Artificial intelligence-based policy classification method, device, equipment and storage medium

Abstract

The application relates to the fields of artificial intelligence and finance and insurance, and provides a policy classification method and device based on artificial intelligence, equipment and a storage medium, the method comprising the following steps: training a preset first machine learning model according to a preset first sample set to obtain a first classification model; training a preset second machine learning model according to a preset second sample set to obtain a second classification model; fusing the first classification model and the second classification model to obtain a candidate classification model; training the candidate classification model according to a preset third sample set to obtain a target classification model; and determining a classification result of a to-be-claimed policy according to enterprise claim data corresponding to the to-be-claimed policy and the target classification model. The application greatly improves the efficiency and accuracy of policy classification. The application also relates to the field of blockchains, and the storage medium can store data created according to the use of a blockchain node.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a policy classification method and device based on artificial intelligence, equipment and a storage medium. BACKGROUND

[0002] Group insurance is an insurance that takes a group as the insurance object, insures on behalf of the collective, and issues a total insurance contract by the insurer, and the insurer provides protection to the members in the group according to the contract. When the group insurance policy needs to be settled, the claim settlement personnel manually analyzes the enterprise claim settlement data corresponding to the group insurance policy to determine whether the group insurance policy that needs to be settled is a risk-free policy or a risky policy, and then determines the settlement strategy based on the classification of the group insurance policy. However, the claim settlement personnel needs to spend a lot of time on classification, which is low in efficiency, and it is difficult for the claim settlement personnel to discover the hidden risks of the enterprise through the enterprise claim settlement data, resulting in inaccurate classification and causing losses to the enterprise. Therefore, how to improve the efficiency and accuracy of policy classification is a problem to be solved at present. SUMMARY

[0003] The embodiments of the present application provide a policy classification method and device based on artificial intelligence, equipment and a storage medium, which aims to improve the efficiency and accuracy of policy classification.

[0004] In a first aspect, the embodiments of the present application provide a policy classification method based on artificial intelligence, comprising:

[0005] training a preset first machine learning model according to a preset first sample set to obtain a first classification model, the first sample set comprising a plurality of first samples, the first sample comprising enterprise claim settlement data corresponding to a first group insurance policy that has completed claim settlement and a labeled first label, the first label being used to identify the first group insurance policy as a risky group insurance policy;

[0006] training a preset second machine learning model according to a preset second sample set to obtain a second classification model, the second sample set comprising a plurality of second samples, the second sample comprising enterprise claim settlement data corresponding to a second group insurance policy that has completed claim settlement and a labeled second label, the second label being used to identify the second group insurance policy as a risk-free group insurance policy;

[0007] fusing the first classification model and the second classification model to obtain a candidate classification model;

[0008] training the candidate classification model according to a preset third sample set to obtain a target classification model, the third sample set comprising a plurality of the first samples and a plurality of the second samples;

[0009] According to the enterprise claim settlement data corresponding to the to-be-claimed policy and the target classification model, a classification result of the to-be-claimed policy is determined.

[0010] In a second aspect, the embodiments of the present application further provide a policy classification device based on artificial intelligence, which comprises:

[0011] The model training module is configured to train a preset first machine learning model according to a preset first sample set, to obtain a first classification model, wherein the first sample set comprises a plurality of first samples, and each first sample comprises enterprise claim settlement data corresponding to a first group insurance policy for which claim settlement has been completed and a labeled first label, and the first label is used to identify that the first group insurance policy is a high-risk group insurance policy.

[0012] The model training module is further configured to train a preset second machine learning model according to a preset second sample set, to obtain a second classification model, wherein the second sample set comprises a plurality of second samples, and each second sample comprises enterprise claim settlement data corresponding to a second group insurance policy for which claim settlement has been completed and a labeled second label, and the second label is used to identify that the second group insurance policy is a low-risk group insurance policy.

[0013] The model fusion module is configured to fuse the first classification model and the second classification model, to obtain a candidate classification model.

[0014] The model training module is further configured to train the candidate classification model according to a preset third sample set, to obtain a target classification model, wherein the third sample set comprises a plurality of the first samples and a plurality of the second samples.

[0015] The policy classification module is configured to determine a classification result of the to-be-claimed policy according to enterprise claim settlement data corresponding to the to-be-claimed policy and the target classification model.

[0016] In a third aspect, the embodiments of the present application further provide a computer device, which comprises a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the computer program implements the policy classification method based on artificial intelligence as described in the first aspect.

[0017] In a fourth aspect, the embodiments of the present application further provide a storage medium, which stores a computer program, wherein when the computer program is executed by a processor, the computer program implements the policy classification method based on artificial intelligence as described in the first aspect.

[0018] This application provides an artificial intelligence-based policy classification method, apparatus, device, and storage medium. Since a first label identifies a first group insurance policy as a risky group insurance policy, a first machine learning model is trained using a first sample set including samples with the first label to obtain a first classification model that accurately classifies risky group insurance policies. Similarly, since a second label identifies a second group insurance policy as a risky group insurance policy, a second machine learning model is trained using a second sample set including samples with the second label to obtain a second classification model that accurately classifies risk-free group insurance policies. The first and second classification models are then fused, and a third sample set including samples with the first and second labels is used to train the fused classification model to obtain an accurate target classification model. When a claim needs to be processed, the corresponding corporate claim data is input into this target classification model for classification, resulting in the classification result for the policy. This eliminates the need for claims personnel to perform risk analysis on the corporate claim data, significantly improving the efficiency and accuracy of policy classification. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating an artificial intelligence-based policy classification method provided in an embodiment of this application;

[0021] Figure 2 This is a schematic block diagram of an artificial intelligence-based policy classification device provided in an embodiment of this application;

[0022] Figure 3 yes Figure 2 A schematic block diagram of a sub-module of the policy sorting device in the system;

[0023] Figure 4 yes Figure 2 A schematic block diagram of another sub-module of the policy sorting device in the system;

[0024] Figure 5 This is a schematic block diagram of the structure of a computer device provided in an embodiment of this application.

[0025] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and the accompanying drawings. Detailed Implementation

[0026] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0027] The flowcharts shown in the drawings are only illustrative, and do not necessarily include all the contents and operations / steps, nor are they necessarily executed in the described order. For example, some operations / steps can be further decomposed, combined or partially merged, so the actual execution order can be changed according to the actual situation.

[0028] Group insurance is an insurance that takes a group as the insurance object, insures on behalf of the group, and issues a total insurance contract by the insurer, and the insurer provides protection to the members in the group according to the contract. When a group insurance policy needs to be settled, the claim settlement personnel manually analyzes the enterprise claim settlement data corresponding to the group insurance policy to determine whether the group insurance policy that needs to be settled is a risk-free policy or a risky policy, and then determines the settlement strategy based on the classification of the group insurance policy. However, the claim settlement personnel needs to spend a lot of time to classify, and the efficiency is low. Moreover, it is difficult for the claim settlement personnel to discover the hidden risks of the enterprise through the enterprise claim settlement data, resulting in inaccurate classification and causing losses to the enterprise.

[0029] To solve the above problems, the present application provides a policy classification method and device based on artificial intelligence, equipment and storage medium. Since the first label is used to identify the first group insurance policy as a risky group insurance policy, the first sample set including the sample of the first label is used to train the first machine learning model, and the first classification model for accurately classifying the risky group insurance policy can be obtained. Since the second label is used to identify the second group insurance policy as a risky group insurance policy, the second sample set including the sample of the second label is used to train the second machine learning model, and the second classification model for accurately classifying the risk-free group insurance policy can be obtained. Then, the first classification model and the second classification model are fused, and the third sample set including the sample of the first label and the sample of the second label is used to train the fused classification model, and the target classification model can be obtained. In this way, in the case of needing to settle the claim settlement policy, the corresponding enterprise claim settlement data is input into the target classification model for classification processing to obtain the classification result of the claim settlement policy. The claim settlement personnel does not need to analyze the risk of the enterprise claim settlement data, which greatly improves the efficiency and accuracy of policy classification.

[0030] The embodiments of the present application can acquire and process related data based on artificial intelligence technology. Among them, artificial intelligence (AI) is to use 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. Theory, method, technology and application system. Artificial intelligence basic technologies generally include technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology, robot technology, biometric technology, speech processing technology, natural language processing technology, and machine learning / deep learning, etc. Among them, the computer device can be a server or a terminal device, and the server can be a standalone server or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content distribution networks (CDN), and big data and artificial intelligence platforms. Basic cloud computing services such as platform, and the like.

[0031] Some embodiments of the present application will be described in detail below with reference to the accompanying drawings. The following embodiments and features in the embodiments can be combined with each other without conflict.

[0032] Please refer to Figure 1 , Figure 1 is a flowchart of a policy classification method based on artificial intelligence provided by the embodiments of the present application.

[0033] As Figure 1 shown, the policy classification method includes steps S101 to S105.

[0034] Step S101, training a first machine learning model according to a preset first sample set to obtain a first classification model.

[0035] In this embodiment, the preset first sample set is pre-generated, and the first sample set includes a plurality of first samples. The first sample includes enterprise claim data corresponding to a first group insurance policy that has completed claim settlement and a labeled first label. The first label is used to identify that the first group insurance policy is a risky group insurance policy. The group insurance policy is a policy of group insurance. The enterprise claim data corresponding to the first group insurance policy includes policy information and enterprise information of the first group insurance policy. The policy information can include a policy number, a premium, a policyholder, an insured person, and a beneficiary, etc. The enterprise information can include the name, attribute information of the enterprise, and the name and attribute information of the associated enterprise. The attribute information can include the registered address, registered time, registered capital, claim rate, shareholders, industry, size, and blacklist identifier of the enterprise. The blacklist identifier can be a first identifier or a second identifier. The first identifier is used to indicate that the enterprise is located in a preset blacklist. The second identifier is used to indicate that the enterprise is not located in the preset blacklist.

[0036] In some embodiments, the first machine learning model includes a first single Gaussian model (SGM) or a first neural network model. The first neural network model can include one or more of a convolutional neural network (CNN), a deep neural network (DNN), a fully neural network (FNN), a recurrent neural network (RNN), a long short-term memory (LSTM), etc.

[0037] In some embodiments, the first machine learning model is trained according to the preset first sample set to obtain a first classification model. The manner can be that: each time a different first sample is selected from the preset first sample set until each first sample in the first sample set is selected once; after each time a first sample is selected, the enterprise claim data in the selected first sample is input into the first machine learning model for processing to obtain a first prediction probability distribution; according to the first prediction probability distribution and the first label, a first cross-entropy loss value is calculated, and according to the first cross-entropy loss value, the parameters of the first machine learning model are updated. The first prediction probability distribution includes a first prediction probability and a second prediction probability. The first prediction probability is the probability that the group insurance policy corresponding to the first sample is risky. The second prediction probability is the probability that the group insurance policy corresponding to the first sample is not risky.

[0038] For example, the first sample set includes first sample A, first sample B, first sample C and first sample D. In the first training, first sample A is selected to train the preset first machine learning model Q0 to update the parameters of the first machine learning model Q0, and the first machine learning model Q1 is obtained. In the second training, first sample C is selected to train the first machine learning model Q1 to update the parameters of the first machine learning model Q1, and the first machine learning model Q2 is obtained. In the third training, first sample B is selected to train the first machine learning model Q2 to update the parameters of the first machine learning model Q2, and the first machine learning model Q3 is obtained. In the fourth training, first sample D is selected to train the first machine learning model Q3 to update the parameters of the first machine learning model Q4, and the first machine learning model Q4 is obtained. Since each first sample in the first sample set is selected once, the first machine learning model Q4 can be used as the first classification model.

[0039] In some embodiments, the manner of training the preset first machine learning model according to the preset first sample set to obtain the first classification model can also be: selecting a different first sample each time from the preset first sample set until each first sample in the first sample set is selected once; after selecting the first sample each time, inputting the enterprise claim data in the selected first sample into the preset risk score prediction model to perform risk score prediction, and obtaining a first risk score; inputting the first risk score into the first single Gaussian model for processing to obtain a second prediction probability distribution; calculating a second cross-entropy loss value according to the second prediction probability distribution and the first label, updating the parameters of the first single Gaussian model according to the second cross-entropy loss value, and obtaining a first target single Gaussian model. The parameters of the first single Gaussian model include a first expectation and a first variance. Since the enterprise claim data includes multi-dimensional data, it takes a long time to analyze and process the multi-dimensional data when training the single Gaussian model using the multi-dimensional data, and the efficiency is low. By normalizing the enterprise claim data into risk scores, the dimension of the data can be reduced, the data processing efficiency can be improved, and the training efficiency of the single Gaussian model is improved.

[0040] In some embodiments, the risk score prediction model is a pre-trained neural network model, and the training process of the neural network model includes: obtaining a training sample set, the training sample set including a plurality of training samples, each training sample including enterprise claim data corresponding to a completed group insurance policy and a labeled true risk score; selecting a training sample from the training sample set as a target training sample; inputting the enterprise claim data in the target training sample into the neural network model for risk score prediction processing to obtain a predicted risk score; calculating a model loss value according to the predicted risk score and the true risk score in the target training sample; updating parameters of the neural network model according to the model loss value; and returning to the step of selecting a training sample from the training sample set as a target training sample until each training sample in the training sample set is selected once. The neural network model can include one or more of a convolutional neural network (CNN), a deep neural network (DNN), a fully neural network (FNN), a recurrent neural network (RNN), a long short-term memory (LSTM), etc.

[0041] In some embodiments, the model loss value can be calculated using the predicted risk score and the true risk score in the target training sample by using a preset loss function. The preset loss function can be set based on actual conditions, and the present embodiment does not make specific limitations thereto. For example, the preset loss function can be any one of a mean square error (MSE) function, a root mean square error (RMSE) function, a mean absolute error (MAE) function, a least absolute deviations (LAD) function, a least square estimate (LSE) function, or a loss function formed by a combination of multiple functions.

[0042] In step S102, a second machine learning model is trained according to a preset second sample set to obtain a second classification model.

[0043] In the present embodiment, the second sample set includes a plurality of second samples, each second sample including enterprise claim data corresponding to a completed second group insurance policy and a labeled second label, the second label being used to identify the second group insurance policy as a risk-free group insurance policy.

[0044] In some embodiments, the second machine learning model comprises a second single Gaussian model (SGM) or a second neural network model, which can comprise one or more of a convolutional neural network (CNN), a deep neural network (DNN), a fully neural network (FNN), a recurrent neural network (RNN), a long short-term memory (LSTM), etc. The parameters of the first single Gaussian model are different from those of the second single Gaussian model, the network structure of the first neural network model is the same as that of the second neural network model, and the parameters of the first neural network model are different from those of the second neural network model.

[0045] In some embodiments, the second machine learning model is trained according to the preset second sample set, and the manner of obtaining the second classification model can be: each time a different second sample is selected from the preset second sample set until each second sample in the second sample set is selected once; after each time a second sample is selected, the enterprise claim data in the selected second sample is input into the second machine learning model for processing to obtain a third predicted probability distribution; according to the third predicted probability distribution and the second label, a third cross-entropy loss value is calculated, and the parameters of the second machine learning model are updated according to the third cross-entropy loss value.

[0046] For example, the second sample set includes second sample a, second sample b, second sample c, and second sample d. In the first training, the second sample a is selected to train the preset second machine learning model P0 to update the parameters of the second machine learning model P0, and the second machine learning model P1 is obtained. In the second training, the second sample c is selected to train the second machine learning model P1 to update the parameters of the second machine learning model P1, and the second machine learning model P2 is obtained. In the third training, the second sample b is selected to train the second machine learning model P2 to update the parameters of the second machine learning model P2, and the second machine learning model P3 is obtained. In the fourth training, the second sample d is selected to train the second machine learning model P3 to update the parameters of the second machine learning model P4, and the second machine learning model P4 is obtained. Since each second sample in the second sample set is selected once, the second machine learning model P4 can be used as the second classification model.

[0047] In some embodiments, the manner of training the preset second machine learning model according to the preset second sample set to obtain the second classification model can also be: selecting a different first sample each time from the preset second sample set until each second sample in the second sample set is selected once; after each selection of a second sample, inputting the enterprise claim data in the selected second sample into the preset risk score prediction model to perform risk score prediction, to obtain a third risk score; inputting the third risk score into the second single Gaussian model for processing, to obtain a fourth prediction probability distribution; according to the fourth prediction probability distribution and the second label, calculating a fourth cross-entropy loss value, and according to the fourth cross-entropy loss value, updating the parameters of the second single Gaussian model to obtain a second target single Gaussian model. The parameters of the second single Gaussian model include a second expectation and a second variance.

[0048] In step S103, the first classification model and the second classification model are fused to obtain a candidate classification model.

[0049] In this embodiment, the first classification model can include a first target single Gaussian model or a first target neural network model, and the second classification model can include a second target single Gaussian model or a second target neural network model. The types of the first classification model and the second classification model are the same. For example, the first classification model is a first target single Gaussian model, and the second classification model is a second target single Gaussian model. For another example, the first classification model is a first target neural network model, and the second classification model is a second target neural network model.

[0050] In an embodiment, the manner of fusing the first classification model and the second classification model to obtain the candidate classification model can be: obtaining a first weighting coefficient and a second weighting coefficient, the sum of the first weighting coefficient and the second weighting coefficient being equal to 1; calculating the product of the first weighting coefficient and the first target single Gaussian model to obtain a first weighted single Gaussian model; calculating the product of the second weighting coefficient and the second target single Gaussian model to obtain a second weighted single Gaussian model; and summing the first weighted single Gaussian model and the second weighted single Gaussian model to obtain a candidate Gaussian mixture model. By weighting and summing the first target single Gaussian model and the second target single Gaussian model using the corresponding weighting coefficients, an accurate candidate Gaussian mixture model can be obtained.

[0051] The first weighting coefficient and the second weighting coefficient can be set based on actual conditions, and the embodiments of the present application do not make specific limitations thereto. For example, the first weighting coefficient and the second weighting coefficient are both 0.5. For example, the first target single Gaussian model is f1(x), the second target single Gaussian model is f2(x), the first weighting coefficient is a1, and the second weighting coefficient is a2, then the candidate Gaussian mixture model f(x) = a1*f1(x) + a2*f2(x), and a1 + a2 = 1.

[0052] In some embodiments, the manner of fusing the first classification model and the second classification model to obtain the candidate classification model can also be: obtaining the first preset coefficient and the second preset coefficient, and obtaining the parameters of the first target neural network model and the parameters of the second target neural network model; calculating the product of the first preset coefficient and the parameters of the first target neural network model to obtain the first weighted parameter; calculating the product of the second preset coefficient and the parameters of the second target neural network model to obtain the second weighted parameter; summing the first weighted parameter and the second weighted parameter to obtain the target parameter; updating the parameters of the first target neural network model or the second target neural network model to the target parameter to obtain the candidate neural network model. The first preset coefficient and the second preset coefficient can be set based on actual conditions, and the embodiments of the present application do not make specific limitations thereto. For example, the first preset coefficient is 0.6, and the second preset coefficient is 0.45.

[0053] Step S104, training the candidate classification model according to the preset third sample set to obtain a target classification model.

[0054] In the embodiment, the third sample set includes a plurality of first samples and a plurality of second samples, the first sample includes enterprise claim data corresponding to a first group insurance policy that has completed claim and a labeled first label, the second sample includes enterprise claim data corresponding to a second group insurance policy that has completed claim and a labeled second label, and the candidate classification model can include a candidate Gaussian mixture model or a candidate neural network model.

[0055] In some embodiments, the manner of training the candidate classification model according to the preset third sample set to obtain the target classification model can be: selecting a different sample each time from the preset third sample set until each sample in the third sample set is selected once; after selecting the sample each time, inputting the enterprise claim data in the selected sample into the candidate Gaussian mixture model for processing to obtain a first predicted label; determining a loss value according to the first predicted label and the true label in the selected sample, and updating the first weighting coefficient and the second weighting coefficient of the candidate Gaussian mixture model according to the loss value. After each sample in the third sample set is selected once, the last updated candidate Gaussian mixture model is determined as the target Gaussian mixture model, the loss value can be determined based on the first predicted label and the true label in the selected sample by using a cross-entropy loss function, and the first weighting coefficient and the second weighting coefficient can be updated by a gradient descent algorithm.

[0056] In some embodiments, the manner of training the candidate classification model according to the preset third sample set to obtain the target classification model can also be: each time a different sample is selected from the preset third sample set until each sample in the third sample set is selected once; after each time a sample is selected, the enterprise claim data in the selected sample is input into the risk score prediction model for risk score prediction to obtain a fifth risk score; the fifth risk score is input into the candidate Gaussian mixture model for processing to obtain a fifth prediction label; according to the fifth prediction label and the true label in the selected sample, a loss value is determined, and the first weight coefficient and the second weight coefficient of the candidate Gaussian mixture model are updated according to the loss value. Wherein, after each sample in the third sample set is selected once, the last updated candidate Gaussian mixture model is determined as the target Gaussian mixture model.

[0057] Step S105, according to the enterprise claim data corresponding to the to-be-claimed policy and the target classification model, determining a classification result of the to-be-claimed policy.

[0058] In this embodiment, the enterprise claim data corresponding to the to-be-claimed policy is input into the target classification model for classification prediction processing, and the classification result of the to-be-claimed policy can be obtained. Wherein, the classification result can be that the to-be-claimed policy is a high-risk group insurance policy or that the to-be-claimed policy is a low-risk group insurance policy. The target classification model can include a target Gaussian mixture model or a target neural network model.

[0059] In some embodiments, the manner of determining the classification result of the to-be-claimed policy according to the enterprise claim data corresponding to the to-be-claimed policy and the target classification model can be: inputting the enterprise claim data corresponding to the to-be-claimed policy into the risk score prediction model for risk score prediction to obtain a second risk score; inputting the second risk score into the target Gaussian mixture model for processing to obtain the classification result of the to-be-claimed policy. By normalizing the multi-dimensional enterprise claim data into one-dimensional risk score, the classification result of the to-be-claimed policy can be quickly and accurately determined based on the risk score and the target Gaussian mixture model.

[0060] Please refer to Figure 2 , Figure 2 is a schematic block diagram of a policy classification device based on artificial intelligence provided by the embodiments of the present application.

[0061] As Figure 2 shown, the policy classification device 200 includes:

[0062] The model training module 210 is configured to train a preset first machine learning model according to a preset first sample set to obtain a first classification model, the first sample set including a plurality of first samples, each first sample including enterprise claim data corresponding to a first group insurance policy that has completed claim settlement and a labeled first label, the first label being used to identify the first group insurance policy as a risky group insurance policy.

[0063] The model training module 210 is further configured to train a preset second machine learning model according to a preset second sample set to obtain a second classification model, the second sample set including a plurality of second samples, each second sample including enterprise claim data corresponding to a second group insurance policy that has completed claim settlement and a labeled second label, the second label being used to identify the second group insurance policy as a non-risky group insurance policy.

[0064] The model fusion module 220 is configured to fuse the first classification model and the second classification model to obtain a candidate classification model.

[0065] The model training module 210 is further configured to train the candidate classification model according to a preset third sample set to obtain a target classification model, the third sample set including a plurality of the first samples and a plurality of the second samples.

[0066] The policy classification module 230 is configured to determine a classification result of a to-be-claimed policy according to enterprise claim data corresponding to the to-be-claimed policy and the target classification model.

[0067] In some embodiments, the first machine learning model includes a first single Gaussian model or a first neural network model, the second machine learning model includes a second single Gaussian model or a second neural network model, parameters of the first single Gaussian model are different from parameters of the second single Gaussian model, a network structure of the first neural network model is the same as a network structure of the second neural network model, and parameters of the first neural network model are different from parameters of the second neural network model.

[0068] In some embodiments, as shown in Figure 3 The model training module 210 includes:

[0069] The selection sub-module 211 is configured to select one different first sample each time from the preset first sample set until each first sample in the first sample set is selected once.

[0070] The processing sub-module 212 is configured to input the enterprise claim data in the selected first sample into the first machine learning model for processing to obtain a first predicted probability distribution after each first sample is selected.

[0071] The processing submodule 212 is further configured to calculate a first cross-entropy loss value according to the first predicted probability distribution and the first label.

[0072] The updating submodule 213 is configured to update parameters of the first machine learning model according to the first cross-entropy loss value.

[0073] In some embodiments, as shown in Figure 4 The model fusion module 220 includes:

[0074] The obtaining submodule 221 is configured to obtain a first weighting coefficient and a second weighting coefficient, wherein a sum of the first weighting coefficient and the second weighting coefficient is equal to 1.

[0075] The calculation submodule 222 is configured to calculate a product of the first weighting coefficient and the first target single Gaussian model to obtain a first weighted single Gaussian model.

[0076] The calculation submodule 222 is further configured to calculate a product of the second weighting coefficient and the second target single Gaussian model to obtain a second weighted single Gaussian model.

[0077] The fusion submodule 223 is configured to sum the first weighted single Gaussian model and the second weighted single Gaussian model to obtain a candidate Gaussian mixture model.

[0078] In some embodiments, the model training module 210 is further configured to:

[0079] A different sample is selected each time from a preset third sample set until each sample in the third sample set is selected once.

[0080] After each selection of a sample, enterprise claim data in the selected sample is input into the candidate Gaussian mixture model for processing to obtain a first predicted label.

[0081] According to the first predicted label and a true label in the selected sample, a loss value is determined, and the first weighting coefficient and the second weighting coefficient of the candidate Gaussian mixture model are updated according to the loss value.

[0082] In some embodiments, the model training module 210 is further configured to:

[0083] A different first sample is selected each time from a preset first sample set until each first sample in the first sample set is selected once.

[0084] After each selection of a first sample, enterprise claim data in the selected first sample is input into a preset risk score prediction model for risk score prediction to obtain a first risk score.

[0085] input the first risk score into the first single Gaussian model for processing to obtain a second predicted probability distribution;

[0086] According to the second predicted probability distribution and the first label, a second cross-entropy loss value is calculated, and parameters of the first single Gaussian model are updated according to the second cross-entropy loss value.

[0087] In some embodiments, the insurance policy classification module 230 is further configured to:

[0088] inputting enterprise claim data corresponding to the to-be-claimed insurance policy into the risk score prediction model for risk score prediction to obtain a second risk score;

[0089] inputting the second risk score into the target Gaussian mixture model for processing to obtain a classification result of the to-be-claimed insurance policy.

[0090] It should be noted that, for the convenience and brevity of description, the specific working processes of the above-described apparatuses and modules and units can refer to the corresponding processes in the foregoing insurance policy classification method embodiments based on artificial intelligence, which will not be described herein.

[0091] The apparatuses provided in the foregoing embodiments can be implemented as a computer program in the form of a program product, which can run on a computer device as shown in Figure 5 .

[0092] Please refer to Figure 5 , Figure 5 is a structural schematic block diagram of a computer device provided in an embodiment of the present application. The computer device can be a server or a terminal.

[0093] As shown in Figure 5 , the computer device includes a processor, a memory and a network interface connected through a system bus, wherein the memory can include a storage medium and an internal memory.

[0094] The storage medium can store an operating system and a computer program. The computer program includes program instructions, which, when executed, can cause the processor to perform any one of the insurance policy classification methods based on artificial intelligence.

[0095] The processor is configured to provide computing and control capabilities to support the operation of the entire computer device.

[0096] The network interface is configured to perform network communication, such as sending assigned tasks. Those skilled in the art can understand that Figure 5The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0097] It should be understood that the processor can be a central processing unit (CPU), and the processor can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0098] Among them, in an embodiment, the processor is configured to run a computer program stored in the memory to implement the following steps:

[0099] According to the preset first sample set, a preset first machine learning model is trained to obtain a first classification model, the first sample set includes a plurality of first samples, the first sample includes enterprise claim data corresponding to a first group insurance policy that has completed claim and a labeled first label, and the first label is used to identify that the first group insurance policy is a risky group insurance policy;

[0100] According to the preset second sample set, a preset second machine learning model is trained to obtain a second classification model, the second sample set includes a plurality of second samples, the second sample includes enterprise claim data corresponding to a second group insurance policy that has completed claim and a labeled second label, and the second label is used to identify that the second group insurance policy is a non-risk group insurance policy;

[0101] The first classification model and the second classification model are fused to obtain a candidate classification model;

[0102] According to the preset third sample set, the candidate classification model is trained to obtain a target classification model, the third sample set includes a plurality of first samples and a plurality of second samples;

[0103] According to the enterprise claim data corresponding to the to-be-claimed policy and the target classification model, a classification result of the to-be-claimed policy is determined.

[0104] In some embodiments, the first machine learning model comprises a first single Gaussian model or a first neural network model, the second machine learning model comprises a second single Gaussian model or a second neural network model, parameters of the first single Gaussian model are different from parameters of the second single Gaussian model, a network structure of the first neural network model is the same as a network structure of the second neural network model, and parameters of the first neural network model are different from parameters of the second neural network model.

[0105] In some embodiments, when implementing training of a preset first machine learning model according to a preset first sample set to obtain a first classification model, the processor is configured to implement:

[0106] selecting a different first sample each time from the preset first sample set until each of the first samples in the first sample set is selected once;

[0107] after selecting the first sample each time, inputting enterprise claim data in the selected first sample into the first machine learning model for processing to obtain a first prediction probability distribution;

[0108] calculating a first cross-entropy loss value according to the first prediction probability distribution and the first label, and updating parameters of the first machine learning model according to the first cross-entropy loss value.

[0109] In some embodiments, the first classification model comprises a first target single Gaussian model, the second classification model comprises a second target single Gaussian model, and when implementing fusion of the first classification model and the second classification model to obtain a candidate classification model, the processor is configured to implement:

[0110] obtaining a first weighting coefficient and a second weighting coefficient, wherein a sum of the first weighting coefficient and the second weighting coefficient is equal to 1;

[0111] calculating a product of the first weighting coefficient and the first target single Gaussian model to obtain a first weighted single Gaussian model;

[0112] calculating a product of the second weighting coefficient and the second target single Gaussian model to obtain a second weighted single Gaussian model;

[0113] summing the first weighted single Gaussian model and the second weighted single Gaussian model to obtain a candidate Gaussian mixture model.

[0114] In some embodiments, when implementing training of the candidate classification model according to a preset third sample set to obtain a target classification model, the processor is configured to implement:

[0115] select a different sample from the preset third sample set each time until each sample in the third sample set is selected once;

[0116] After each sample is selected, the enterprise claim data in the selected sample is input into the candidate Gaussian mixture model for processing to obtain a first predicted label;

[0117] According to the first predicted label and the true label in the selected sample, a loss value is determined, and the first weighting coefficient and the second weighting coefficient of the candidate Gaussian mixture model are updated according to the loss value.

[0118] In some embodiments, when implementing training of a preset first machine learning model according to a preset first sample set to obtain a first classification model, the processor is configured to implement:

[0119] select a different first sample from the preset first sample set each time until each first sample in the first sample set is selected once;

[0120] After each first sample is selected, the enterprise claim data in the selected first sample is input into a preset risk score prediction model for risk score prediction to obtain a first risk score;

[0121] The first risk score is input into the first single Gaussian model for processing to obtain a second predicted probability distribution;

[0122] According to the second predicted probability distribution and the first label, a second cross-entropy loss value is calculated, and parameters of the first single Gaussian model are updated according to the second cross-entropy loss value.

[0123] In some embodiments, the target classification model includes a target Gaussian mixture model, and when implementing determination of a classification result of a to-be-claimed policy according to enterprise claim data corresponding to the to-be-claimed policy and the target classification model, the processor is configured to implement:

[0124] The enterprise claim data corresponding to the to-be-claimed policy is input into the risk score prediction model for risk score prediction to obtain a second risk score;

[0125] The second risk score is input into the target Gaussian mixture model for processing to obtain the classification result of the to-be-claimed policy.

[0126] It should be noted that, for the convenience and brevity of description, the specific working process of the computer device described above can refer to the corresponding process in the foregoing policy classification method embodiments based on artificial intelligence, which will not be described herein.

[0127] Those skilled in the art can clearly understand the application by the description of the above embodiments that the application can be implemented by means of software and the necessary universal hardware platform. Based on such understanding, the technical solutions of the application can be embodied in the form of a software product, which can be stored in a storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in various embodiments or some parts of the embodiments of the application.

[0128] The application also provides a storage medium having a computer program stored thereon, wherein the computer program includes program instructions, and the method implemented when the program instructions are executed can refer to the various embodiments of the application.

[0129] The storage medium can be volatile or non-volatile. The storage medium can be an internal storage unit of the computer device, such as a hard disk or a memory of the computer device. The storage medium can also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc.

[0130] Further, the storage medium can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function, etc., and the data storage area can store data created according to the use of the blockchain node, etc.

[0131] The blockchain referred to in the application is a new application mode of distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. The blockchain is essentially a decentralized database, which is a series of data blocks associated by using cryptographic methods, each data block contains the information of a batch of network transactions, and is used to verify the validity (anti-fake) of the information and generate the next block. The blockchain can include a blockchain underlying platform, a platform product service layer, and an application service layer, etc.

[0132] It should be understood that the terms used in this specification of the application are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in the specification and the appended claims of the application, the singular forms "a", "an" and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0133] It should also be understood that, in the claims hereinafter presented by way of example, any means for performing an operation and any reference characters for performing an operation are intended to encompass any means for performing the operation and any reference characters for performing the operation, even if performing the operation is not explicitly disclosed in the claims. It is intended that the application be considered as encompassing any such modified or equivalent forms of the application disclosed and the right to equiva lent modifications and variations of the present application is reserved.

[0134] The above embodiments of the present application are only used for description, and are not used to judge the embodiment of the present application. The above description is only the specific embodiment of the present application, and the protection scope of the present application is not limited to this. Any modification or replacement within the technical range disclosed by the present application can be easily thought by the person skilled in the art, and these modifications or replacements should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An artificial intelligence-based insurance policy classification method, characterized by, The method comprises the following steps: training a preset first machine learning model according to a preset first sample set to obtain a first classification model, wherein the first sample set comprises a plurality of first samples, each first sample comprises enterprise claim data corresponding to a first group insurance policy that has completed claim and a labeled first label, and the first label is used to identify that the first group insurance policy is a risky group insurance policy; training a preset second machine learning model according to a preset second sample set to obtain a second classification model, wherein the second sample set comprises a plurality of second samples, each second sample comprises enterprise claim data corresponding to a second group insurance policy that has completed claim and a labeled second label, and the second label is used to identify that the second group insurance policy is a non-risky group insurance policy; fusing the first classification model and the second classification model to obtain a candidate classification model; training the candidate classification model according to a preset third sample set to obtain a target classification model, wherein the third sample set comprises a plurality of first samples and a plurality of second samples; determining a classification result of a to-be-claimed policy according to enterprise claim data corresponding to the to-be-claimed policy and the target classification model; wherein the first classification model comprises a first target single Gaussian model, the second classification model comprises a second target single Gaussian model, the candidate classification model comprises a candidate Gaussian mixture model, and the fusing the first classification model and the second classification model to obtain a candidate classification model comprises: obtaining a first weighting coefficient and a second weighting coefficient, wherein the sum of the first weighting coefficient and the second weighting coefficient is equal to 1; calculating the product of the first weighting coefficient and the first target single Gaussian model to obtain a first weighted single Gaussian model; calculating the product of the second weighting coefficient and the second target single Gaussian model to obtain a second weighted single Gaussian model; summing the first weighted single Gaussian model and the second weighted single Gaussian model to obtain a candidate Gaussian mixture model; the training the candidate classification model according to a preset third sample set to obtain a target classification model comprises: selecting a different sample each time from the preset third sample set until each sample in the third sample set is selected once; after selecting a sample each time, inputting the enterprise claim data in the selected sample into a risk score prediction model to perform risk score prediction and obtain a fifth risk score, wherein the risk score prediction model is obtained by training a neural network model according to a training sample set in advance, and each training sample in the training sample set comprises enterprise claim data corresponding to a group insurance policy that has completed claim and a labeled true risk score; inputting the fifth risk score into the candidate Gaussian mixture model for processing to obtain a fifth predicted label; According to the fifth prediction label and the true label in the selected sample, a loss value is determined, and the first weighting coefficient and the second weighting coefficient are updated according to the loss value, so as to update the candidate Gaussian mixture model, wherein after each sample in the third sample set is selected once, the last updated candidate Gaussian mixture model is determined as the target Gaussian mixture model.

2. The policy classification method of claim 1, wherein, The first machine learning model comprises a first single Gaussian model, and the second machine learning model comprises a second single Gaussian model, and parameters of the first single Gaussian model are different from parameters of the second single Gaussian model.

3. The policy classification method of claim 2, wherein, The training of the preset first machine learning model according to the preset first sample set to obtain the first classification model comprises: Each time, a different first sample is selected from the preset first sample set until each first sample in the first sample set is selected once. After each time of selecting a first sample, the enterprise claim data in the selected first sample is input into the first machine learning model for processing to obtain a first prediction probability distribution. According to the first prediction probability distribution and the first label, a first cross-entropy loss value is calculated, and parameters of the first machine learning model are updated according to the first cross-entropy loss value.

4. The policy classification method of claim 2, wherein, The training of the preset first machine learning model according to the preset first sample set to obtain the first classification model comprises: Each time, a different first sample is selected from the preset first sample set until each first sample in the first sample set is selected once. After each time of selecting a first sample, the enterprise claim data in the selected first sample is input into the first machine learning model for processing to obtain a first prediction probability distribution. According to the first prediction probability distribution and the first label, a first cross-entropy loss value is calculated, and parameters of the first machine learning model are updated according to the first cross-entropy loss value. The determination of the classification result of the to-be-claimed policy according to the enterprise claim data corresponding to the to-be-claimed policy and the target classification model comprises:

5. The policy classification method of claim 1, wherein, The enterprise claim data corresponding to the to-be-claimed policy is input into the risk score prediction model for risk score prediction to obtain a second risk score. The second risk score is input into the target Gaussian mixture model for processing to obtain the classification result of the to-be-claimed policy. The policy classification device comprises: 6.A policy classification device based on artificial intelligence, characterized by, A model training module is configured to train a preset first machine learning model according to a preset first sample set to obtain a first classification model, the first sample set comprises a plurality of first samples, the first sample comprises enterprise claim data corresponding to a first group insurance policy that has completed claim and a labeled first label, and the first label is used to identify that the first group insurance policy is a risky group insurance policy. ​ The model training module is further configured to train a preset second machine learning model according to a preset second sample set to obtain a second classification model, the second sample set includes a plurality of second samples, the second sample includes enterprise claim data corresponding to a second group insurance policy that has completed claim and a labeled second label, and the second label is used to identify that the second group insurance policy is a risk-free group insurance policy. The model fusion module is configured to fuse the first classification model and the second classification model to obtain a candidate classification model. The model training module is further configured to train the candidate classification model according to a preset third sample set to obtain a target classification model, the third sample set includes a plurality of first samples and a plurality of second samples. The policy classification module is configured to determine a classification result of a to-be-claimed policy according to enterprise claim data corresponding to the to-be-claimed policy and the target classification model. The first classification model includes a first target single Gaussian model, the second classification model includes a second target single Gaussian model, and the candidate classification model includes a candidate Gaussian mixture model. The model fusion module is further configured to: obtain a first weighting coefficient and a second weighting coefficient, a sum of the first weighting coefficient and the second weighting coefficient is equal to 1; calculate a product of the first weighting coefficient and the first target single Gaussian model to obtain a first weighted single Gaussian model; calculate a product of the second weighting coefficient and the second target single Gaussian model to obtain a second weighted single Gaussian model; perform summation on the first weighted single Gaussian model and the second weighted single Gaussian model to obtain a candidate Gaussian mixture model. The model training module is further configured to: select a different sample each time from the preset third sample set until each sample in the third sample set is selected once. After selecting a sample each time, input enterprise claim data in the selected sample into a risk score prediction model to perform risk score prediction to obtain a fifth risk score, wherein the risk score prediction model is obtained by pre-training a neural network model according to a training sample set, and each training sample in the training sample set includes enterprise claim data corresponding to a group insurance policy that has completed claim and a labeled real risk score; input the fifth risk score into the candidate Gaussian mixture model for processing to obtain a fifth predicted label; 7. A computer device, comprising: determine a loss value according to the fifth predicted label and a real label in the selected sample, and update the first weighting coefficient and the second weighting coefficient according to the loss value to update the candidate Gaussian mixture model, wherein after each sample in the third sample set is selected once, the last updated candidate Gaussian mixture model is determined as the target Gaussian mixture model. The computer device includes a processor, a memory, and a computer program stored on the memory and executable by the processor, wherein when the computer program is executed by the processor, the computer program implements the artificial intelligence-based policy classification method in any one of claims 1 to 5.

8. A storage medium for computer-readable storage, characterized in that, The storage medium has stored thereon a computer program, and the computer program, when executed by a processor, implements the artificial intelligence-based insurance policy classification method according to any one of claims 1 to 5.

Images