized to obtain an evaluation view; The underwriting process for the target insured object is performed based on the assessment view to obtain underwriting information.

6. The method according to claim 1, characterized in that, The step of performing a target early warning operation on the target insured object based on the claim risk prediction data includes: The aforementioned claims risk prediction data is processed to obtain an early warning level; The target warning operation is selected from the preset candidate warning operations based on the warning level; Perform the target early warning operation on the target insured object.

7. The method according to claim 6, characterized in that, After the underwriting process for the target insured object is performed based on the insured risk assessment data and the object profile to obtain underwriting information, the method further includes: Receive the claim request from the target insured; Based on the claim request, obtain the insurance case characteristics of the target insured; Based on the characteristics of the insurance cases and the preset historical claims information, a claims level is obtained by classifying the claims into different levels. The selected claim processing mode is selected from the preset candidate claim processing modes according to the claim level; wherein the selected claim processing mode includes at least one of the following: manual processing mode and intelligent processing mode; The claims of the target insured are processed according to the selected claims processing mode.

8. The method according to claim 7, characterized in that, The characteristics of the insurance cases include: insurance category, insured resource value, compensation resource value, insured risk assessment data, and object characteristics; The process of classifying claims based on the characteristics of the insurance cases and pre-defined historical claims information to obtain claim levels includes: The case complexity is assessed based on the insurance category, the insured resource value, the compensation resource value, the insured risk assessment data, and the object characteristics to obtain case complexity assessment data; The difficulty of case processing is assessed based on the insurance category, the insured resource value, the compensation resource value, the insured risk assessment data, the object characteristics, and the historical claims information, resulting in case processing difficulty assessment data. The claim level is determined by classifying cases based on the case complexity assessment data and the case processing difficulty assessment data.

9. A risk warning device, characterized in that, The device includes: The profile acquisition module is used to acquire the profile of the target insured object; The insurance risk assessment module is used to assess the insurance risk of the target insured object through a preset target risk assessment model and the object profile, and obtain insurance risk assessment data. The underwriting module is used to perform underwriting processing on the target insured object based on the insured risk assessment data and the object profile, and obtain underwriting information; The data acquisition module is used to acquire real-time insurance-related data of the target insured object and the risk category of the real-time insurance-related data based on the underwriting information; The model filtering module is used to select a risk prediction model from a preset pool of candidate risk prediction models based on the risk category. The claims risk prediction module is used to predict claims risk using the selected risk prediction model and the real-time insurance-related data, and to obtain claims risk prediction data. The risk warning module is used to perform target warning operations on the target insured object based on the claim risk prediction data.

10. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the risk warning method according to any one of claims 1 to 8.

Images