Electronic insurance policy intelligent identification method

Abstract

The invention provides an electronic insurance policy intelligent identification method, which comprises the following steps of obtaining comprehensive data information of an electronic insurance policy, and performing preprocessing to obtain first data in a set format; constructing a multi-party ontology and an attribute unit of the multi-party ontology by using the first data, and carrying out relation representation among the multi-party ontology through a trained graph attention network model to obtain a group of ontology with attention weight values and a fraud estimated value; extracting a driver body and an insurance salesman body, and performing potential fraud evaluation by adopting a neural network model according to previous historical data to obtain a reference estimated value; and performing linear modeling on the fraud estimated value and a reference estimated value to obtain a comprehensive estimated value, and if the comprehensive estimated value is greater than a set threshold value, determining that the electronic insurance policy is a fraud type electronic insurance policy. According to the method, a graph attention neural network is adopted to represent the relationship among multiple parties of ontologies to obtain a fraud estimated value, and the behavior habits of a subject person are taken as reference factors of estimation to comprehensively evaluate the potential fraud of the electronic insurance policy.

Description

Technical Field

[0001] The present application relates to the field of smart financial services technology, and in particular to a method for intelligently identifying electronic insurance policies. Background Art

[0002] With the rapid development of the financial services market, the insurance industry's scope and the scale of insurance institutions are also gradually expanding. Auto insurance is a key area of focus for the industry. Premium income has been declining for several consecutive years, while claims costs have continued to rise. Insurance fraud cases exist, with some criminals employing covert tactics, violating insurance regulations, and defrauding insurance companies of benefits by fabricating insurance labels, insurance incidents, or even fabricating insurance incidents. Accurate detection and identification of fraudulent activity in auto insurance claims is essential. Current solutions include binary classification detection models for various fraud behaviors and neural network-based modeling. However, there is room for improvement in the efficiency and accuracy of data correlations. Furthermore, effective relational modeling is still required to assess the risk of potential fraud. Summary of the Invention

[0003] In view of the above problems, the present invention provides an intelligent identification method for electronic insurance policies, which uses a graph attention neural network to characterize the relationship between multiple ontologies to obtain a fraud estimate, and uses the behavioral habits of the subject as a reference factor for the valuation to comprehensively evaluate the potential fraud of electronic insurance policies.

[0004] In a first aspect, the present invention provides an electronic insurance policy intelligent recognition method, which obtains comprehensive data information of the electronic insurance policy and pre-processes it to obtain first data in a set format; wherein, the multi-party ontology includes an insurance policy ontology, a driver ontology, an insurance salesperson ontology, a case ontology, a repair information ontology, and a vehicle ontology;

[0005] Constructing multiple ontologies and their attribute units based on the first data, and representing the relationships between the multiple ontologies using the trained graph attention network model to obtain a set of ontologies with attention weights and fraud estimates;

[0006] The driver entity and insurance salesperson entity are extracted and evaluated for potential fraud using a neural network model based on historical data to obtain a reference valuation.

[0007] The fraud estimate value and the reference estimate value are linearly modeled to obtain a comprehensive estimate value. If the comprehensive estimate value is greater than a set threshold value, it is determined to be a fraudulent electronic insurance policy.

[0008] Furthermore, the graph attention network model obtained after the training has the following specific training steps:

[0009] Collect the entity node set ε={E1,E2,…,E n}, extract the M-dimensional feature value h of each entity node from the attribute unit i ;

[0010] Construct a multi-layer graph neural network model, and the attention between each entity node is expressed as The attention weight is expressed as The final output feature of each entity node is expressed as

[0011] The collected K samples are used to train a multi-layer graph neural network model to obtain a converged graph attention network model.

[0012] Furthermore, the driver's historical insurance records are collected and feature engineering is performed to obtain the first data to be processed;

[0013] Construct a Conv-LSTM time series model and train it to obtain a converged Conv-LSTM model;

[0014] The data to be processed is fed into the Conv-LSTM model for prediction to obtain a reference estimate of potential fraud.

[0015] Furthermore, the neural network model is used to evaluate the potential fraud of insurance sales personnel, specifically including:

[0016] Collect the insurance salesperson's previously issued insurance records, perform feature engineering, and obtain the second data to be processed;

[0017] Construct a Conv-LSTM time series model and train it to obtain a converged Conv-LSTM model;

[0018] The second data to be processed is introduced into the Conv-LSTM model for prediction to obtain a second reference estimate of potential fraud.

[0019] Furthermore, the comprehensive valuation includes the estimated value obtained by the graph neural network model and the reference valuation one and reference valuation two obtained by the potential fraud assessment of the driver entity and the insurance salesperson entity, and the estimated value, reference valuation one and reference valuation two are linearly modeled, expressed as comprehensive valuation = α·estimated value + β·parameter valuation one + γ·parameter valuation two; wherein α is the proportion of positive sample data to the total sample data, β is the proportion of positive sample data of the driver entity to the total sample data, and γ is the proportion of positive sample data of the insurance salesperson entity to the total sample data.

[0020] In a second aspect, a computer-readable storage medium stores instructions, and when the instructions are executed by a processor, the electronic insurance policy intelligent identification method as described in any one of the above items is implemented.

[0021] The above embodiment has the following advantages or beneficial effects:

[0022] (1) Consider the impact of multiple factors on model valuation in more dimensions;

[0023] (2) Improve the recognition accuracy of fraud models for electronic insurance policies.

[0024] Other features and advantages of the present application will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present application. The purposes and other advantages of the present application can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0026] Figure 1 This is a flow chart of an electronic insurance policy intelligent identification method provided by the present invention. DETAILED DESCRIPTION

[0027] To make the objectives, technical solutions, and advantages of this application more clearly understood, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0028] Example 1

[0029] like Figure 1 The flowchart of a method for intelligent identification of electronic insurance policies is shown in FIG. The specific steps are as follows:

[0030] Step S01: Obtain comprehensive data information of the electronic insurance policy and pre-process it to obtain first data in a set format; wherein the multi-party ontology includes the insurance policy ontology, the driver ontology, the insurance salesperson ontology, the case ontology, the repair information ontology, and the vehicle ontology.

[0031] The policy body contains key information such as the insured, insurance type, insurance amount, premium, and insurance period. The driver body contains basic information and insurance history data.

[0032] Insurance salesperson's ontological information includes communication records with insurance companies, including emails, chat logs, phone records, contracts, insurance terms, exemption clauses, etc.

[0033] Case information includes photos of the accident scene, diagnostic records provided by medical institutions, medical imaging and other data;

[0034] The repair information includes data such as the vehicle damage repair bill of the insured vehicle;

[0035] The vehicle body includes data such as basic information of the vehicle.

[0036] The multi-party ontology information mentioned above is feature-vectorized and extracted and stored in the form of a dictionary; for example: data D = {policy ontology, driver ontology, insurance salesperson ontology, case ontology, repair information ontology, vehicle ontology}; the specific example is: {{"insured person's gender": 1, "insurance type": 2, "insurance amount": 9, "premium": 6, "insurance period": 2}; {"driver's gender": 2, "driver's status": 2, "insured years": 4; {"historical data 1": 0, "insured amount": 7, "insurance willingness": 1, "whether to claim this year": Y, "whether there is insurance fraud": Y}; {"historical data 2": 1, ...}}}; where, for example: 1 represents male, 0 represents female; 9 represents the highest insurance amount.

[0037] The collected data is characterized and saved in the form of a dictionary.

[0038] Step S02: construct a multi-party ontology and its attribute units using the first data, and characterize the relationship between the multi-party ontologies through the trained graph attention network model to obtain a set of ontologies with attention weight values and fraud estimation values.

[0039] The graph attention network model, the specific training steps are as follows:

[0040] Collect the entity node set ε={E1,E2,…,E n}, extract the M-dimensional feature value h of each entity node from the attribute unit i ;

[0041] Construct a multi-layer graph neural network model, and the attention between each entity node is expressed as The attention weight is expressed as The final output feature of each entity node is expressed as

[0042] The collected K samples are used to train a multi-layer graph neural network model to obtain a converged graph attention network model.

[0043] Entity node e={E,a}, where {E1,E2,…,E n} is a set of n ontology types, a is a specific attribute of a single entity. The relationship between entities is a triple {e1, r, e2}, where {R1, R2, ..., R m} is a set of m entities and relationships between them. The key subgraph g = {E, R} is selected through the graph attention mechanism, where E is a set of single entity nodes and R is the relationship between entity nodes in the entity node set.

[0044] Step S03: Extract the driver entity and insurance salesperson entity and use a neural network model to conduct potential fraud assessment based on historical data to obtain a reference valuation;

[0045] The neural network model is used to evaluate the potential fraud of the driver, specifically including:

[0046] Collect the driver's historical insurance records, perform feature engineering, and obtain the first data to be processed;

[0047] Construct a Conv-LSTM time series model and train it to obtain a converged Conv-LSTM model;

[0048] The data to be processed is fed into the Conv-LSTM model for prediction to obtain a reference estimate of potential fraud.

[0049] In practice, the system extracts features from the driver's previous insurance records to estimate whether there are any anomalies in the current insurance policy and predict potential fraud. Using the driver's behavioral habits as a reference, it also assists in fraud prediction for current electronic insurance policies.

[0050] Then, conduct a fraud assessment on the insurance salesperson's main behavior. The specific operations are as follows:

[0051] Collect the insurance salesperson's previously issued insurance records, perform feature engineering, and obtain the second data to be processed;

[0052] Construct a Conv-LSTM time series model and train it to obtain a converged Conv-LSTM model;

[0053] The second data to be processed is introduced into the Conv-LSTM model for prediction to obtain a second reference estimate of potential fraud.

[0054] In the specific implementation, the historical business records of insurance sales personnel are collected. The data contains some fraud samples. Features are extracted through the Conv convolutional network, and the LSTM time series model is used to infer and predict the correlation between the features. It detects whether there are outliers in the current business of the insurance sales personnel and predicts the risk of fraud.

[0055] Step S04: Linearly modeling is performed on the fraud estimate value and the reference estimate value to obtain a comprehensive estimate value. If the comprehensive estimate value is greater than a set threshold value, it is determined to be a fraudulent electronic insurance policy.

[0056] The above makes a comprehensive fraud analysis and assessment of current electronic insurance policies by evaluating the overall multi-party ontology relationship and taking the behavioral inertia of the subjects in the ontology as a reference.

[0057] The comprehensive valuation includes the estimated value obtained by the graph neural network model and the reference valuation one and reference valuation two obtained by the potential fraud assessment of the driver entity and the insurance salesperson entity, and linear modeling is performed on the estimated value, reference valuation one and reference valuation two, which is expressed as comprehensive valuation = α·estimated value + β·parameter valuation one + γ·parameter valuation two; wherein α is the proportion of positive sample data to the total sample data, β is the proportion of positive sample data of the driver entity to the total sample data, and γ is the proportion of positive sample data of the insurance salesperson entity to the total sample data.

[0058] Example 2

[0059] A computer-readable storage medium having instructions stored thereon, wherein the instructions, when executed by a processor, implement the electronic insurance policy intelligent identification method as described in any one of the above items.

[0060] Through the above scheme, the following advantages or beneficial effects are achieved:

[0061] (1) Consider the impact of multiple factors on model valuation in more dimensions;

[0062] (2) Improve the recognition accuracy of fraud models for electronic insurance policies.

[0063] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for intelligent identification of electronic insurance policies, characterized in that: These include: Obtaining comprehensive data information of the electronic insurance policy and preprocessing it to obtain first data in a set format; wherein the multi-party ontology includes the insurance policy ontology, the driver ontology, the insurance salesperson ontology, the case ontology, the repair information ontology, and the vehicle ontology; Constructing multiple ontologies and their attribute units based on the first data, and representing the relationships between the multiple ontologies using the trained graph attention network model to obtain a set of ontologies with attention weights and fraud estimates; The driver entity and insurance salesperson entity are extracted and evaluated for potential fraud using a neural network model based on historical data to obtain a reference valuation. The fraud estimate value and the reference estimate value are linearly modeled to obtain a comprehensive estimate value. If the comprehensive estimate value is greater than a set threshold value, it is determined to be a fraudulent electronic insurance policy.

2. The electronic insurance policy intelligent identification method according to claim 1, characterized in that: The graph attention network model obtained after the training, the specific training steps are as follows: Collect the entity node set ε={E1,E2,…,E n }, extract the M-dimensional feature value h of each entity node from the attribute unit i ; Construct a multi-layer graph neural network model, and the attention between each entity node is expressed as The attention weight is expressed as The final output feature of each entity node is expressed as The collected K samples are used to train a multi-layer graph neural network model to obtain a converged graph attention network model.

3. The method for intelligently identifying electronic insurance policies according to claim 1, wherein: The neural network model is used to evaluate the potential fraud of the driver, specifically including: Collect the driver's historical insurance records, perform feature engineering, and obtain the first data to be processed; Construct a Conv-LSTM time series model and train it to obtain a converged Conv-LSTM model; The data to be processed is fed into the Conv-LSTM model for prediction to obtain a reference estimate of potential fraud.

4. The method for intelligently identifying electronic insurance policies according to claim 1, wherein: The use of a neural network model to assess potential fraud of insurance sales personnel specifically includes: Collect the insurance salesperson's previously issued insurance records, perform feature engineering, and obtain the second data to be processed; Construct a Conv-LSTM time series model and train it to obtain a converged Conv-LSTM model; The second data to be processed is introduced into the Conv-LSTM model for prediction to obtain a second reference estimate of potential fraud.

5. The electronic insurance policy intelligent identification method according to claim 1, characterized in that: The comprehensive valuation includes the estimated value obtained by the graph neural network model and the reference valuation one and reference valuation two obtained by the potential fraud assessment of the driver entity and the insurance salesperson entity, and linear modeling is performed on the estimated value, reference valuation one and reference valuation two, which is expressed as comprehensive valuation = α·estimated value + β·parameter valuation one + γ·parameter valuation two; wherein α is the proportion of positive sample data to the total sample data, β is the proportion of positive sample data of the driver entity to the total sample data, and γ is the proportion of positive sample data of the insurance salesperson entity to the total sample data.

6. A computer-readable storage medium having instructions stored thereon, wherein the instructions, when executed by a processor, implement the electronic insurance policy intelligent identification method according to any one of claims 1 to 5.

Images