CN-122023022-A - Method and device for identifying insurance attribute of vehicle insurance user

CN122023022ACN 122023022 ACN122023022 ACN 122023022ACN-122023022-A

Abstract

In order to identify the insurance attribute of the vehicle insurance user, the expected total sum of claims (total sum of risk) can be split into the number of risk-out times and the single average sum of claims and modeled respectively based on the insurance information of the vehicle insurance user. The first model is used for predicting the number of times of insurance based on the insurance information of the vehicle insurance user, the second model is used for predicting the single average claim settlement based on the insurance information of the vehicle insurance user, and then, the insurance attribute of the vehicle insurance user is determined based on the predicted number of times of insurance and the single average claim settlement, so that insurance business is developed for the vehicle insurance user, and personalized service of the vehicle insurance user is realized. Thus, the bottleneck of long tail distribution modeling can be broken through.

Inventors

XIONG ZHILI
SU YU
ZHANG PENG

Assignees

蚂蚁区块链科技(上海)有限公司

Dates

Publication Date: 20260512
Application Date: 20260105

Claims (10)

1. A method of identifying insurance attributes of a vehicle insurance user, the method comprising: Acquiring first insurance information corresponding to a first insurance user, wherein the first insurance information is determined based on at least one of vehicle information and insurance applicant information; predicting a first risk occurrence number of the first risk user in a policy period through a pre-trained first model based on the first insurance application information; predicting a single average claim amount for the first vehicle insurance user based on the first application information by a pre-trained second model; based on the first number of adventure and the single average claim amount, a first insurance attribute describing a risk of the first vehicle insurance user being at risk of the first vehicle insurance user is determined.
2. The method of claim 1, wherein the training of the first model comprises the steps of: Acquiring a first training sample, wherein the first training sample comprises sample application information and a label of the number of times of risk; sample application information in the first training sample is processed through a first model, and a predicted value of the risk number is obtained; determining a first model loss based on the predicted value of the number of risk excursions and the label of the number of risk excursions, wherein the first model loss is determined based on a negative binomial distribution satisfied by the number of risk excursions relative to the predicted value of the number of risk excursions; the undetermined parameters in the first model are updated in a direction to minimize loss of the first model.
3. The method of claim 2, wherein the negative binomial distribution satisfied by the predicted value of the risk occurrence times relative to the risk occurrence times is described by a dispersion coefficient and a probability quality function in the form of a mean value, wherein the mean value takes the predicted value of the risk occurrence times, and the dispersion coefficient is determined by: counting probability distribution under each risk occurrence number k based on each training sample; Fitting a probability distribution curve of the number of risk emerging k under negative binomial distribution based on a probability quality function described by the number of risk emerging k, the probability of risk emerging p and a parameter r describing the number of risk emerging not under the condition of occurrence of the kth time of risk emerging, so as to obtain the probability of risk emerging p and the parameter r; determining a dispersion coefficient based on the inverse of the parameter r 。
4. The method of claim 1, wherein a single training sample for training the second model comprises sample application information and a label for a sample single claim amount; in the process of training the second model, taking the logarithm of the predicted value of the second model as a dependent variable, taking the logarithm of the label of the sample single claim amount as a label, and carrying out linear regression fitting based on sample application information so as to update parameters in the second model.
5. The method of claim 1, wherein the determining a risk attribute describing a risk of the first risk user being at risk based on the first number of risk exits and the single average claim amount comprises: Determining a first risk-out total expected for the first risk user according to the product of the first risk-out times and the single average claim amount; and judging the first risk attribute based on the first risk total.
6. The method of claim 5, wherein the first risk attribute comprises a risk level of the first risk user, and wherein the determining the first risk attribute based on the first risk total comprises: Detecting a first range in which the first risk-out total falls in each of a plurality of preset monetary ranges corresponding to a plurality of preset grades, determining the risk grade of the first risk user as a first grade corresponding to the first range, or And mapping the first risk total amount to probability values on various preset levels, and determining the level with the maximum probability value as the risk level of the first risk user.
7. The method of claim 5, wherein the first risk-out attribute comprises the first premium amount; the determining the first risk attribute based on the first sum of claims includes: determining a ratio of the first claim sum to a predetermined reference value as a first maintenance coefficient; and determining the premium share of the first vehicle insurance user based on the product of the first premium coefficient and the reference premium.
8. An apparatus for identifying insurance attributes of a vehicle insurance user, the apparatus comprising: the system comprises an acquisition unit, a first insurance application unit and a second insurance application unit, wherein the acquisition unit is configured to acquire first insurance application information corresponding to a first vehicle insurance user, and the first insurance application information comprises at least one item of vehicle information and insurance application information; a first processing unit configured to predict a first number of times of risk of the first vehicle insurance user in a policy period by a first model trained in advance based on the first application information; A second processing unit configured to predict a single average claim amount for the first vehicle insurance user by a pre-trained second model based on the first application information; and a determination unit configured to determine a first insurance attribute describing a risk occurrence of the first insurance user based on the first number of times of risk occurrence and the single average claim amount.
9. A computer readable storage medium having stored thereon a computer program which, when executed in a computer, causes the computer to perform the method of any of claims 1-7.
10. A computing device comprising a processor and a memory having executable code stored therein which when executed by the processor implements the method of any of claims 1-7.

Description

Method and device for identifying insurance attribute of vehicle insurance user Technical Field One or more embodiments of the present disclosure relate to the field of computer technology, and in particular, to a method and apparatus for identifying insurance attributes of a vehicle insurance user. Background With the popularization of vehicles, vehicle insurance is a popular demand which is not separated in daily life. The insurance premium is typically the cost of an owner purchasing insurance for his vehicle at an insurance company. The car risk category may include, for example, but not limited to, traffic risk, xun risk, personal injury risk, and the like. The insurance premium is usually calculated by an insurance company according to information such as vehicle states (e.g., vehicle model number, vehicle value) and vehicle owner requirements (e.g., insurance risk, insurance amount). The pricing of the insurance premium is a key link of risk management in the insurance industry, and the key aim is to realize the accuracy and differentiation of the premium by scientifically evaluating the risk level of the vehicle and driving behaviors. Premium determination for vehicle insurance services relies on accurate predictions of the amount of overall claims expected by the applicant. Conventional predictive models can be divided into two categories, namely, predicting the total amount of claims and predicting the risk of claims. The scheme for predicting the risk of the claim is often to predict whether the risk is raised or not by adopting a classification model, users with medium and high risks cannot be distinguished, and the premium classification is rough. The traditional generalized linear model can be adopted for directly predicting the total claim amount, however, the total claim amount has the typical long tail characteristic that most vehicles cannot be in danger, and the number of times of danger occurrence of a few vehicles can be very large. Therefore, a few classes (heads) occupy most of the samples, while most classes (tails) have few samples, which are limited by complex distribution of claim data and limited sample size, long tail distribution results in difficult model fitting, sensitivity to hyper-parameters, and generally poor modeling effect. Therefore, how to predict the insurance attribute of the vehicle insurance user more accurately, especially the risk estimation of high-value insurance policy (such as high odds) is an important technical problem in the vehicle insurance business. Disclosure of Invention One or more embodiments of the present specification describe a method of identifying insurance attributes of a vehicle insurance user to address one or more of the problems mentioned in the background. According to a first aspect, a method for identifying insurance attributes of a vehicle insurance user is provided, and the method comprises the steps of obtaining first insurance information corresponding to the first vehicle insurance user, determining the first insurance information based on at least one of vehicle information and insurance applicant information, predicting first insurance times of the first vehicle insurance user in an insurance policy period through a first model trained in advance based on the first insurance information, predicting single average claim settlement amount for the first vehicle insurance user through a second model trained in advance based on the first insurance information, and determining the first insurance attribute describing the insurance risk of the first vehicle insurance user based on the first insurance times and the single average claim settlement amount. In one embodiment, training of the first model comprises the steps of obtaining a first training sample, processing sample application information in the first training sample through the first model to obtain a predicted value of the risk occurrence, determining first model loss based on the predicted value of the risk occurrence and the label of the risk occurrence, wherein the first model loss is determined based on a negative binomial distribution satisfied by the risk occurrence relative to the predicted value of the risk occurrence, and updating undetermined parameters in the first model towards a direction of minimizing the first model loss. In a further embodiment, the negative binomial distribution of the number of bets relative to the predicted value of the number of bets is described by a dispersion coefficient and a probability mass function in the form of a mean value, wherein the mean value takes the predicted value of the number of bets, the dispersion coefficient is determined by counting the probability distribution of each number of bets k based on each training sample, fitting a probability distribution curve of the number of bets k under the negative binomial distribution based on the probability mass function described by the number of bets k, the betting probabil