Insurance policy claim amount prediction method and device, electronic equipment and storage medium
By acquiring and utilizing insurance policy information and data, combined with prediction models and discriminant models, the problem of inaccurate insurance claim amounts was solved, and accurate predictions of insurance policy claim amounts were achieved.
Patent Information
- Application Number
- CN202510779157.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-11
- Publication Date
- 2025-09-16
AI Technical Summary
In the existing technology, the determination of insurance claim amounts relies on manual experience estimation, resulting in inaccurate claim amounts.
By obtaining reference object information, reference policy information and reference claims data, the preset policy amount prediction model and discriminant model are used to predict the policy claim amount, and the prediction accuracy is improved by training the target policy amount prediction model.
It achieves accurate prediction of insurance policy claim amounts, solves the problem of inaccurate claim amounts caused by relying on manual experience, and improves the accuracy of prediction results.
Smart Images
Figure CN120655433A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of insurance amount prediction, is applicable to the financial field, and particularly relates to a method and device for predicting insurance policy claim amounts, an electronic device, and a storage medium. Background Art
[0002] The policy claim amount refers to the amount payable to the insured after an accident. For example, in the event of a traffic accident, if the vehicle repair cost is 80,000 yuan and the policy's maximum claim amount is greater than 80,000 yuan, the claim amount will be 80,000 yuan. When filling in the policy's maximum claim amount, staff typically estimate and set the claim amount based on experience, which can easily lead to inaccurate claims. Therefore, how to accurately determine insurance claim amounts has become a pressing technical issue. Summary of the Invention
[0003] The main purpose of the embodiments of the present application is to propose a method and device for predicting the insurance claim amount, an electronic device and a storage medium, so as to accurately determine the insurance claim amount.
[0004] To achieve the above objectives, a first aspect of an embodiment of the present application provides a method for predicting insurance policy claim amounts, the method comprising:
[0005] Obtain reference object information, reference policy information, and reference claims data;
[0006] Predicting the insurance claim amount of the reference object information and the reference policy information using a preset insurance claim amount prediction model to obtain a predicted insurance claim amount;
[0007] performing accuracy discrimination on the reference object information, the reference policy information, and the reference claims data using a preset policy amount discrimination model to obtain a policy amount prediction accuracy of the predicted policy amount; wherein the policy amount accuracy is used to represent the probability that the predicted policy amount is the same as the reference claims data;
[0008] Training the preset policy amount prediction model based on the reference claim data, the predicted policy amount, and the policy amount accuracy to obtain a target policy amount prediction model;
[0009] Obtain target object information and target policy information;
[0010] The target policy amount prediction model is used to predict the policy claim amount for the target object information and the target policy information.
[0011] In some embodiments, the training of the preset policy amount prediction model based on the reference claim data, the predicted policy amount, and the policy amount accuracy to obtain a target policy amount prediction model includes:
[0012] Constructing a loss value based on the reference claim data and the predicted policy amount to obtain a first loss value;
[0013] Constructing a loss value according to the accuracy of the policy amount to obtain a second loss value;
[0014] Performing aggregation calculation based on the first loss value and the second loss value to obtain a generator loss value;
[0015] The parameters of the preset policy amount prediction model are optimized according to the generator loss value to obtain the target policy amount prediction model.
[0016] In some embodiments, after optimizing the parameters of the preset policy amount prediction model according to the generator loss value to obtain the target policy amount prediction model, the method further includes:
[0017] The target policy amount prediction model is used to predict the policy claim amount based on the reference object information and the reference policy information to obtain the target policy amount;
[0018] Constructing a loss value based on the reference object information, the reference policy information, the reference claim data, and the target policy amount to obtain a discriminator loss value;
[0019] The preset policy amount discrimination model is optimized according to the discriminator loss value to obtain a target policy amount discrimination model.
[0020] In some embodiments, constructing a loss value based on the reference object information, the reference policy information, the reference claim data, and the target policy amount to obtain a discriminator loss value includes:
[0021] Using the preset policy amount discrimination model, performing accuracy discrimination on the target policy amount according to the reference object information, the reference policy information, and the reference claims data to obtain a first accuracy;
[0022] Using the preset policy amount discrimination model, performing accuracy discrimination on the reference claim data according to the reference object information, the reference policy information, and the reference claim data to obtain a second accuracy;
[0023] A binary cross entropy is calculated based on the first accuracy and the second accuracy to obtain the discriminator loss value.
[0024] In some embodiments, after optimizing the preset policy amount discrimination model according to the discriminator loss value to obtain a target policy amount discrimination model, the method further includes:
[0025] performing accuracy judgment on the target policy amount according to the reference object information, the reference policy information, and the reference claims data using the target policy amount judgment model to obtain a third accuracy;
[0026] Parameters of the target policy amount prediction model are optimized according to the target policy amount, the reference claims data and the third accuracy.
[0027] In some embodiments, predicting the policy claim amount of the reference object information and the reference policy information using a preset policy amount prediction model to obtain the predicted policy amount includes:
[0028] Encoding the reference policy information to obtain a reference policy code;
[0029] Classifying the reference policy codes according to the reference object information to obtain a classified policy code for each reference object information;
[0030] For each reference object information, sort the classified policy codes according to the timestamp of the reference policy information to obtain the position identifiers of the classified policy codes, and concatenate the classified policy codes according to the position identifiers to obtain a combined policy code;
[0031] For each reference object information, the policy claim amount is predicted using the preset policy amount prediction model, the reference object information and the combined policy code to obtain the predicted policy amount.
[0032] In some embodiments, the preset policy amount prediction model includes a self-attention network and a prediction network; and predicting the policy claim amount using the preset policy amount prediction model, the reference object information, and the combined policy code to obtain the predicted policy amount includes:
[0033] Calculating the contribution of the joint policy code through the self-attention network to obtain a policy weight value;
[0034] Concatenate the policy weight value and the combined policy code to obtain a target policy code;
[0035] The claim amount is predicted using the prediction network, the target policy code and the reference object information to obtain the predicted policy amount.
[0036] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application provides a device for predicting insurance claim amounts, the device comprising:
[0037] A first acquisition module is used to obtain reference object information, reference policy information and reference claim data;
[0038] A first prediction module is configured to predict the insurance claim amount of the reference object information and the reference policy information using a preset insurance claim amount prediction model to obtain a predicted insurance claim amount;
[0039] an accuracy determination module, configured to perform an accuracy determination on the reference object information, the reference policy information, and the reference claims data using a preset policy amount determination model to obtain a policy amount prediction accuracy of the predicted policy amount; wherein the policy amount accuracy is used to represent the probability that the predicted policy amount is the same as the reference claims data;
[0040] A model training module is used to train the preset policy amount prediction model based on the reference claim data, the predicted policy amount and the policy amount accuracy to obtain a target policy amount prediction model;
[0041] The second acquisition module is used to obtain target object information and target policy information;
[0042] The second prediction module is used to predict the insurance policy claim amount for the target object information and the target insurance policy information through the target insurance policy amount prediction model.
[0043] To achieve the above-mentioned purpose, the 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 the processor implements the method described in the first aspect when executing the computer program.
[0044] To achieve the above-mentioned purpose, the fourth aspect of the embodiments of the present application proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method described in the first aspect.
[0045] The present application proposes a method and device for predicting the policy claim amount, an electronic device, and a storage medium. The method obtains reference object information, reference policy information, and reference claim data, and predicts the policy claim amount for the reference object information and reference policy information based on a preset policy amount prediction model. The accuracy of the prediction result is judged by a preset policy amount discrimination model to obtain a policy amount prediction accuracy that characterizes the probability that the predicted policy amount is consistent with the actual claim data. Furthermore, the policy amount prediction model is trained in combination with the reference claim data, the predicted policy amount, and the prediction accuracy to obtain a target policy amount prediction model that can improve the prediction accuracy. The model is then applied to the target object information and the target policy information to achieve accurate prediction of the policy claim amount. In this way, the embodiment of the present application improves the prediction accuracy of the prediction result by introducing the discrimination of prediction accuracy and the optimization of the model, solving the problem of inaccurate policy claim amounts caused by relying on manual experience settings. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] Figure 1 This is a flow chart of the insurance policy claim amount prediction method provided in an embodiment of the present application;
[0047] Figure 2 yes Figure 1 Flowchart of step S102 in FIG.
[0048] Figure 3 yes Figure 2 Flowchart of step S204 in FIG.
[0049] Figure 4 yes Figure 1 Flowchart of step S104 in FIG.
[0050] Figure 5 This is a flow chart of a method for predicting insurance claim amounts provided by another embodiment of the present application;
[0051] Figure 6 yes Figure 5 Flowchart of step S502 in FIG.
[0052] Figure 7 This is a flow chart of a method for predicting insurance claim amounts provided by another embodiment of the present application;
[0053] Figure 8 This is a schematic diagram of the structure of the insurance policy claim amount prediction device provided in an embodiment of the present application;
[0054] Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0055] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.
[0056] It should be noted that although the device schematics illustrate functional module divisions and the flowcharts illustrate logical sequences, in certain circumstances, the steps shown or described may be performed in a sequence that differs from the module divisions in the device or the sequence in the flowcharts. The terms "first," "second," and so on, in the specification, claims, and drawings, are used to distinguish similar items and are not necessarily used to describe a specific sequence or precedence.
[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.
[0058] First, let’s analyze some of the terms used in this application:
[0059] Discriminator: The discriminator is one of the components of the Generative Adversarial Network (GAN). It is used to identify and classify the source of input samples and is an important component of the generative model structure in artificial intelligence. The discriminator is usually composed of a multi-layer neural network. Its function is to distinguish between samples from real datasets and samples output by the generator, and output corresponding discrimination results to measure the quality of the data generated by the generator. During the training process, the discriminator and the generator are iteratively optimized in an adversarial manner. The discriminator improves its ability to identify sample features, which prompts the generator to continuously optimize its generation strategy, thereby improving the convergence and generation effect of the overall model. The discriminator is widely used in artificial intelligence sub-fields such as image generation, speech synthesis, data augmentation, and style transfer. It is an important technical unit for building high-quality generative models, achieving data distribution approximation, and restoring sample structure.
[0060] Generator: A core module in a generative adversarial network (GAN), the generator is used to generate synthetic samples with specific distributional characteristics within an unsupervised or semi-supervised learning framework. It is a crucial component of generative modeling techniques in artificial intelligence. The generator typically consists of a multi-layer neural network, taking random noise or latent variables as input and generating output samples with the target data distribution characteristics through a series of nonlinear mappings. The generator and discriminator collaborate to form an adversarial training architecture. During training, the generator continuously optimizes its generation strategy, gradually aligning the generated samples with the distribution of real samples in the feature space, thereby improving the overall model's fitting ability and output quality. Generators are widely used in technical fields such as image generation, speech synthesis, text generation, image super-resolution reconstruction, and sample data augmentation. They are a key technical approach to addressing sample scarcity and improving the diversity and authenticity of generated content. The continuous evolution of generator design and training methods has driven the rapid development and widespread application of generative models in areas such as content generation, style transfer, virtual reality, and digital content production.
[0061] Binary cross-entropy: Binary cross-entropy is a loss function used to evaluate the difference between the predictions of a binary classification model and the true labels. It is a fundamental evaluation method used in supervised learning and deep learning algorithms in artificial intelligence. Based on cross-entropy from information theory, this loss function is optimized for cases where the class labels only have two values (usually 0 and 1). It measures the closeness of the probability distribution between the model output and the target label. Binary cross-entropy is commonly used in technical fields such as image classification, speech recognition, text sentiment analysis, and generative adversarial network training. During training, it minimizes the logarithmic error between the predicted probabilities and the true labels, guiding the continuous update of model parameters to improve classification accuracy. Its mathematical expression is the weighted sum of the negative logarithms of the prediction confidence for each sample in both classes, and it exhibits excellent numerical stability and optimization properties. Binary cross-entropy is not only applicable to neural network models whose output is probabilities, but is also a standard loss function used in the training of models such as logistic regression, convolutional neural networks (CNNs), and generative discriminant networks. It is an important foundational tool for achieving model accuracy convergence and performance improvement in binary classification tasks.
[0062] The policy claim amount refers to the amount payable to the insured after an accident. For example, in the event of a traffic accident, if the vehicle repair cost is 80,000 yuan and the policy's maximum claim amount is greater than 80,000 yuan, the claim amount will be 80,000 yuan. When filling in the policy's maximum claim amount, staff typically estimate and set the claim amount based on experience, which can easily lead to inaccurate claims. Therefore, how to accurately determine insurance claim amounts has become a pressing technical issue.
[0063] Based on this, the embodiments of the present application provide a method and device for predicting the insurance claim amount, an electronic device and a storage medium, which aim to accurately determine the insurance claim amount.
[0064] The embodiments of the present application provide a method and device for predicting the insurance policy claim amount, an electronic device, and a storage medium, which are specifically illustrated through the following embodiments. First, the method for predicting the insurance policy claim amount in the embodiments of the present application is described.
[0065] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. 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 achieve optimal results.
[0066] Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.
[0067] The insurance claim amount prediction method provided in the embodiment of the present application relates to the technical field of insurance amount prediction and is applicable to the financial field. The insurance claim amount prediction method provided in the embodiment of the present application can be applied in a terminal, can be applied in a server side, or can be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or as a server cluster or distributed system composed of multiple physical servers, or as 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 insurance claim amount prediction method, etc., but is not limited to the above forms.
[0068] 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, and the like. 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, and the like 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 via a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.
[0069] It should be noted that in each specific embodiment of the present application, when it comes to the need to perform relevant processing based on data related to the user's 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 such 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.
[0070] Figure 1 This is an optional flowchart of the insurance claim amount prediction method provided in the embodiment of the present application. Figure 1 The method may include but is not limited to steps S101 to S106.
[0071] Step S101, obtaining reference object information, reference policy information and reference claim data;
[0072] Step S102, predicting the policy claim amount based on the reference object information and the reference policy information using a preset policy amount prediction model to obtain a predicted policy amount;
[0073] Step S103: performing accuracy discrimination on the reference object information, reference policy information, and reference claims data using a preset policy amount discrimination model to obtain a policy amount prediction accuracy for predicting the policy amount; wherein the policy amount accuracy is used to represent the probability that the predicted policy amount is the same as the reference claims data;
[0074] Step S104: training a preset policy amount prediction model based on reference claim data, predicted policy amount, and policy amount accuracy to obtain a target policy amount prediction model;
[0075] Step S105, obtaining target object information and target policy information;
[0076] Step S106: predicting the insurance claim amount based on the target object information and the target insurance policy information using the target insurance claim amount prediction model.
[0077] In the steps S101 to S106 shown in the embodiment of the present application, by obtaining reference object information, reference policy information and reference claim data, the reference object information and reference policy information are predicted based on the preset policy amount prediction model, and the accuracy of the prediction result is judged by the preset policy amount discrimination model, thereby obtaining the policy amount prediction accuracy used to characterize the probability that the predicted policy amount is consistent with the actual claim data; further, the policy amount prediction model is trained in combination with the reference claim data, the predicted policy amount and the prediction accuracy to obtain a target policy amount prediction model that can improve the prediction accuracy, and then applied to the target object information and the target policy information to achieve accurate prediction of the policy claim amount. In this way, the embodiment of the present application improves the prediction accuracy of the prediction result by introducing the discrimination of prediction accuracy and the optimization of the model, and solves the problem of inaccurate policy claim amount caused by relying on manual experience setting.
[0078] In step S101 of some embodiments, the reference object information is used to characterize the static attributes and dynamic behavior data of the historical insured object, which may specifically include information such as vehicle type, vehicle age, owner's past accident record, and vehicle usage nature (such as operating vehicle or non-operating vehicle).
[0079] The reference policy information is used to represent the policy content of the historical insured object, which may include the policy effective time, insurance type, insurance amount limit, deductible agreement clauses, etc.
[0080] Reference claims data is used to characterize the actual claims situation of the reference object during the historical policy period, specifically the claim amount.
[0081] For example, the reference object is a historically insured vehicle A. The reference object information might include that vehicle A is a five-year-old mid-size sedan, non-commercial, with a 40-year-old owner and a minor collision within the past two years. The corresponding reference policy information includes vehicle loss insurance and third-party liability insurance purchased by vehicle A in a specific year, with a coverage of 100,000 yuan for vehicle loss insurance and 500,000 yuan for third-party liability insurance. The reference claim data includes a single claim for 55,000 yuan due to a traffic accident involving vehicle A during the policy period.
[0082] See also Figure 2 In some embodiments, step S102 may include but is not limited to steps S201 to S204:
[0083] Step S201, encoding the reference policy information to obtain a reference policy code;
[0084] Step S202: Classify the reference policy codes according to the reference object information to obtain the classified policy codes for each reference object information;
[0085] Step S203: For each reference object information, sort the classified policy codes according to the timestamp of the reference policy information to obtain the position identifiers of the classified policy codes, and concatenate the classified policy codes according to the position identifiers to obtain a combined policy code;
[0086] Step S204: For each reference object information, the policy claim amount is predicted by using the preset policy amount prediction model, the reference object information and the combined policy code to obtain the predicted policy amount.
[0087] In the steps S201 to S204 shown in the embodiment of the present application, a structured reference policy code is generated by encoding the reference policy information; further, the reference policy code is classified according to the reference object information to obtain the classified policy code corresponding to each reference object information, and the classified policy code is sorted in combination with the timestamp of the policy information, thereby constructing a joint policy code reflecting the evolution path of the policy; finally, based on the preset policy amount prediction model, the reference object information and the joint policy code are used as input to complete the prediction of the policy claim amount and obtain the predicted policy amount. In this way, the present embodiment realizes the time series modeling and semantic integration of historical policy data of the same reference object, introduces the dynamic context information of the insurance behavior in the process of predicting the policy claim amount, and improves the prediction model's perception of the insured's behavior trend and risk changes, so that the generated predicted policy amount is closer to the actual claim result, thereby enhancing the accuracy of the prediction.
[0088] In step S201 of some embodiments, the reference policy code is used to convert the key fields in the reference policy information into a numerical representation with a unified format. The reference policy code generally includes, but is not limited to, structured organization and identification coding of elements such as the policy clause type, insurance liability scope, insured amount value, deductible amount, and policy validity period. For example, for the insured amount field, it can be divided into discrete levels according to a preset interval and assigned corresponding codes, such as 0 to 50,000 yuan is coded "1" and 50,000 to 100,000 yuan is coded "2"; for the liability scope field, a unique hot encoding method can be used to encode different types of insurance such as collision insurance, third-party liability insurance, and theft insurance.
[0089] In step S202 of some embodiments, classifying the reference policy codes according to the reference object information means taking each reference object as the classification basis and aggregating all the reference policy codes corresponding to the reference object.
[0090] For example, if the reference object information includes two historically insured objects, Vehicle B and Vehicle C, where Vehicle B has been insured under three policies in the past three years, and the corresponding reference policy codes are [1,0,3], [1,1,2], and [0,1,3], respectively; and Vehicle C has been insured under two policies in the past two years, and the corresponding reference policy codes are [2,0,1] and [2,1,1], then in step S202, after classification based on the reference object information, the classified policy codes corresponding to Vehicle B are [[1,0,3], [1,1,2], [0,1,3]], and the classified policy codes corresponding to Vehicle C are [[2,0,1], [2,1,1]], i.e., the corresponding policy codes of Vehicles B and C are aggregated and classified.
[0091] In step S203 of some embodiments, sorting the classified policy codes according to the timestamp of the reference policy information means arranging the multiple classified policy codes corresponding to the reference object information in an orderly manner based on the time sequence, thereby obtaining the relative position of each policy code in the time dimension, and then obtaining the position identifier of the classified policy code. Subsequently, the classified policy codes are spliced according to the position identifier, that is, the classified policy codes are sequentially connected in chronological order to form a complete joint policy code. The joint policy code represents the policy evolution process of a certain reference object information in different periods in the form of a sequence structure, so that the model can simultaneously perceive the content of the policy terms and their evolution trend in the subsequent claim amount prediction.
[0092] The reason for sorting insurance policies is that the ordered splicing of policy timestamps enhances the temporal expression capability of policy data, enabling the model to capture the potential correlation between policy evolution and claim amounts when receiving joint policy codes, thereby improving prediction accuracy.
[0093] See also Figure 3 In some embodiments, the preset policy amount prediction model includes a self-attention network and a prediction network. Step S204 may include but is not limited to steps S301 to S303:
[0094] Step S301: Calculate the contribution of the joint policy code through the self-attention network to obtain the policy weight value;
[0095] Step S302: Concatenate the policy weight value and the combined policy code to obtain a target policy code;
[0096] Step S303: predict the claim amount using the prediction network, target policy code and reference object information to obtain the predicted policy amount.
[0097] In steps S301 to S303 shown in the embodiment of the present application, the contribution of the joint policy code is calculated through the self-attention network to generate a policy weight value that can characterize the strength of the correlation between various types of policies and the target claim situation; further, the policy weight value is spliced with the joint policy code to form a target policy code that comprehensively considers the policy structure characteristics and the importance assessment results; finally, based on the prediction network, the target policy code and the reference object information are input together to predict the claim amount to obtain the predicted policy amount. In this way, the embodiment of the present application realizes the quantitative distinction of the influence of multi-time series policy information by introducing the self-attention mechanism, effectively improving the prediction network's ability to identify key policy features, so that the generated predicted policy amount can better reflect the actual claim risk level of the insured object, and enhance the accuracy of the prediction results.
[0098] In step S301 of some embodiments, the self-attention network is used to identify the importance and weight the individual policy fragments in the joint policy code. The self-attention network can model the mutual influence relationship between elements at different positions in the input sequence by constructing an attention mechanism between the query vector, the key vector, and the value vector, thereby determining the actual contribution of each policy fragment to the claim amount prediction in the overall insurance behavior. The self-attention network can be composed of a multi-head attention module in the Transformer structure, or a single-layer structure based on Scaled Dot-Product Attention, and this application does not impose any specific restrictions.
[0099] The contribution calculation is implemented as follows: the joint policy code is sent as the input sequence into the self-attention network, the importance distribution of each code segment relative to the overall sequence is obtained, and then the attention score corresponding to each code segment is output, that is, the policy weight value.
[0100] The numerical form of the policy weight value is typically a real number vector with the same length as the number of combined policy code segments, ranging from 0 to 1, and ensuring that the sum of all segment weights is 1. It reflects the influence of each policy segment on the predicted policy amount in the current prediction task. The higher the weight value, the more important the policy segment is to the final prediction result, and vice versa. This mechanism enables the model to automatically identify and emphasize historical policy information that has a critical impact on the target claim risk, thereby improving the targeted and accurate prediction.
[0101] In step S302 of some embodiments, concatenating the policy weight value with the combined policy code means that, based on the combined policy code, the respective policy weight values are connected as additional dimensions to the original coding features in the order of the corresponding policy segments, thereby forming a target policy code that includes both structural and weight features. The target policy code retains the content information of each policy segment in the original combined policy code and also introduces weight information indicating the importance of each segment, thereby enhancing the discriminative ability of the code in subsequent predictions, enabling the prediction model to more accurately focus on the effective features carried by key policy segments, thereby improving the expression efficiency and information utilization of claim amount predictions.
[0102] In step S303 of some embodiments, the prediction network may use a fully connected neural network, a residual structure network, or other modeling methods with nonlinear fitting capabilities to perform a fusion calculation on the target policy code and the reference object information to output the predicted policy amount.
[0103] In step S103 of some embodiments,
[0104] See also Figure 4 In some embodiments, step S104 may include but is not limited to steps S401 to S404:
[0105] Step S401, constructing a loss value based on reference claim data and predicted policy amount to obtain a first loss value;
[0106] Step S402: constructing a loss value based on the accuracy of the policy amount to obtain a second loss value;
[0107] Step S403, performing aggregation calculation based on the first loss value and the second loss value to obtain a generator loss value;
[0108] Step S404: Optimize the parameters of the preset policy amount prediction model according to the generator loss value to obtain a target policy amount prediction model.
[0109] In the steps S401 to S404 shown in the embodiment of the present application, a first loss value is constructed by referring to the difference between the claim data and the predicted policy amount, which is used to reflect the degree of numerical deviation between the predicted policy amount and the actual claim result; further, a second loss value is constructed according to the accuracy of the policy amount, which is used to measure the probability performance of the predicted policy amount being consistent with the true value; then, an aggregation calculation is performed based on the first loss value and the second loss value to obtain a generator loss value that comprehensively reflects the numerical error and probability consistency; finally, the parameters of the preset policy amount prediction model are optimized according to the generator loss value to obtain a target policy amount prediction model. In this way, the embodiment of the present application integrates direct numerical deviation and probability accuracy evaluation to enable the target policy amount prediction model to have a higher prediction accuracy.
[0110] In some embodiments, step S401 may construct the first loss value using a mean squared error method. Specifically, the squared difference between the predicted policy amount and reference claims data is averaged to form a loss metric, quantifying the degree of deviation between the predicted policy amount and the actual claim amount. In other embodiments, other robust loss functions such as mean absolute error or Huber loss may be used instead.
[0111] In step S402 of some embodiments, the second loss value can be constructed using a mean square error method, where the difference between the policy amount accuracy and the ideal accuracy (i.e., the policy amount accuracy is used to characterize the probability that the predicted policy amount is the same as the reference claim data is 100%) is squared and averaged to serve as a loss indicator for measuring the credibility deviation of the prediction result.
[0112] Specifically, the ideal accuracy can be set to 1 (that is, the target probability that the predicted policy amount is completely consistent with the reference claims data). The policy amount accuracy is the probability value output by the model. The smaller the difference between the two, the closer the prediction is to the ideal state.
[0113] In step S403 of some embodiments, the aggregate calculation of the first loss value and the second loss value can be implemented by weighted summation. Specifically, a set of weight parameters is set to assign certain weight coefficients to the first loss value and the second loss value respectively. The calculation formula can be expressed as: generator loss value = α × first loss value + β × second loss value, where α and β are pre-set or dynamically adjusted hyperparameters.
[0114] See also Figure 5 In some embodiments, after step S404, the following steps may also include but are not limited to steps S501 to S503:
[0115] Step S501: using a target policy amount prediction model, the policy claim amount is predicted based on the reference object information and the reference policy information to obtain the target policy amount;
[0116] Step S502 , constructing a loss value based on the reference object information, reference policy information, reference claim data, and target policy amount to obtain a discriminator loss value;
[0117] Step S503: Optimize the preset policy amount discrimination model according to the discriminator loss value to obtain a target policy amount discrimination model.
[0118] In steps S501 to S503 shown in the embodiment of the present application, the target policy amount prediction model is used to predict the policy claim amount based on the reference object information and the reference policy information to obtain the target policy amount; further, a discriminator loss value is constructed based on the reference object information, the reference policy information, the reference claim data and the target policy amount to quantify the degree of match between the current prediction result and the actual claim data; finally, the parameters of the preset policy amount discrimination model are optimized based on the discriminator loss value to form a target policy amount discrimination model with stronger recognition ability. In this way, the embodiment of the present application enables the policy amount discrimination model to learn and adjust based on the output of the prediction model, thereby effectively improving the consistency recognition ability between the prediction result and the actual claim data.
[0119] In step S501 of some embodiments, reference object information and reference policy information are input into a target policy amount prediction model to obtain a target policy amount.
[0120] See also Figure 6 In some embodiments, step S502 includes but is not limited to steps S601 to S603:
[0121] Step S601: Using a preset policy amount discrimination model, the target policy amount is subjected to accuracy discrimination based on reference object information, reference policy information, and reference claim data to obtain a first accuracy.
[0122] Step S602: Using a preset policy amount discrimination model, the reference claim data is subjected to accuracy discrimination based on the reference object information, the reference policy information, and the reference claim data to obtain a second accuracy.
[0123] Step S603: performing binary cross entropy calculation based on the first accuracy and the second accuracy to obtain a discriminator loss value.
[0124] In the steps S601 to S603 shown in the embodiment of the present application, the target policy amount is subjected to accuracy judgment according to the reference object information, reference policy information and reference claim data through the preset policy amount judgment model, and a first accuracy is obtained, which is used to measure the rationality of the target policy amount under the existing reference data conditions, that is, whether the judgment model can identify that the target policy amount is not derived from a real claim behavior; at the same time, the reference claim data itself is subjected to accuracy judgment under the same reference object information, reference policy information and reference claim data conditions through the same preset policy amount judgment model, and a second accuracy is obtained, which is used to characterize the matching strength presented by the model when facing real claim data, that is, the recognition ability of the judgment model for real data. Subsequently, the first accuracy and the second accuracy are used as the output results in the binary classification judgment scenario, and the error is calculated based on the binary cross entropy function to obtain the discriminator loss value used to train the optimized discrimination model.
[0125] Thus, the present embodiment constructs a dual-path discrimination mechanism, enabling the preset policy amount discrimination model to receive two sets of inputs: first, reference object information, reference policy information, and reference claims data, combined with a target policy amount, to determine whether the target policy amount has the characteristics of real claims data; and second, reference object information, reference policy information, and reference claims data, combined with the reference claims data itself, to determine whether the reference claims data is indeed real historical data. A loss function is constructed based on the output values of these two sets of inputs, enabling the preset policy amount discrimination model to better distinguish between non-real and real values.
[0126] In step S601 of some embodiments, reference object information, reference policy information, and reference claim data are input into a preset policy amount discrimination model to obtain a first accuracy.
[0127] In some embodiments, the principle of step S602 is the same as that of step S601 and is not repeated here.
[0128] In step S603 of some embodiments, the first and second accuracies are used as prediction results in a binary classification task, and a loss value is constructed using binary cross-entropy. Specifically, the first accuracy represents the probability output of the predicted policy amount being identified as "not true" under the reference data conditions, and the second accuracy represents the probability output of the reference claims data being identified as "true" under the same conditions. The binary cross-entropy uses the true label as the standard and quantifies the deviation between the predicted output and the ideal label as the discriminator loss value.
[0129] See also Figure 7 In some embodiments, after step S603, the following steps may also include but are not limited to steps S701 to S702:
[0130] Step S701: performing accuracy judgment on the target policy amount using a target policy amount judgment model based on reference object information, reference policy information, and reference claims data to obtain a third accuracy.
[0131] Step S702: Optimize the parameters of the target policy amount prediction model according to the target policy amount, reference claim data, and the third accuracy.
[0132] In steps S701 to S702, the target policy amount discrimination model is used to discriminate the accuracy of the target policy amount based on reference object information, reference policy information, and reference claims data, thereby generating a third accuracy used to characterize the consistency between the predicted policy amount and the actual claims data. Furthermore, the target policy amount, reference claims data, and the third accuracy are combined as a feedback basis to optimize the parameters of the target policy amount prediction model. In this way, the embodiment of the present application continuously improves the prediction accuracy of the target policy amount prediction model by constructing an iterative optimization mechanism for the prediction model based on discrimination feedback.
[0133] In some embodiments, the principle of step S701 is the same as that of step S601 and will not be repeated here.
[0134] In some embodiments, the principle of step S702 is the same as that of step S104 and will not be repeated here.
[0135] See also Figure 8 The present application also provides an insurance claim amount prediction device that can implement the above-mentioned insurance claim amount prediction method. The device includes:
[0136] The first acquisition module 801 is used to acquire reference object information, reference policy information and reference claim data;
[0137] The first prediction module 802 is configured to predict the insurance claim amount based on the reference object information and the reference policy information using a preset insurance claim amount prediction model to obtain a predicted insurance claim amount.
[0138] The accuracy determination module 803 is configured to perform an accuracy determination on the reference object information, reference policy information, and reference claims data using a preset policy amount determination model to obtain a policy amount prediction accuracy for the predicted policy amount; wherein the policy amount accuracy is used to indicate the probability that the predicted policy amount is the same as the reference claims data;
[0139] The model training module 804 is used to train the preset policy amount prediction model based on the reference claim data, the predicted policy amount, and the policy amount accuracy to obtain a target policy amount prediction model;
[0140] The second acquisition module 805 is used to acquire target object information and target policy information;
[0141] The second prediction module 806 is used to predict the insurance policy claim amount based on the target object information and the target insurance policy information using the target insurance policy amount prediction model.
[0142] The specific implementation of the insurance policy claim amount prediction device is basically the same as the specific embodiment of the above-mentioned insurance policy claim amount prediction method, and will not be repeated here.
[0143] The present application also provides an electronic device comprising a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned insurance claim amount prediction method. The electronic device can be any smart terminal, including a tablet computer and an in-vehicle computer.
[0144] See also Figure 9 , Figure 9 The hardware structure of an electronic device according to another embodiment is shown. The electronic device includes:
[0145] The processor 901 can be implemented as a general-purpose CPU (Central Processing Unit), a microprocessor, an 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;
[0146] The memory 902 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 902 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and is called by the processor 901 to execute the insurance policy claim amount prediction method of the embodiments of this application;
[0147] Input / output interface 903, used to implement information input and output;
[0148] Communication interface 904, used to implement communication interaction between this device and other devices, which can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WiFi, Bluetooth, etc.);
[0149] Bus 905 , which transmits information between various components of the device (e.g., processor 901 , memory 902 , input / output interface 903 , and communication interface 904 );
[0150] The processor 901 , the memory 902 , the input / output interface 903 and the communication interface 904 are connected to each other in communication within the device via a bus 905 .
[0151] 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 insurance policy claim amount prediction method.
[0152] 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 arranged 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.
[0153] The embodiment of the present application provides a method for predicting the insurance claim amount, an insurance claim amount prediction device, an electronic device, and a storage medium. The method obtains reference object information, reference policy information, and reference claim data, predicts the insurance claim amount for the reference object information and reference policy information based on a preset insurance claim amount prediction model, and discriminates the accuracy of the prediction result through a preset insurance claim amount discrimination model, thereby obtaining an insurance claim amount prediction accuracy used to characterize the probability that the predicted insurance claim amount is consistent with the actual claim data; further, the insurance claim amount prediction model is trained in combination with the reference claim data, the predicted insurance policy amount, and the prediction accuracy to obtain a target insurance policy amount prediction model that can improve the prediction accuracy, and then applies it to the target object information and the target insurance policy information to achieve accurate prediction of the insurance claim amount. In this way, the embodiment of the present application improves the prediction accuracy of the prediction result by introducing the discrimination of prediction accuracy and the optimization of the model, and solves the problem of inaccurate insurance claim amounts caused by relying on manual experience settings.
[0154] The embodiments described in the embodiments of this application are intended to more clearly illustrate the technical solutions of the embodiments of this application and do not constitute a limitation on the technical solutions provided by the embodiments of this 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 this application are also applicable to similar technical problems.
[0155] Those skilled in the art will understand 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.
[0156] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.
[0157] Those skilled in the art will appreciate that all or some of the steps in the methods, systems, and functional modules / units in the devices disclosed above may be implemented as software, firmware, hardware, or appropriate combinations thereof.
[0158] 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 sequential order. 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 variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes 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.
[0159] It should be understood that in this 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 previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items 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.
[0160] In the several embodiments provided in this 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 merely schematic. For example, the division of the above-mentioned units is only a logical function division. In actual implementation, there may be other division methods, 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.
[0161] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0162] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.
[0163] 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 is essentially 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. The computer software product is stored in a storage medium and includes multiple instructions for enabling a computer device (which can be a personal computer, server, or 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.
[0164] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and essence of the present invention should be within the scope of the present invention.
Claims
1. A method for predicting insurance claim amounts, characterized in that: The method comprises: Obtain reference object information, reference policy information, and reference claims data; Predicting the insurance claim amount of the reference object information and the reference policy information using a preset insurance claim amount prediction model to obtain a predicted insurance claim amount; performing accuracy discrimination on the reference object information, the reference policy information, and the reference claims data using a preset policy amount discrimination model to obtain a policy amount prediction accuracy of the predicted policy amount; wherein the policy amount accuracy is used to represent the probability that the predicted policy amount is the same as the reference claims data; Training the preset policy amount prediction model based on the reference claim data, the predicted policy amount, and the policy amount accuracy to obtain a target policy amount prediction model; Obtain target object information and target policy information; The target policy amount prediction model is used to predict the policy claim amount for the target object information and the target policy information.
2. The method according to claim 1, characterized in that The step of training the preset policy amount prediction model based on the reference claim data, the predicted policy amount, and the policy amount accuracy to obtain a target policy amount prediction model includes: Constructing a loss value based on the reference claim data and the predicted policy amount to obtain a first loss value; Constructing a loss value according to the accuracy of the policy amount to obtain a second loss value; Performing aggregation calculation based on the first loss value and the second loss value to obtain a generator loss value; The parameters of the preset policy amount prediction model are optimized according to the generator loss value to obtain the target policy amount prediction model.
3. The method according to claim 2, characterized in that After optimizing the parameters of the preset policy amount prediction model according to the generator loss value to obtain the target policy amount prediction model, the method further includes: The target policy amount prediction model is used to predict the policy claim amount based on the reference object information and the reference policy information to obtain the target policy amount; Constructing a loss value based on the reference object information, the reference policy information, the reference claim data, and the target policy amount to obtain a discriminator loss value; The preset policy amount discrimination model is optimized according to the discriminator loss value to obtain a target policy amount discrimination model.
4. The method according to claim 3, characterized in that The constructing of a loss value based on the reference object information, the reference policy information, the reference claim data, and the target policy amount to obtain a discriminator loss value includes: Using the preset policy amount discrimination model, performing accuracy discrimination on the target policy amount according to the reference object information, the reference policy information, and the reference claims data to obtain a first accuracy; Using the preset policy amount discrimination model, performing accuracy discrimination on the reference claim data according to the reference object information, the reference policy information, and the reference claim data to obtain a second accuracy; A binary cross entropy is calculated based on the first accuracy and the second accuracy to obtain the discriminator loss value.
5. The method according to claim 3, characterized in that After optimizing the preset policy amount discrimination model according to the discriminator loss value to obtain a target policy amount discrimination model, the method further includes: performing accuracy judgment on the target policy amount according to the reference object information, the reference policy information, and the reference claims data using the target policy amount judgment model to obtain a third accuracy; Parameters of the target policy amount prediction model are optimized according to the target policy amount, the reference claims data and the third accuracy.
6. The method according to any one of claims 1 to 5, characterized in that The step of predicting the insurance claim amount of the reference object information and the reference policy information using a preset insurance claim amount prediction model to obtain the predicted insurance claim amount includes: Encoding the reference policy information to obtain a reference policy code; Classifying the reference policy codes according to the reference object information to obtain a classified policy code for each reference object information; For each reference object information, sort the classified policy codes according to the timestamp of the reference policy information to obtain the position identifiers of the classified policy codes, and concatenate the classified policy codes according to the position identifiers to obtain a combined policy code; For each reference object information, the policy claim amount is predicted using the preset policy amount prediction model, the reference object information and the combined policy code to obtain the predicted policy amount.
7. The method according to claim 6, characterized in that The preset policy amount prediction model includes a self-attention network and a prediction network; the policy claim amount prediction is performed using the preset policy amount prediction model, the reference object information, and the combined policy code to obtain the predicted policy amount, including: Calculating the contribution of the joint policy code through the self-attention network to obtain a policy weight value; Concatenate the policy weight value and the combined policy code to obtain a target policy code; The claim amount is predicted using the prediction network, the target policy code and the reference object information to obtain the predicted policy amount.
8. A device for predicting insurance claim amounts, characterized in that: The device comprises: A first acquisition module is used to obtain reference object information, reference policy information and reference claim data; A first prediction module is configured to predict the insurance claim amount of the reference object information and the reference policy information using a preset insurance claim amount prediction model to obtain a predicted insurance claim amount; an accuracy determination module, configured to perform an accuracy determination on the reference object information, the reference policy information, and the reference claims data using a preset policy amount determination model to obtain a policy amount prediction accuracy of the predicted policy amount; wherein the policy amount accuracy is used to represent the probability that the predicted policy amount is the same as the reference claims data; A model training module is used to train the preset policy amount prediction model based on the reference claim data, the predicted policy amount and the policy amount accuracy to obtain a target policy amount prediction model; The second acquisition module is used to obtain target object information and target policy information; The second prediction module is used to predict the insurance policy claim amount for the target object information and the target insurance policy information through the target insurance policy amount prediction model.
9. 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 insurance policy claim amount prediction method according to any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for predicting the insurance policy claim amount according to any one of claims 1 to 7 is implemented.