Insurance intelligent risk control system

By constructing graph structure data and risk score prediction models, the intelligent risk control system solves the problems of cumbersome review process and insufficient preset thresholds in the existing insurance risk control system, realizes more accurate risk assessment and early warning measures, and reduces the insurance company's Economic losses.

CN118941398AInactive Publication Date: 2024-11-12TI RONG INTELLIGENT TECH (SUZHOU) CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202410957453.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-17
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing insurance risk control system relies on manual review and preset threshold conditions, resulting in cumbersome review process, strong subjectivity and low efficiency, and it is difficult to effectively identify potential risks, such as the policyholder or relative who has a bad consumption record but does not exceed the preset amount. Scene.

Method used

设计一种保险智能风控系统,通过信息采集模块收集投保人和亲属的基本信息、财务信息和交易信息,构建图结构数据,并利用风险评分预测模型生成风险评分,根据评分划分区间设定预警措施,包括定期审查、增加审查频率和提高保证金或拒绝承保。

Benefits of technology

Improve the accuracy of risk score prediction for policyholders, reduce the economic losses of insurers, and more effectively identify and respond to potential risks through automated and data-driven methods.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118941398A_ABST
    Figure CN118941398A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of insurance risk control, and discloses an insurance intelligent risk control system, which comprises an information acquisition module used for acquiring basic information, financial information and transaction information of an insurance applicant and relatives of the insurance applicant; the information processing module is used for generating initial vectors corresponding to the insurance applicant and the relatives of the insurance applicant; the graph structure data generation module is used for generating graph structure data according to the initial vectors corresponding to the insurance applicant and the relatives of the insurance applicant; the risk score prediction module is used for inputting the graph structure data into a risk score prediction model, and an output value represents a risk score of the insurance applicant; the early warning measure generation module is used for generating an early warning measure according to the risk score of the insurance applicant; according to the invention, the basic information, financial information and transaction information of the insurance applicant and the relatives of the insurance applicant are constructed into the graph structure data, and information aggregation is carried out through the risk score prediction model, so that the prediction accuracy of the risk score of the insurance applicant is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of insurance risk control technology, and more specifically, to an insurance intelligent risk control system. Background Art

[0002] Existing insurance risk control systems usually determine whether there is a risk in insurance through manual review or preset threshold conditions; manual review checks the documents submitted by the insured (such as identity documents, financial status and health status, etc.) and background checks (such as historical consumption records, criminal records, etc.), and determines whether the insurance is risky based on manual experience. However, this method has a cumbersome review process, a highly subjective decision-making process, and low efficiency; preset threshold conditions include: insurance that exceeds the preset large insurance amount threshold requires additional review; the insured has multiple transactions or claims in a short period of time, which requires key review; however, although the preset threshold conditions for determining whether there is a risk in insurance solve the drawbacks of manual review to a certain extent, there may be situations where the insured circumvents the preset threshold conditions, for example: the insured amount does not exceed the insured amount threshold, but the insured or the insured's relatives have potential bad consumption records, then the insurance may also be risky.

[0003] Therefore, there is an urgent need for an insurance intelligent risk control system to solve the above problems. Summary of the invention

[0004] The present invention provides an insurance intelligent risk control system to solve the technical problems in the above-mentioned background technology.

[0005] The present invention provides an insurance intelligent risk control system, comprising:

[0006] An information collection module, which is used to collect basic information, financial information and transaction information of the policyholder and the policyholder's relatives;

[0007] Basic information includes: marital status, education, occupation and health status;

[0008] Financial information includes: annual income, total assets, total liabilities, credit score, number of bank accounts, number of credit cards, and average credit card limit;

[0009] Transaction information includes: historical insurance times, historical total claim amounts, historical claims times, average monthly deposit amounts, average monthly withdrawal amounts, and number of transaction location changes;

[0010] An information processing module, which is used to process the basic information, financial information and transaction information of the insured and the insured's relatives to generate an initial vector;

[0011] A graph structure data generation module, which is used to generate graph structure data according to the initial vectors corresponding to the insured and the insured's relatives;

[0012] Graph structure data includes: nodes, node features, and edges between nodes;

[0013] The node establishes data links with the policyholder and the policyholder’s relatives;

[0014] The node features are represented by the initial vectors corresponding to the policyholders and the policyholders’ relatives;

[0015] The edges between nodes include: the insured person and the insured person's relatives mapped by the node data have a direct family relationship; the insured person and the insured person's relatives mapped by the node data have transactions in the past year;

[0016] A risk score prediction module is used to input the graph structure data into the risk score prediction model, and the output value represents the risk score of the insured;

[0017] The risk score ranges from 1 to 10, with a higher risk score indicating a higher risk for the policyholder;

[0018] The early warning measure generation module is used to generate early warning measures according to the risk score of the insured.

[0019] Furthermore, marital status includes: married and single; educational background includes: postgraduate, undergraduate and below undergraduate; health status includes: healthy, sub-healthy and sick; among which marital status, educational background and health status are all represented by real number codes.

[0020] Furthermore, occupations are represented by occupational classification codes.

[0021] Furthermore, the basic information, financial information and transaction information of the insured and the insured's relatives are processed to generate an initial vector, including the following steps:

[0022] Step S201, directly fill in the missing values ​​in the basic information, financial information and transaction information with zero values;

[0023] Step S202, normalizing the basic information, financial information and transaction information by using the maximum and minimum normalization method;

[0024] Step S203, concatenating the normalized basic information, financial information and transaction information to generate an initial vector.

[0025] Further, the risk score prediction model includes: a first hidden layer, a second hidden layer, an extraction layer, and a first classifier;

[0026] The first hidden layer sequentially inputs the node feature of each node of the graph structure data, outputs the first updated feature, and represents the node feature of each node of the graph structure data through the first updated feature;

[0027] The second hidden layer inputs graph structure data and outputs graph update matrix;

[0028] A row vector of the graph update matrix corresponds to the second updated feature of a node;

[0029] The extraction layer is used to extract the second updated features of the nodes for establishing data mapping with the policyholder and input them into the first classifier;

[0030] The classification space of the first classifier represents the risk score of the policyholder.

[0031] Furthermore, the calculation formula of the first hidden layer is as follows:

[0032] F1=sigmoid[W2*(ReLU(W1*X+b1))+b2];

[0033] Wherein F1 represents the first updated feature output by the first hidden layer, X represents the node feature of the first hidden layer input, W1 and W1 represent the first weight parameter and the second weight parameter respectively, b1 and b2 represent the first bias parameter and the second bias parameter respectively, ReLU represents the ReLU activation function, and sigmoid represents the sigmoid activation function.

[0034] Furthermore, the calculation formula of the second hidden layer includes:

[0035]

[0036] Where 1≤i≤M, M represents the number of nodes in the graph structure data, S represents the graph update matrix output by the second hidden layer, and N i represents the set of nodes that have edges connected to the i-th node, represents the second updated feature of the i-th node, h i and h j Represents the first updated features of the i-th node and the j-th node, Nattion ij represents the attention score between the i-th node and the j-th node, D ij represents the number of edges between the i-th node and the j-th node, and Respectively represent the first weight parameter, the second weight parameter and the third weight parameter between the i-th node and the j-th node, W i represents the weight parameter corresponding to the i-th node, b ij represents the bias parameter between the i-th node and the j-th node, pile represents the stacking operation, T represents the transposition operation, and sigmoid represents the sigmoid activation function.

[0037] Furthermore, the risk score prediction model is pre-trained before training the risk score prediction model. During the pre-training process, the credit score of the policyholder is masked. The extraction layer is used to extract the second updated feature of the node that establishes data mapping with the policyholder and input it into the second classifier. The classification space of the second classifier represents the credit score of the policyholder.

[0038] Furthermore, the sample labels of the training samples used to train the risk score prediction model are obtained by taking the average value of the results simultaneously labeled by multiple experts.

[0039] Furthermore, the insured's risk score is divided into three intervals, including: low-risk interval, medium-risk interval and high-risk interval; the risk score of the low-risk interval is between 1 and 3, and the corresponding early warning measure is to regularly review the insured's financial status and credit record; the risk score of the medium-risk interval is between 4 and 7, and the corresponding early warning measure is to increase the review frequency and ask the insured to provide additional financial documents and credit certificates; the risk score of the high-risk interval is between 8 and 10, and the corresponding early warning measure is to immediately require the insured to provide a new guarantor and increase the deposit.

[0040] The beneficial effects of the present invention are as follows: the present invention improves the prediction accuracy of the risk score of the insured by constructing the basic information, financial information and transaction information of the insured and the insured's relatives into graph structure data, and aggregates the information through a risk score prediction model. In addition, different early warning measures are set according to the risk score of the insured, which can effectively reduce the economic losses of the insurance company. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic diagram of an insurance intelligent risk control system of the present invention;

[0042] Figure 2 It is a flow chart of the present invention for processing the basic information, financial information and transaction information of the insured and the insured's relatives to generate an initial vector.

[0043] In the figure: information collection module 101, information processing module 102, graph structure data generation module 103, risk score prediction module 104, early warning measure generation module 105. DETAILED DESCRIPTION

[0044] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0045] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in one or more embodiments of the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in one or more embodiments of the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0046] like Figure 1-2 As shown, an insurance intelligent risk control system includes:

[0047] An information collection module 101 is used to collect basic information, financial information and transaction information of the policyholder and the policyholder's relatives;

[0048] Basic information includes: marital status, education, occupation and health status;

[0049] Financial information includes: annual income, total assets, total liabilities, credit score, number of bank accounts, number of credit cards, and average credit card limit;

[0050] Transaction information includes: historical insurance times, historical total claim amounts, historical claims times, average monthly deposit amounts, average monthly withdrawal amounts, and number of transaction location changes;

[0051] An information processing module 102, which is used to process the basic information, financial information and transaction information of the insured and the insured's relatives to generate an initial vector;

[0052] A graph structure data generating module 103, which is used to generate graph structure data according to the initial vectors corresponding to the insured and the insured's relatives;

[0053] Graph structure data includes: nodes, node features, and edges between nodes;

[0054] The node establishes data links with the policyholder and the policyholder’s relatives;

[0055] The node features are represented by the initial vectors corresponding to the policyholders and the policyholders’ relatives;

[0056] The edges between nodes include: the insured person and the insured person's relatives mapped by the node data have a direct family relationship; the insured person and the insured person's relatives mapped by the node data have transactions in the past year;

[0057] The risk score prediction module 104 is used to input the graph structure data into the risk score prediction model, and the output value represents the risk score of the insured;

[0058] The risk score ranges from 1 to 10, with a higher risk score indicating a higher risk for the policyholder;

[0059] The early warning measure generation module 105 is used to generate early warning measures according to the risk score of the insured.

[0060] In one embodiment of the present invention, the basic information may also include: whether there is a chronic disease, the number of historical surgeries and the number of annual physical examinations, etc.

[0061] In one embodiment of the present invention, the financial information may also include: monthly expenditure amount, number of properties and number of vehicles in the past year, etc.

[0062] In one embodiment of the present invention, the transaction information may also include: online shopping frequency, participation in high-risk activities (such as gambling, extreme sports, etc.) and high-value commodity purchase records (such as luxury goods, works of art), etc.

[0063] In one embodiment of the present invention, marital status includes: married and single; educational background includes: postgraduate, undergraduate and below undergraduate; health status includes: healthy, sub-healthy and diseased; wherein marital status, educational background and health status are all represented by real number codes.

[0064] For example, the marital status of married is represented by the real number 1.1, the educational background of bachelor degree is represented by the real number 2.2, and the health condition of illness is represented by the real number 3.3.

[0065] In one embodiment of the present invention, occupations are represented by occupational classification codes, for example, 0112 represents a planting technician.

[0066] In one embodiment of the present invention, Figure 2 As shown, the basic information, financial information and transaction information of the insured and the insured's relatives are processed to generate an initial vector, including the following steps:

[0067] Step S201, directly fill in the missing values ​​in the basic information, financial information and transaction information with zero values;

[0068] Step S202, normalizing the basic information, financial information and transaction information by using the maximum and minimum normalization method;

[0069] Step S203, concatenating the normalized basic information, financial information and transaction information to generate an initial vector.

[0070] In one embodiment of the present invention, the risk score prediction model includes: a first hidden layer, a second hidden layer, an extraction layer, and a first classifier;

[0071] The first hidden layer sequentially inputs the node feature of each node of the graph structure data, outputs the first updated feature, and represents the node feature of each node of the graph structure data through the first updated feature;

[0072] The second hidden layer inputs graph structure data and outputs graph update matrix;

[0073] A row vector of the graph update matrix corresponds to the second updated feature of a node;

[0074] The extraction layer is used to extract the second updated features of the nodes for establishing data mapping with the policyholder and input them into the first classifier;

[0075] The classification space of the first classifier represents the risk score of the policyholder.

[0076] In one embodiment of the present invention, the calculation formula of the first hidden layer is as follows:

[0077] F1=sigmoid[W2*(ReLU(W1*X+b1))+b2];

[0078] Wherein F1 represents the first updated feature output by the first hidden layer, X represents the node feature of the first hidden layer input, W1 and W1 represent the first weight parameter and the second weight parameter respectively, b1 and b2 represent the first bias parameter and the second bias parameter respectively, ReLU represents the ReLU activation function, and sigmoid represents the sigmoid activation function.

[0079] In one embodiment of the present invention, the calculation formula of the second hidden layer includes:

[0080]

[0081] Where 1≤i≤M, M represents the number of nodes in the graph structure data, S represents the graph update matrix output by the second hidden layer, and N i represents the set of nodes that have edges connected to the i-th node, represents the second updated feature of the i-th node, h i and h j Represents the first updated features of the i-th node and the j-th node, Nattion ij represents the attention score between the i-th node and the j-th node, D ij represents the number of edges between the i-th node and the j-th node, and Respectively represent the first weight parameter, the second weight parameter and the third weight parameter between the i-th node and the j-th node, W i represents the weight parameter corresponding to the i-th node, b ij represents the bias parameter between the i-th node and the j-th node, pile represents the stacking operation, T represents the transposition operation, and sigmoid represents the sigmoid activation function.

[0082] It should be noted that the present invention aggregates the basic information, financial information and transaction information of the insured and the insured's relatives by constructing graph structure data, and comprehensively considers the relationship between the insured and the insured's relatives, thereby improving the accuracy of the prediction of the insured's risk score.

[0083] In one embodiment of the present invention, the risk score prediction model is pre-trained before training the risk score prediction model. During the pre-training process, the credit score of the policyholder is masked. The extraction layer is used to extract the second updated feature of the node that establishes data mapping with the policyholder and input it into the second classifier. The classification space of the second classifier represents the credit score of the policyholder.

[0084] In one embodiment of the present invention, the total liabilities may also be masked during the pre-training process, and the extraction layer is used to extract the second updated features of the nodes that establish data mapping with the policyholder and input them into the third classifier, and the classification space of the third classifier represents the total liabilities of the policyholder.

[0085] It should be noted that pre-training can improve the ability to extract features from information. Fine-tuning the operation after pre-training and then training can speed up the training of the model. Similarly, it can also prevent the model from overfitting to a certain extent and improve the generalization ability of the model.

[0086] In one embodiment of the present invention, the sample labels of the training samples used to train the risk score prediction model are obtained by taking the average value of the results simultaneously labeled by multiple experts.

[0087] It should be noted that during the training process of the risk score prediction model, the difference between the value output by the risk score prediction model in each iteration and the sample label of the training sample is specified as the loss function, and the weight parameters and bias parameters of the risk score prediction model are updated through back propagation and gradient descent algorithms. The training of the neural network model is a conventional technical means and will not be elaborated here.

[0088] In one embodiment of the present invention, the risk score of the insured is divided into three intervals, including: a low-risk interval, a medium-risk interval and a high-risk interval; the risk score of the low-risk interval is between 1 and 3, and the corresponding early warning measure is to regularly review the financial status and credit record of the insured; the risk score of the medium-risk interval is between 4 and 7, and the corresponding early warning measure is to increase the review frequency and ask the insured to provide additional financial documents and credit certificates; the risk score of the high-risk interval is between 8 and 10, and the corresponding early warning measure is to immediately require the insured to re-provide a guarantor and increase the deposit, or re-evaluate the insurance terms, or even refuse to continue to underwrite.

[0089] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. An insurance intelligent risk control system, It is characterized in that include: An information collection module, which is used to collect basic information, financial information and transaction information of the policyholder and the policyholder's relatives; Basic information includes: marital status, education, occupation and health status; Financial information includes: annual income, total assets, total liabilities, credit score, number of bank accounts, number of credit cards, and average credit card limit; Transaction information includes: historical insurance times, historical total claim amounts, historical claims times, average monthly deposit amounts, average monthly withdrawal amounts, and number of transaction location changes; An information processing module, which is used to process the basic information, financial information and transaction information of the insured and the insured's relatives to generate an initial vector; A graph structure data generation module, which is used to generate graph structure data according to the initial vectors corresponding to the insured and the insured's relatives; Graph structure data includes: nodes, node features, and edges between nodes; The node establishes data links with the policyholder and the policyholder’s relatives; The node features are represented by the initial vectors corresponding to the policyholders and the policyholders’ relatives; The edges between nodes include: the insured person and the insured person's relatives mapped by the node data have a direct family relationship; the insured person and the insured person's relatives mapped by the node data have transactions in the past year; A risk score prediction module is used to input the graph structure data into the risk score prediction model, and the output value represents the risk score of the insured; The risk score ranges from 1 to 10, with a higher risk score indicating a higher risk for the policyholder; The early warning measure generation module is used to generate early warning measures according to the risk score of the insured.

2. An insurance intelligent risk control system according to claim 1, It is characterized in that Marital status includes: married and single; educational background includes: postgraduate, undergraduate and below undergraduate; health status includes: healthy, sub-healthy and sick; marital status, educational background and health status are all represented by real number codes.

3. An insurance intelligent risk control system according to claim 1, It is characterized in that Occupations are represented by occupational classification codes.

4. An insurance intelligent risk control system according to claim 1, It is characterized in that The basic information, financial information and transaction information of the insured and the insured's relatives are processed to generate an initial vector, including the following steps: Step S201, directly fill in the missing values ​​in the basic information, financial information and transaction information with zero values; Step S202, normalizing the basic information, financial information and transaction information by using the maximum and minimum normalization method; Step S203, concatenating the normalized basic information, financial information and transaction information to generate an initial vector.

5. An insurance intelligent risk control system according to claim 1, It is characterized in that The risk score prediction model includes: a first hidden layer, a second hidden layer, an extraction layer and a first classifier; The first hidden layer sequentially inputs the node feature of each node of the graph structure data, outputs the first updated feature, and represents the node feature of each node of the graph structure data through the first updated feature; The second hidden layer inputs graph structure data and outputs graph update matrix; A row vector of the graph update matrix corresponds to the second updated feature of a node; The extraction layer is used to extract the second updated features of the nodes for establishing data mapping with the policyholder and input them into the first classifier; The classification space of the first classifier represents the risk score of the policyholder.

6. An insurance intelligent risk control system according to claim 5, It is characterized in that The calculation formula for the first hidden layer is as follows: F 1 =sigmoid[W 2 *(ReLU(W 1 *X+b 1 ))+b 2 ]; Among them, F 1 represents the first updated feature of the first hidden layer output, X represents the node feature of the first hidden layer input, and W 1 and W 1 Represent the first weight parameter and the second weight parameter, b 1 and b 2 Represent the first bias parameter and the second bias parameter respectively, ReLU represents the ReLU activation function, and sigmoid represents the sigmoid activation function.

7. An insurance intelligent risk control system according to claim 5, It is characterized in that The calculation formula for the second hidden layer includes: Where 1≤i≤M, M represents the number of nodes in the graph structure data, S represents the graph update matrix output by the second hidden layer, and N i represents the set of nodes that have edges connected to the i-th node, represents the second updated feature of the i-th node, h i and h j Represents the first updated features of the i-th node and the j-th node, Nattion ij represents the attention score between the i-th node and the j-th node, D ij represents the number of edges between the i-th node and the j-th node, and Respectively represent the first weight parameter, the second weight parameter and the third weight parameter between the i-th node and the j-th node, W i represents the weight parameter corresponding to the i-th node, b ij represents the bias parameter between the i-th node and the j-th node, pile represents the stacking operation, T represents the transposition operation, and sigmoid represents the sigmoid activation function.

8. An insurance intelligent risk control system according to claim 5, It is characterized in that The risk score prediction model is pre-trained before training. During the pre-training process, the credit score of the policyholder is masked. The extraction layer is used to extract the second updated feature of the node that establishes data mapping with the policyholder and input it into the second classifier. The classification space of the second classifier represents the credit score of the policyholder.

9. An insurance intelligent risk control system according to claim 1, It is characterized in that The sample labels of the training samples used to train the risk score prediction model are obtained by taking the average of the results simultaneously labeled by multiple experts.

10. An insurance intelligent risk control system according to claim 1, It is characterized in that The insured's risk score is divided into three intervals, including: low-risk interval, medium-risk interval and high-risk interval; the risk score of the low-risk interval is between 1 and 3, and the corresponding early warning measure is to regularly review the insured's financial status and credit record; the risk score of the medium-risk interval is between 4 and 7, and the corresponding early warning measure is to increase the review frequency and ask the insured to provide additional financial documents and credit certificates; the risk score of the high-risk interval is between 8 and 10, and the corresponding early warning measure is to immediately require the insured to provide a new guarantor and increase the deposit.

Citation Information

Patent Citations

  • Medical insurance underwriting risk control method and device and storage medium

    CN111179102A

  • Underwriting risk control monitoring method, device and equipment and storage medium

    CN112330474A

  • Account risk model training method and risk user group determination method

    CN114187112A

  • Training method of transaction risk prediction model and transaction risk prediction method

    CN117522410A