Vehicle insurance fraud behavior identification method and device, electronic equipment, medium and product

Abstract

The embodiment of the invention provides a vehicle insurance fraud behavior identification method and device, electronic equipment, a medium and a product, and belongs to the technical field of artificial intelligence. The method comprises the following steps: inputting information of each car insurance claim settlement case into a neural network model to obtain a selection probability of each claim settlement action of each case; and determining a target claim settlement action according to the selection probability. And determining reward values of the target claim settlement action according to the target claim settlement action and the reference claim settlement action, and determining a sum value of the reward values as an accumulated reward of the model. A second desired gradient is determined from the cumulative reward and the first desired gradient. And finally, iteratively updating the model parameters according to the second expected gradient to obtain a vehicle insurance fraud behavior identification model. Therefore, the accumulated reward of the model is obtained by inputting the information of each car insurance claim settlement case into the neural network model, and the model parameters are updated according to the expected gradient of the model parameters through reinforcement learning, so that the anti-fraud strategy of the model can be optimized, and the accuracy of car insurance fraud behavior recognition is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, electronic equipment, medium and product for identifying auto insurance fraud. Background Art

[0002] At present, traditional methods for identifying auto insurance fraud mainly focus on manually annotated structured data, or use traditional binary classification models to detect auto insurance fraud.

[0003] A binary classification model is a machine learning model that is used to classify data into two categories. The goal of a binary classification model is to predict the output label, usually labeled 0 and 1 or negative and positive, based on the input features.

[0004] However, when the number of samples of different categories in the data set varies greatly, the binary classification model will tend to predict the sample category with the majority number, resulting in low accuracy of the prediction results.

[0005] Therefore, how to improve the accuracy of identifying auto insurance fraud has become a technical problem that needs to be solved urgently. Summary of the invention

[0006] The main purpose of the embodiments of the present application is to propose a method, device, electronic device, medium and product for identifying auto insurance fraud, aiming to improve the accuracy of identifying auto insurance fraud.

[0007] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a method for identifying auto insurance fraud, the method comprising:

[0008] Input the information of each auto insurance claim case into the neural network model respectively, and obtain the selection probability of each claim action corresponding to each auto insurance claim case;

[0009] Determine the target claim action corresponding to each auto insurance claim case according to the selection probability of each claim action corresponding to each auto insurance claim case;

[0010] Determine the reward value of the target claim action according to the target claim action and the reference claim action corresponding to the target claim action;

[0011] The sum of the reward values ​​of the target claim actions corresponding to each auto insurance claim case is determined as the cumulative reward of the neural network model;

[0012] Determine a second expected gradient according to the accumulated reward and the first expected gradient, the first expected gradient refers to the expected gradient of each target claim settlement action with respect to the parameters of the neural network model, and the second expected gradient refers to the expected gradient of the expected reward of the neural network model with respect to the parameters of the neural network model;

[0013] The parameters of the neural network model are updated according to the second expected gradient until the parameters of the neural network model meet the iteration termination condition, thereby obtaining a motor vehicle insurance fraud behavior identification model, which is used to identify motor vehicle insurance fraud behavior.

[0014] In some embodiments, determining the second expected gradient according to the accumulated reward and the first expected gradient includes:

[0015] Multiply the cumulative reward by the expected gradient of each target claim action to obtain the expected gradient of the parameters of the neural network model corresponding to each auto insurance claim case;

[0016] The expected gradients of the parameters of the neural network model corresponding to each motor insurance claim case are summed to obtain a second expected gradient.

[0017] In some embodiments, before determining the second expected gradient according to the accumulated reward and the first expected gradient, the identifying method includes:

[0018] Determine the log probability of the selection probability of the target claim action;

[0019] The gradient of the logarithmic probability is determined based on the chain rule to obtain the first expected gradient.

[0020] In some embodiments, updating the initial parameters of the neural network model according to the second expected gradient includes:

[0021] Determine the product of the second expected gradient and the learning rate of the neural network model;

[0022] The sum of the product and the parameters of the neural network model is determined as the updated parameters of the neural network model.

[0023] In some embodiments, updating the parameters of the neural network model according to the second expected gradient until the parameters of the neural network model meet the iteration termination condition to obtain a vehicle insurance fraud behavior recognition model includes:

[0024] Determine the current cumulative return of the neural network model based on the reward value of the target claim action corresponding to each auto insurance claim case and the discount factor of the neural network model;

[0025] When the current cumulative return is greater than the cumulative return before the update, reduce the learning rate;

[0026] When the current cumulative reward is less than the cumulative reward before the update, increase the learning rate.

[0027] In some embodiments, determining a reward value of a target claim settlement action according to the target claim settlement action and a reference claim settlement action corresponding to the target claim settlement action includes:

[0028] When the target claim action is the same as the reference claim action, the reward value is determined to be a positive number;

[0029] In the case where the target claim action is different from the reference claim action, the reward value is determined to be a negative number.

[0030] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application provides a device for identifying auto insurance fraud, the device comprising:

[0031] An input module, used to input the information of each auto insurance claim case into the neural network model respectively, and obtain the selection probability of each claim action corresponding to each auto insurance claim case;

[0032] The first determination module is used to determine the target claim action corresponding to each motor insurance claim case according to the selection probability of each claim action corresponding to each motor insurance claim case;

[0033] A second determination module is used to determine the reward value of the target claim settlement action according to the target claim settlement action and the reference claim settlement action corresponding to the target claim settlement action;

[0034] A third determination module is used to determine the sum of the reward values ​​of the target claim settlement actions corresponding to each motor insurance claim settlement case as the cumulative reward of the neural network model;

[0035] A fourth determination module is used to determine a second expected gradient according to the accumulated reward and the first expected gradient, wherein the first expected gradient refers to the expected gradient of each target claim settlement action with respect to the parameters of the neural network model, and the second expected gradient refers to the expected gradient of the expected reward of the neural network model with respect to the parameters of the neural network model;

[0036] An updating module is used to update the parameters of the neural network model according to the second expected gradient until the parameters of the neural network model meet the iteration termination condition, thereby obtaining a motor vehicle insurance fraud behavior identification model, which is used to identify motor vehicle insurance fraud behavior.

[0037] To achieve the above objectives, a third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method described in the first aspect is implemented.

[0038] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.

[0039] To achieve the above objectives, the fifth aspect of an embodiment of the present application proposes a computer program product. When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device executes the method for identifying auto insurance fraud as described above.

[0040] The present application proposes a method, device, electronic device, medium and product for identifying auto insurance fraud. The method inputs the information of each auto insurance claim case into a neural network model to obtain the selection probability of each claim action corresponding to each auto insurance claim case; and determines the target claim action corresponding to each auto insurance claim case based on the selection probability of each claim action corresponding to each auto insurance claim case. Then, the reward value of the target claim action is determined based on the target claim action and the reference claim action corresponding to the target claim action, and the sum of the reward values ​​of the target claim actions corresponding to each auto insurance claim case is determined as the cumulative reward of the neural network model. The second expected gradient is determined based on the cumulative reward and the first expected gradient. Finally, the parameters of the neural network model are updated based on the second expected gradient until the parameters of the neural network model meet the iteration termination condition, thereby obtaining an auto insurance fraud identification model, which is used to identify auto insurance fraud. In this way, by inputting the information of each auto insurance claim case into the neural network model, the cumulative reward of the neural network model is obtained, and through reinforcement learning, the parameters of the neural network model are updated according to the expected gradient of the parameters of the neural network model, so that the neural network model can continuously update the parameters, optimize the anti-fraud strategy, and improve the accuracy of identifying auto insurance fraud behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is one of the flow charts of the method for identifying auto insurance fraud provided in the embodiment of the present application;

[0042] Figure 2 yes Figure 1 Flow chart of step S105 in FIG.

[0043] Figure 3 yes Figure 1 Flow chart of step S106 in FIG.

[0044] Figure 4 This is the second flow chart of the method for identifying auto insurance fraud provided in the embodiment of the present application;

[0045] Figure 5 It is a schematic diagram of the structure of a device for identifying auto insurance fraud provided in an embodiment of the present application;

[0046] Figure 6 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0048] It should be noted that, although the functional modules are divided in the device schematic diagram and the 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 above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0050] First, some nouns involved in this application are analyzed:

[0051] Artificial intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0052] Information Extraction: A text processing technology that extracts specified types of entity, relationship, event and other factual information from natural language text and forms structured data output. Information extraction is a technology that extracts specific information from text data. Text data is composed of some specific units, such as sentences, paragraphs, and chapters. Text information is composed of some small specific units, such as characters, words, phrases, sentences, paragraphs, or a combination of these specific units. Extracting noun phrases, names, place names, etc. from text data are all text information extraction. Of course, the information extracted by text information extraction technology can be various types of information.

[0053] Based on this, the embodiments of the present application provide a method, device, electronic device, medium and product for identifying auto insurance fraud, aiming to improve the accuracy of identifying auto insurance fraud.

[0054] The method, device, electronic device, medium and product for identifying auto insurance fraud provided in the embodiments of the present application are specifically illustrated through the following embodiments. First, the identification method in the embodiments of the present application is described.

[0055] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0056] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0057] The method for identifying auto insurance fraud provided in the embodiment of the present application relates to the field of artificial intelligence technology. The method for identifying auto insurance fraud provided in the embodiment of the present application can be applied to a terminal, can be applied to a server, or can be software running in a terminal or a server. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server can be configured as an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the method for identifying auto insurance fraud, etc., but is not limited to the above forms.

[0058] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present 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. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0059] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.

[0060] Figure 1 This is one of the flow charts of the method for identifying auto insurance fraud provided in the embodiment of the present application. Figure 1 The method may include but is not limited to steps S101 to S106.

[0061] Step S101, inputting the information of each motor vehicle insurance claim case into the neural network model respectively, and obtaining the selection probability of each claim action corresponding to each motor vehicle insurance claim case.

[0062] In an embodiment of the present application, information on a car insurance claim case may include, but is not limited to, the insurer's personal information, vehicle information, policy information, claim history, accident type, and accident severity.

[0063] In some embodiments, the neural network model may use a multi-layer perceptron (MLP). MLP is an important neural network model, which is composed of multiple perceptron layers, and is therefore also called a deep neural network.

[0064] In some embodiments, information of each auto insurance claim case is input into a neural network model. The neural network model can output selection probabilities of each claim action corresponding to each auto insurance claim case. In one example, the claim actions corresponding to the auto insurance claim case can include claim settlement and claim rejection.

[0065] In some embodiments, the information of each auto insurance claim case can be input into the neural network model based on the corresponding time step. The time step can be understood as corresponding to a specific moment in the input data sequence. That is, each time step corresponds to the input of the information of a auto insurance claim case.

[0066] Step S102, determining a target claim action corresponding to each motor vehicle insurance claim case according to the selection probability of each claim action corresponding to each motor vehicle insurance claim case.

[0067] In the embodiment of the present application, the target claim action can be understood as the claim action to be executed selected by the neural network model. After the neural network model outputs the selection probabilities of each claim action corresponding to each auto insurance claim case, the target claim action corresponding to each auto insurance claim case can be determined based on the output selection probabilities of each claim action. In one example, the claim action with the highest selection probability can be determined as the target claim action.

[0068] Step S103, determining the reward value of the target claim settlement action according to the target claim settlement action and the reference claim settlement action corresponding to the target claim settlement action.

[0069] In the embodiment of the present application, the reference claim action corresponding to the target claim action can be understood as the actual claim action of the motor vehicle insurance claim case. The reward value can be understood as the feedback signal obtained by the neural network model at each time step (or state), which is usually a numerical value. The reward value can reflect the quality of a certain action taken by the neural network model in the current state.

[0070] In some embodiments, the reward value can be determined by determining whether the target claim action and the reference claim action corresponding to the target claim action are consistent. If the target claim action and the reference claim action corresponding to the target claim action are consistent, the reward value of the target claim action can be determined to be a positive number. If the target claim action and the reference claim action corresponding to the target claim action are inconsistent, the reward value of the target claim action can be determined to be a negative number.

[0071] In one example, the actual claim action of a car insurance claim case is "claim settlement". The information of the car insurance claim case is input into the neural network model, and the probability of outputting the "claim settlement" action is 70%, and the probability of the "claim rejection" action is 30%. The "claim settlement" action is determined as the target claim settlement action. The target claim settlement action is compared with the actual claim settlement action. If both are "claim settlement", the reward value of +10 is set for the target claim settlement action.

[0072] Step S104, determining the sum of the reward values ​​of the target claim actions corresponding to each motor insurance claim case as the cumulative reward of the neural network model.

[0073] In an embodiment of the present application, when training the neural network model, the neural network model is trained based on multiple auto insurance claims cases. Each target claim action corresponding to each auto insurance claim case corresponds to a reward value. After obtaining the reward value of the target claim action corresponding to each auto insurance claim case, the reward values ​​of the target claim actions corresponding to each auto insurance claim case are summed, and the obtained sum is determined as the cumulative reward of the neural network model. The cumulative reward can be understood as the sum of all rewards obtained by the neural network model after executing a series of actions.

[0074] Step S105, determining the second expected gradient based on the accumulated reward and the first expected gradient, the first expected gradient refers to the expected gradient of each target claim action with respect to the parameters of the neural network model, and the second expected gradient refers to the expected gradient of the expected reward of the neural network model with respect to the parameters of the neural network model.

[0075] In the embodiment of the present application, the first expected gradient can be understood as the expected gradient of each target claim settlement action with respect to the parameters of the neural network model. The first expected gradient can be expressed as That is, in a given state s t When action a t The gradient of the logarithm of the probability density function of with respect to the policy parameter θ. State s t It can be understood as information about various car insurance claims. Action a t It can be understood as the target claim action. The strategy parameter θ can be understood as the parameter of the neural network model.

[0076] In some embodiments, the second expected gradient can be understood as the expected gradient of the expected reward of the neural network model with respect to the parameters of the neural network model. The second expected gradient can be expressed as in, It can be understood as the second expected gradient. T can be understood as the time period, and t can be understood as each time step within the time period. It can be understood as the first expected gradient. G t It can be understood as cumulative rewards.

[0077] Step S106, updating the parameters of the neural network model according to the second expected gradient until the parameters of the neural network model meet the iteration termination condition, thereby obtaining a motor vehicle insurance fraud behavior identification model, which is used to identify motor vehicle insurance fraud behavior.

[0078] In the embodiment of the present application, the goal of reinforcement learning is to maximize the cumulative reward. By calculating the expected gradient, the optimal strategy can be effectively found to maximize the reward obtained.

[0079] After obtaining the second expected gradient, the parameters of the neural network model can be updated according to the second expected gradient. In one example, the parameters of the neural network model can be updated using a gradient ascent method. The gradient ascent method is an optimization algorithm for finding a maximum value.

[0080] In some embodiments, the parameters of the neural network model can be iteratively updated according to the second expected gradient until the parameters of the neural network model meet the iteration termination condition, thereby obtaining a vehicle insurance fraud behavior recognition model. In one example, the iteration termination condition can be that the parameters of the neural network model can make the accuracy of the neural network model reach a preset threshold.

[0081] By updating the parameters of the neural network model, the probability of selecting high-return actions can be increased and the probability of selecting low-return actions can be reduced, thereby obtaining a more accurate auto insurance fraud behavior identification model.

[0082] Steps S101 to S106 shown in the embodiment of the present application are to obtain the selection probability of each claim action corresponding to each auto insurance claim case by inputting the information of each auto insurance claim case into the neural network model respectively; and determine the target claim action corresponding to each auto insurance claim case according to the selection probability of each claim action corresponding to each auto insurance claim case. Then, determine the reward value of the target claim action according to the target claim action and the reference claim action corresponding to the target claim action, and determine the sum of the reward values ​​of the target claim actions corresponding to each auto insurance claim case as the cumulative reward of the neural network model. Determine the second expected gradient according to the cumulative reward and the first expected gradient, the first expected gradient refers to the expected gradient of each target claim action, and the second expected gradient refers to the expected gradient of the parameters of the neural network model. Finally, update the parameters of the neural network model according to the second expected gradient until the parameters of the neural network model meet the iteration termination condition, and obtain a motor insurance fraud behavior recognition model, which is used to identify motor insurance fraud behavior. In this way, by inputting the information of each auto insurance claim case into the neural network model, the cumulative reward of the neural network model is obtained, and through reinforcement learning, the parameters of the neural network model are updated according to the expected gradient of the parameters of the neural network model, so that the neural network model can continuously update the parameters, optimize the anti-fraud strategy, and improve the accuracy of identifying auto insurance fraud behavior.

[0083] See also Figure 2 In some embodiments, step S105 may include but is not limited to steps S201 to S202:

[0084] Step S201, multiplying the accumulated reward by the expected gradient of the parameters of the neural network model for each target claim action, to obtain the expected gradient of the parameters of the neural network model corresponding to each motor insurance claim case;

[0085] Step S202, summing up the expected gradients of the parameters of the neural network model corresponding to each motor insurance claim case to obtain a second expected gradient.

[0086] Specifically, the cumulative reward can be understood as the sum of all rewards obtained by the neural network model after executing a series of actions, that is, the sum of the reward values ​​of the target claim actions of each claim case.

[0087] In some embodiments, when calculating the second expected gradient, the cumulative reward is first multiplied by the expected gradient of each target claim action with respect to the parameters of the neural network model, that is, The expected gradient of the parameters of the neural network model corresponding to each auto insurance claim case is obtained. Then, the expected gradient of the parameters of the neural network model corresponding to each auto insurance claim case is summed to obtain a second expected gradient.

[0088] In some embodiments, the second desired gradient is expressed as in, It can be understood as the second expected gradient, N can be understood as the number of samples, that is, the number of auto insurance claims, and T can be understood as the time period. It can be understood as the first expected gradient. G t It can be understood as cumulative rewards.

[0089] In this embodiment, by determining the second expected gradient according to the expected gradient of each target claim settlement action with respect to the parameters of the neural network model and the cumulative reward of the neural network model, the accuracy of determining the second expected gradient can be improved.

[0090] In some embodiments, before determining the second expected gradient according to the accumulated reward and the first expected gradient, the identifying method includes:

[0091] Determine the log probability of the selection probability of the target claim action;

[0092] The gradient of the logarithmic probability is determined based on the chain rule to obtain the first expected gradient.

[0093] Specifically, before determining the second expected gradient according to the accumulated reward and the first expected gradient, the first expected gradient needs to be calculated first. The first expected gradient can be understood as the expected gradient of each target claim action with respect to the parameters of the neural network model.

[0094] In some embodiments, the first desired gradient can be expressed as The logarithmic probability of the selection probability of the target claim action. Then, the gradient of the logarithmic probability is determined based on the chain rule to obtain the first expected gradient. π θ (α t |s t ) can be understood as the strategy π in a given state s t Next select action a t When calculating, given the state s t It can be understood as information about car insurance claims. Action a t This is the target action corresponding to the car insurance claim case. θ (α t |s t ) Take the logarithm and get logπ θ (α t |s t ). After taking the logarithm, the calculation convenience can be improved. Then logπ θ (α t |s t ) to find the gradient, we can get the first expected gradient

[0095] In this embodiment, by taking the logarithm of the selection probability of the target claim settlement action and then calculating the first expected gradient, the convenience of calculation can be improved.

[0096] See also Figure 3 In some embodiments, step S106 may include but is not limited to steps S301 to S302:

[0097] Step S301: determining the product of the second expected gradient and the learning rate of the neural network model;

[0098] Step S302: The sum of the product and the parameters of the neural network model is determined as the updated parameters of the neural network model.

[0099] Specifically, when updating the initial parameters of the neural network model according to the second expected gradient, the gradient ascent method can be used. The gradient ascent method aims to maximize the objective function by iteratively updating the parameters. The gradient ascent method focuses on finding the maximum value of the function. The gradient ascent method is usually used to optimize a function, such as optimizing the likelihood function or other loss functions in machine learning.

[0100] In some embodiments, the learning rate of the neural network model is first determined. The learning rate determines the size of each update step. In one example, if the learning rate is too small, the convergence speed will be very slow; in another example, if the learning rate is too large, it may cause parameter oscillation or non-convergence.

[0101] In some embodiments, after determining the learning rate of the neural network model, the product of the second expected gradient and the learning rate of the neural network model can be determined, and then the sum of the product and the initial parameters is determined as the updated parameters of the neural network model.

[0102] In some embodiments, the update of the parameters of the neural network model using the gradient ascent method can be expressed as: Among them, θ can be understood as the updated parameters of the neural network model. 0 It can be understood as the parameters of the neural network model before updating. α can be understood as the learning rate of the neural network model. It can be understood as the second expected gradient.

[0103] In this embodiment, the accuracy of parameter updating can be improved by using the gradient ascent method to update the parameters of the neural network model.

[0104] In some embodiments, updating the parameters of the neural network model according to the second expected gradient until the parameters of the neural network model meet the iteration termination condition, thereby obtaining a vehicle insurance fraud behavior recognition model, including:

[0105] Determine the current cumulative return of the neural network model based on the reward value of the target claim action corresponding to each auto insurance claim case and the discount factor of the neural network model;

[0106] When the current cumulative return is greater than the cumulative return before the update, reduce the learning rate;

[0107] When the current cumulative reward is less than the cumulative reward before the update, increase the learning rate.

[0108] Specifically, the learning rate determines the size of each update. If the learning rate is too small, the convergence speed will be very slow; if the learning rate is too large, it may cause parameter oscillation or non-convergence. Therefore, it is necessary to adjust the learning rate of the neural network model in real time during the update of the parameters of the neural network model.

[0109] In some embodiments, the current cumulative reward of the neural network model can be determined based on the reward value of the target claim action corresponding to each auto insurance claim case and the discount factor of the neural network model. The cumulative reward can be understood as the sum of the rewards of all future time steps from the current moment.

[0110] In some embodiments, the current cumulative reward can be expressed as Among them, G t It can be expressed as cumulative return. γ can represent the discount factor of the neural network model. In one example, the discount factor γ ranges between [0,1). t can be understood as the time step. k can be understood as the number of reward values ​​of the target claim action corresponding to the car insurance claim case.

[0111] In some embodiments, the learning rate can be adjusted by comparing the magnitude relationship between the current cumulative reward and the reward before the update. If the current cumulative reward is greater than the cumulative reward before the update, the learning rate is reduced; if the current cumulative reward is less than the cumulative reward before the update, the learning rate is increased.

[0112] In some embodiments, the learning rate can also be automatically adjusted by setting the loss function to a negative policy gradient.

[0113] In this embodiment, by adjusting the learning rate in real time during the updating process of the parameters of the neural network model, the accuracy of updating the parameters of the neural network model can be improved.

[0114] In some embodiments, determining a reward value of a target claim settlement action according to the target claim settlement action and a reference claim settlement action corresponding to the target claim settlement action includes:

[0115] When the target claim action is the same as the reference claim action, the reward value is determined to be a positive number;

[0116] In the case where the target claim action is different from the reference claim action, the reward value is determined to be a negative number.

[0117] Specifically, the reward value can be understood as the feedback signal obtained by the neural network model at each time step (or state), which is usually a numerical value. The reward value can reflect the quality of a certain action taken by the neural network model in the current state. The time step can be understood as corresponding to a specific moment in the input data sequence. That is, each time step corresponds to the input of information on a car insurance claim case.

[0118] In some embodiments, when the target claim action is the same as the reference claim action, it means that the claim action selected by the system is consistent with the actual reference claim work of the motor insurance claim case, and the reward value can be determined to be a positive number, that is, the probability of the selection is increased. When the target claim action is different from the reference claim action, it means that the claim action selected by the system is inconsistent with the actual reference claim work of the motor insurance claim case, and the reward value can be determined to be a negative number, that is, the probability of the selection is reduced.

[0119] In this embodiment, by setting different reward values ​​according to different situations, the selection probability of each claim action can be adjusted in a timely manner, thereby making the parameter update of the neural network model more accurate.

[0120] See also Figure 4 , is the second flow chart of the method for identifying auto insurance fraud provided in the embodiment of the present application. Figure 4 As shown, the method for identifying auto insurance fraud includes the following steps:

[0121] Step S401, state definition: obtaining key information required in the auto insurance claim process.

[0122] Step S402, MLP network prediction: based on the state input, the model outputs the probability distribution of the action.

[0123] Step S403: Action execution: perform corresponding operations according to the action (claim settlement / claim rejection) selected by the model.

[0124] Step S404, reward feedback: feedback the reward value based on the actual results as a basis for subsequent optimization.

[0125] Step S405, strategy update: adjust the MLP network parameters through the strategy update step of reinforcement learning to optimize future decisions.

[0126] In this specific embodiment, the state and action are defined first. The state can be understood as containing all the key information in the insurance claim risk control problem, such as the insurer's personal information, vehicle information, policy information, claim history, accident type, accident severity, etc. MLP can solve the state that is both continuous (such as the insurer's renewal period, premium increase caused by past claims, vehicle age, etc.) or discrete (such as vehicle type, accident type, whether the number of claims exceeds 3, whether driving on holidays, etc.). Actions can be understood as defining all possible actions that the model can take, such as claims or rejections, changes in the amount of compensation, etc. (continuous or discrete).

[0127] In this specific embodiment, the policy network can use REINFORCE in the policy gradient algorithm, and MLP is used as the policy network. The input is the state, and the output is the probability distribution over each possible action. For example, the output of auto insurance (claim or rejection) is two neurons, representing the probability of each action. At the same time, the expected return after each action is set. For example, the reward value given by auto insurance (claim or rejection) is: -10 for claim, +10 for rejection.

[0128] In this specific embodiment, the expected return (i.e., the second expected gradient in this application) can be maximized by calculating the gradient of each time step multiplied by the corresponding advantage function or cumulative reward, and then calculating the expectation for all time steps. The expected gradient can be used to update the parameters of the policy function to improve the policy. That is, the probability of selecting a high-return action is increased, and the probability of selecting a low-return action is reduced, thereby improving the policy and maximizing the overall return.

[0129] In this specific embodiment, in reinforcement learning, an "agent" refers to an intelligent entity that makes decisions and takes actions in a certain environment to achieve a specific goal. The agent runs a piece of data, and after selecting the last action and obtaining the corresponding reward, it can compare it with the annotated answer and increase the output probability of all actions that are consistent with it, otherwise it will decrease. If the agent determines that the claim is rejected, and the annotated data also shows that the claim is rejected, then the probability of rejection of all agents who choose the action of rejection in this state will be increased.

[0130] In this specific embodiment, the neural network model can be trained on the Aether platform, deployed on the risk control platform, and called during the auto insurance claims process (reporting, investigation, and damage assessment), and the caller is the auto insurance claims payer.

[0131] See also Figure 5 , is a schematic diagram of the structure of a device for identifying auto insurance fraud provided in an embodiment of the present application. The device for identifying auto insurance fraud can implement the above-mentioned method for identifying auto insurance fraud. The device includes:

[0132] An input module 501 is used to input information of each motor insurance claim case into the neural network model to obtain selection probabilities of each claim action corresponding to each motor insurance claim case;

[0133] The first determination module 502 is used to determine the target claim settlement action corresponding to each motor vehicle insurance claim settlement case according to the selection probability of each claim settlement action corresponding to each motor vehicle insurance claim settlement case;

[0134] A second determination module 503, configured to determine a reward value of a target claim settlement action according to the target claim settlement action and a reference claim settlement action corresponding to the target claim settlement action;

[0135] The third determination module 504 is used to determine the sum of the reward values ​​of the target claim settlement actions corresponding to each motor insurance claim settlement case as the cumulative reward of the neural network model;

[0136] A fourth determination module 505, configured to determine a second expected gradient according to the accumulated reward and the first expected gradient, wherein the first expected gradient refers to the expected gradient of each target claim settlement action, and the second expected gradient refers to the expected gradient of the parameters of the neural network model;

[0137] The updating module 506 is used to update the parameters of the neural network model according to the second expected gradient until the parameters of the neural network model meet the iteration termination condition, thereby obtaining a motor vehicle insurance fraud behavior identification model, which is used to identify motor vehicle insurance fraud behavior.

[0138] The specific implementation of the device for identifying auto insurance fraud is basically the same as the specific implementation of the method for identifying auto insurance fraud described above, and will not be described in detail here.

[0139] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned method for identifying auto insurance fraud when executing the computer program. The electronic device can be any smart terminal including a tablet computer, a car computer, etc.

[0140] See also Figure 6 , Figure 6 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:

[0141] The processor 601 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (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 the present application;

[0142] The memory 602 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 602 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 602, and the processor 601 calls and executes the method for identifying auto insurance fraud in the embodiment of this application;

[0143] Input / output interface 603, used to implement information input and output;

[0144] Communication interface 604, used to realize communication interaction between the device and other devices, which can be realized by wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0145] Bus 605 , which transmits information between various components of the device (e.g., processor 601 , memory 602 , input / output interface 603 , and communication interface 604 );

[0146] The processor 601 , the memory 602 , the input / output interface 603 and the communication interface 604 are connected to each other in communication within the device via a bus 605 .

[0147] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned method for identifying auto insurance fraud.

[0148] An embodiment of the present application also provides a computer program product. When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device executes the method for identifying auto insurance fraud as described above.

[0149] The present application proposes a method, device, electronic device, medium and product for identifying auto insurance fraud. The method inputs the information of each auto insurance claim case into a neural network model to obtain the selection probability of each claim action corresponding to each auto insurance claim case; and determines the target claim action corresponding to each auto insurance claim case based on the selection probability of each claim action corresponding to each auto insurance claim case. Then, the reward value of the target claim action is determined based on the target claim action and the reference claim action corresponding to the target claim action, and the sum of the reward values ​​of the target claim actions corresponding to each auto insurance claim case is determined as the cumulative reward of the neural network model. The second expected gradient is determined based on the cumulative reward and the first expected gradient. Finally, the parameters of the neural network model are updated based on the second expected gradient until the parameters of the neural network model meet the iteration termination condition, thereby obtaining an auto insurance fraud identification model, which is used to identify auto insurance fraud. In this way, by inputting the information of each auto insurance claim case into the neural network model, the cumulative reward of the neural network model is obtained, and through reinforcement learning, the parameters of the neural network model are updated according to the expected gradient of the parameters of the neural network model, so that the neural network model can continuously update the parameters, optimize the anti-fraud strategy, and improve the accuracy of identifying auto insurance fraud behavior.

[0150] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0151] The method, device, electronic device, medium and product for identifying auto insurance fraud provided in the embodiment of the present application respectively input the information of each auto insurance claim case into the neural network model to obtain the selection probability of each claim action corresponding to each auto insurance claim case; determine the target claim action corresponding to each auto insurance claim case according to the selection probability of each claim action corresponding to each auto insurance claim case. Then determine the reward value of the target claim action according to the target claim action and the reference claim action corresponding to the target claim action, and determine the sum of the reward values ​​of the target claim actions corresponding to each auto insurance claim case as the cumulative reward of the neural network model. Determine the second expected gradient according to the cumulative reward and the first expected gradient, the first expected gradient refers to the expected gradient of each target claim action, and the second expected gradient refers to the expected gradient of the parameters of the neural network model. Finally, update the parameters of the neural network model according to the second expected gradient until the parameters of the neural network model meet the iteration termination condition, and obtain the auto insurance fraud behavior identification model, which is used to identify auto insurance fraud behavior. In this way, by inputting each auto insurance claim case into the neural network model, the cumulative reward of the neural network model is obtained, and through reinforcement learning, the parameters of the neural network model are updated according to the expected gradient of the parameters of the neural network model, so that the neural network model can continuously update the parameters, optimize the anti-fraud strategy, and improve the accuracy of identifying auto insurance fraud behavior.

[0152] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0153] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0154] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0155] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0156] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0157] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0158] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0159] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0160] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0161] If the integrated unit is implemented in the form of 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 the present application, 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, which is stored in a storage medium and includes multiple instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store programs.

[0162] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A method for identifying auto insurance fraud, characterized in that: The identification method comprises: Inputting information of each automobile insurance claim case into the neural network model respectively, and obtaining selection probabilities of each claim action corresponding to each automobile insurance claim case; Determining target claims actions corresponding to each of the motor vehicle insurance claims cases according to selection probabilities of each of the claims actions corresponding to each of the motor vehicle insurance claims cases; Determining a reward value for the target claim settlement action according to the target claim settlement action and a reference claim settlement action corresponding to the target claim settlement action; The sum of the reward values ​​of the target claim actions corresponding to each auto insurance claim case is determined as the cumulative reward of the neural network model; Determine a second expected gradient according to the accumulated reward and the first expected gradient, wherein the first expected gradient refers to the expected gradient of each target claim settlement action with respect to the parameters of the neural network model, and the second expected gradient refers to the expected gradient of the expected reward of the neural network model with respect to the parameters of the neural network model; The parameters of the neural network model are updated according to the second expected gradient until the parameters of the neural network model meet the iteration termination condition, thereby obtaining the auto insurance fraud behavior identification model, and the auto insurance fraud behavior identification model is used to identify auto insurance fraud behavior.

2. The identification method according to claim 1, characterized in that: The determining the second expected gradient according to the accumulated reward and the first expected gradient comprises: Multiplying the accumulated reward by the expected gradient of each target claim action with respect to the parameter of the neural network model respectively, to obtain the expected gradient of the parameter of the neural network model corresponding to each motor insurance claim case; The expected gradients of the parameters of the neural network model corresponding to each of the motor insurance claim cases are summed to obtain the second expected gradient.

3. The identification method according to claim 2, characterized in that: Before determining the second expected gradient according to the accumulated reward and the first expected gradient, the identification method includes: Determining the log probability of the selection probability of the target claim settlement action; The gradient of the logarithmic probability is determined based on the chain rule to obtain the first expected gradient.

4. The identification method according to claim 1, characterized in that: The updating of the initial parameters of the neural network model according to the second expected gradient includes: Determine the product of the second expected gradient and the learning rate of the neural network model; The sum of the product and the parameters of the neural network model is determined as the updated parameters of the neural network model.

5. The identification method according to claim 4, characterized in that: The updating of the parameters of the neural network model according to the second expected gradient until the parameters of the neural network model meet the iteration termination condition to obtain the auto insurance fraud behavior identification model includes: Determining the current cumulative return of the neural network model according to the reward value of the target claim action corresponding to each of the motor insurance claim cases and the discount factor of the neural network model; When the current cumulative reward is greater than the cumulative reward before updating, reducing the learning rate; When the current cumulative reward is less than the cumulative reward before the update, the learning rate is increased.

6. The identification method according to claim 1, characterized in that: The step of determining the reward value of the target claim settlement action according to the target claim settlement action and a reference claim settlement action corresponding to the target claim settlement action includes: When the target claim settlement action is the same as the reference claim settlement action, determining the reward value to be a positive number; When the target claim settlement action is different from the reference claim settlement action, the reward value is determined to be a negative number.

7. A device for identifying automobile insurance fraud, characterized in that: The identification device comprises: An input module, used to input information of each motor insurance claim case into the neural network model respectively, and obtain the selection probability of each claim action corresponding to each motor insurance claim case; A first determination module is used to determine the target claim settlement action corresponding to each of the motor vehicle insurance claim settlement cases according to the selection probability of each of the claim settlement actions corresponding to each of the motor vehicle insurance claim settlement cases; A second determination module, configured to determine a reward value of the target claim settlement action according to the target claim settlement action and a reference claim settlement action corresponding to the target claim settlement action; A third determination module is used to determine the sum of the reward values ​​of the target claim settlement actions corresponding to each motor insurance claim settlement case as the cumulative reward of the neural network model; a fourth determination module, configured to determine a second expected gradient according to the accumulated reward and the first expected gradient, wherein the first expected gradient refers to the expected gradient of each target claim settlement action with respect to the parameters of the neural network model, and the second expected gradient refers to the expected gradient of the expected reward of the neural network model with respect to the parameters of the neural network model; An updating module is used to update the parameters of the neural network model according to the second expected gradient until the parameters of the neural network model meet the iteration termination condition, thereby obtaining the auto insurance fraud behavior identification model, and the auto insurance fraud behavior identification model is used to identify auto insurance fraud behavior.

8. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the method for identifying auto insurance fraud as described in any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for identifying automobile insurance fraud according to any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that When the instructions in the computer program product are executed by a processor of an electronic device, the electronic device executes the method for identifying auto insurance fraud as described in any one of claims 1 to 6.

Images