Claim decision-making method and device, electronic equipment and storage medium

Abstract

The invention discloses a claim decision-making method and device, electronic equipment and a storage medium, and belongs to the technical field of financial science and technology. The method comprises steps that a plurality of historical enterprise claim settlement data are acquired, a training set is constructed based on the plurality of historical enterprise claim settlement data, and the historical enterprise claim settlement data comprise enterprise characteristic parameters and claim settlement results corresponding to the enterprise characteristic parameters; establishing an initial strategy network, and constructing a decision optimization factor based on the historical enterprise claim settlement data and the initial strategy network; based on the state factor, the action factor and the reward factor, updating network parameters of the initial policy network by adopting a policy gradient to obtain an optimized policy network; and outputting a claim decision result of the current enterprise based on the optimized strategy network. The technical problem that enterprise insurance claim decision making cannot be accurately carried out in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of financial technology, and in particular to a claims decision-making method, device, electronic device and storage medium in an insurance claims processing process under an applied financial scenario. Background Art

[0002] Enterprises face various risks in the course of operation. In order to ensure the stable operation of the enterprise and the well-being of employees, these risks can be transferred by purchasing insurance. For example, enterprises can purchase property insurance to protect corporate assets and purchase liability insurance to avoid potential legal liabilities.

[0003] There are many personalized factors in the field of corporate insurance, including that the insured belongs to the enterprise. Unlike individual auto insurance, the enterprise factors are relatively few but personalized, because the attributes and relationships involved in each enterprise are different, and the premium policy amount is high, so insurance companies generally "tailor-make" insurance policies for enterprises. Therefore, different types of insurance policies and other multi-dimensional information make traditional risk control models mostly binary classification models. Binary classification models face huge challenges in capturing these complex and highly personalized features, and the accuracy rate is difficult to meet the needs, maintaining at around 16%. Therefore, when the claim results are inaccurate, it may cause huge losses to the insurance company. Summary of the invention

[0004] The purpose of the present invention is to overcome the above-mentioned technical deficiencies and provide a claims decision-making method, device, electronic device and storage medium to solve the technical problem that the prior art cannot accurately make corporate insurance claims decisions.

[0005] In order to achieve the above technical objectives, the present invention adopts the following technical solutions:

[0006] In a first aspect, the present invention provides a claim decision-making method, comprising the following steps:

[0007] Acquire a number of historical enterprise claims data, and construct a training set based on the historical enterprise claims data, wherein the historical enterprise claims data includes enterprise characteristic parameters and claims results corresponding to the enterprise characteristic parameters;

[0008] Establishing an initial strategy network, and constructing a decision optimization factor based on the historical enterprise claims data and the initial strategy network, wherein the decision optimization factor includes a state factor for characterizing a state, an action factor for characterizing an action, and a reward factor for characterizing a reward;

[0009] Based on the state factor, action factor and reward factor, the network parameters of the initial policy network are updated using policy gradient to obtain an optimized policy network;

[0010] Based on the optimized strategy network, the current enterprise's claims decision result is output.

[0011] In some embodiments, the establishing of the initial strategy network, constructing a decision optimization factor based on the historical enterprise claims data and the initial strategy network, includes:

[0012] Establishing an initial strategy network based on a multi-layer perceptron model, and presetting network parameters of the initial strategy network;

[0013] Using the enterprise characteristic parameters in the historical enterprise claims data as state factors;

[0014] Based on the state factor, using the initial strategy network to predict the action corresponding to the state factor and use it as the action factor;

[0015] Based on the action factor, a preset reward value calculation model is adopted to obtain a corresponding reward value, and the reward value is used as the reward factor.

[0016] In some embodiments, the using of a preset reward value calculation model to obtain a corresponding reward value includes:

[0017] Obtaining the action factor, and determining the type of the action factor;

[0018] When the action factor is a claim settlement action, a first reward value is output; when the action factor is a claim rejection action, a second reward value is output, wherein the first reward value is a negative number and the second reward value is a positive number.

[0019] In some embodiments, the updating of the network parameters of the initial policy network using policy gradient based on the state factor, action factor and reward factor to obtain an optimized policy network includes:

[0020] Based on the state factor and the action factor, calculating the probability that the initial strategy network selects the action factor;

[0021] Based on the probability, calculating the gradient of the log probability of the initial policy network;

[0022] Calculating a policy gradient based on the gradient of the log probability of the initial policy network and the reward factor;

[0023] The network parameters of the initial policy network are updated using policy gradient to obtain an optimized policy network.

[0024] In some embodiments, the calculating, based on the state factor and the action factor, the probability that the initial policy network selects the action factor comprises:

[0025] Calculate the index value of each action based on the state factor and the action factor;

[0026] Based on the index value of each action, the probability of the initial policy network selecting the action factor is calculated.

[0027] In some embodiments, the adopting of policy gradient to update the network parameters of the initial policy network to obtain an optimized policy network includes:

[0028] Use the gradient ascent algorithm to maximize the policy gradient value;

[0029] Based on the maximized policy gradient value and the preset learning rate, the network parameters of the initial policy network are optimized until an optimized policy network is obtained.

[0030] In some embodiments, outputting the current enterprise's claim decision result based on the optimized policy network includes:

[0031] Get the enterprise characteristic parameters of the current enterprise;

[0032] The enterprise characteristic parameters of the current enterprise are input into the optimized strategy network to generate a claim decision result.

[0033] In a second aspect, the present invention further provides a claim decision-making device, comprising:

[0034] A data acquisition module, used to acquire a number of historical enterprise claims data, and construct a training set based on the historical enterprise claims data, wherein the historical enterprise claims data includes enterprise characteristic parameters and claims results corresponding to the enterprise characteristic parameters;

[0035] A decision optimization factor acquisition module, used to establish an initial strategy network, and construct a decision optimization factor based on the historical enterprise claims data and the initial strategy network, wherein the decision optimization factor includes a state factor for representing a state, an action factor for representing an action, and a reward factor for representing a reward;

[0036] An optimization module, configured to update the network parameters of the initial policy network using a policy gradient based on the state factor, the action factor, and the reward factor to obtain an optimized policy network;

[0037] The decision module is used to output the current enterprise's claim decision result based on the optimized strategy network.

[0038] In a third aspect, the present invention further provides a computer device, comprising a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the claim decision method as described above when executing the computer-readable instructions.

[0039] In a fourth aspect, the present invention further provides a computer-readable storage medium having computer-readable instructions stored thereon, and the computer-readable instructions, when executed by a processor, implement the steps of the claim decision method as described above.

[0040] Compared with the prior art, the claims decision method, device, equipment and storage medium provided by the present invention first obtain several historical enterprise claims data, build a training set based on the several historical enterprise claims data, wherein the historical enterprise claims data include enterprise characteristic parameters and claims results corresponding to the enterprise characteristic parameters, then establish an initial policy network, build a decision optimization factor based on the historical enterprise claims data and the initial policy network, wherein the decision optimization factor includes a state factor for characterizing the state, an action factor for characterizing the action and a reward factor for characterizing the reward, then based on the state factor, action factor and reward factor, use the policy gradient to update the network parameters of the initial policy network to obtain an optimized policy network, and finally output the current enterprise's claims decision result based on the optimized policy network. The present invention can automatically optimize the decision model without manual intervention, can process data, is relatively robust to noise, and greatly improves the accuracy of decision making. In addition, it can also process and give answers when the input information is incomplete, thereby enhancing the performance of the decision model. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the scheme in the present invention, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 1 is an exemplary system architecture diagram in which the present invention can be applied;

[0043] Figure 2 is a flow chart of an embodiment of a claim decision-making method according to the present invention;

[0044] Figure 3 yes Figure 2 A flowchart of a specific embodiment of step S200 is shown;

[0045] Figure 4 yes Figure 2 A flowchart of a specific embodiment of step S300 is shown;

[0046] Figure 5 is a schematic diagram of the structure of an embodiment of a claim decision-making device according to the present invention;

[0047] Figure 6 FIG. 1 is a schematic structural diagram of an embodiment of a computer device according to the present invention. DETAILED DESCRIPTION

[0048] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present invention; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention; the terms "including" and "having" and any variations thereof in the specification and claims of the present invention and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present invention or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0049] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0050] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings.

[0051] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0052] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0053] The terminal devices 101, 102, and 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV) players, laptop computers, desktop computers, etc.

[0054] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .

[0055] It should be noted that the claim decision method provided in the embodiment of the present invention is generally executed by a server / terminal device, and accordingly, the claim decision device is generally arranged in the server / terminal device.

[0056] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to the implementation requirements.

[0057] Please continue reading Figure 2 , Figure 2 A flow chart of an embodiment of a method for making a claim settlement decision according to the present invention is shown, comprising steps S100 to S400.

[0058] S100. Acquire a number of historical enterprise claims data, and construct a training set based on the historical enterprise claims data, wherein the historical enterprise claims data include enterprise characteristic parameters and claims results corresponding to the enterprise characteristic parameters.

[0059] In this embodiment, the closed group enterprise claims data is used as training data, and the final claims results are marked, where the claims results are claims or denials. The enterprise feature parameters include insurance features and claims features, where the insurance parameters include insurance time, insured person, insured person address, legal person, enterprise credit code, employee list, etc.; the claims features include reporting time, reporter, inspection time, reporting reason and other data.

[0060] In order to increase the reliability of subsequent models, a validation set is also constructed when constructing a training set, wherein the training set is used for model training, and the validation set is used to evaluate the performance of the trained model, that is, to test the performance of the trained model. For example, taking corporate claims data as a training set and a validation set, a sample set containing a large amount of historical corporate claims data can be obtained from the historical corporate claims data set, and then the sample set can be divided into a validation set and a training set according to a preset allocation ratio. For example, the preset allocation ratio of the training set to the validation set in the sample set is 9:1, that is, 90% of the image samples in the sample set are used as the training set, and the remaining 10% of the image samples in the sample set are used as the validation set, then the obtained sample set is divided into a training set and a validation set according to a ratio of 9:1.

[0061] S200, establishing an initial strategy network, and constructing a decision optimization factor based on the historical enterprise claims data and the initial strategy network, wherein the decision optimization factor includes a state factor for characterizing a state, an action factor for characterizing an action, and a reward factor for characterizing a reward.

[0062] In this embodiment, the initial policy network is based on a multi-layer perceptron (MLP) model, which takes the characteristics of the case as input and outputs the decision of claim settlement or claim rejection. Among them, the MLP model, as a classic artificial neural network (ANN), occupies an important position in the field of deep learning. It can handle nonlinear complex relationships by constructing a network structure including an input layer, multiple hidden layers and an output layer, and is widely used in various tasks such as classification and regression.

[0063] Specifically, MLP consists of an input layer, one or more hidden layers, and an output layer. Each layer consists of several neurons that are connected to each other through weights. In MLP, information is transmitted from the input layer, and after weighting and biasing, nonlinearity is introduced through activation functions, and the result is finally obtained in the output layer. This process allows MLP to learn and simulate complex relationships between input and output.

[0064] S300: Based on the state factor, action factor and reward factor, the network parameters of the initial policy network are updated using policy gradient to obtain an optimized policy network.

[0065] In this embodiment, the policy gradient (PG) is a reinforcement learning algorithm that directly models and optimizes the policy to obtain the optimal policy. Unlike the traditional value function method, the policy gradient does not need to estimate the action value function, but directly learns the mapping from state to action through parameterized policy. This makes it more applicable in high-dimensional action space and complex environment. The policy gradient is an optimization method based on the gradient of the performance metric J(θ), and its purpose is to directly learn a parameterized policy π(a|s,θ) to select the optimal action under a given state. The core of the policy gradient is to maximize the cumulative reward by adjusting the policy parameter θ. Unlike the value function method, the policy gradient does not need to calculate the Q value or V value, but directly applies the gradient rise on the policy parameter. The policy gradient uses a function approximator (such as a deep neural network) to input the current state s and output the probability distribution of taking each action in this state. This policy is parameterized as θ, that is, π(a|s,θ). Given the policy parameter θ, the agent interacts with the environment to generate trajectories and calculates the total reward R for each trajectory. These samples are then used to estimate the gradient of the expected return and update the parameters θ using the policy gradient theorem.

[0066] S400: Based on the optimized strategy network, output the current enterprise's claim decision result.

[0067] In an embodiment of the present invention, a number of historical enterprise claims data are first obtained, and a training set is constructed based on the historical enterprise claims data, wherein the historical enterprise claims data include enterprise characteristic parameters and claims results corresponding to the enterprise characteristic parameters, and then an initial policy network is established, and a decision optimization factor is constructed based on the historical enterprise claims data and the initial policy network, wherein the decision optimization factor includes a state factor for characterizing a state, an action factor for characterizing an action, and a reward factor for characterizing a reward, and then based on the state factor, action factor, and reward factor, the network parameters of the initial policy network are updated using a policy gradient to obtain an optimized policy network, and finally based on the optimized policy network, the current enterprise's claims decision result is output. The present invention can automatically optimize the decision model without manual intervention, can be processed with data, is relatively robust to noise, and greatly improves the accuracy of the decision. In addition, it can also process and give answers when the input information is incomplete, thereby enhancing the performance of the decision model.

[0068] In some embodiments, see Figure 3 , the step S200 specifically includes:

[0069] S210, establishing an initial strategy network based on a multi-layer perceptron model, and presetting network parameters of the initial strategy network;

[0070] S220, using the enterprise characteristic parameters in the historical enterprise claims data as state factors;

[0071] S230, based on the state factor, using the initial strategy network to predict the action corresponding to the state factor and use it as an action factor, wherein the action is a claim settlement action or a claim rejection action;

[0072] S240: Based on the action factor, a preset reward value calculation model is used to obtain a corresponding reward value, and the reward value is used as a reward factor.

[0073] In reinforcement learning, state factor, action factor and reward factor are core concepts, which together form the basis of the reinforcement learning model.

[0074] The state factor represents the specific situation or configuration in the reinforcement learning environment, and the agent will be in a certain state at any given point in time. The state may include but is not limited to position, speed, variables of other environments, etc. In this embodiment, the state is the enterprise claims data. In reinforcement learning, the state is the basis for the agent's decision-making, and the agent needs to choose the appropriate action based on the current state. The state transition in the reinforcement learning model is random and has the Markov property, that is, the next state depends only on the current state and the action taken, and has nothing to do with the previous state and action. This property is widely used in Markov decision processes (MDPs) to simulate and predict the transition rules between states. In practical applications, the state space can be finite or continuous, which determines the complexity of the problem. For example, board games such as chess have a finite state space, while tasks such as autonomous driving have a continuous state space.

[0075] An action factor refers to the actions or decisions that an agent can take in a specific state. In a reinforcement learning environment, the choice of action directly affects the state transition of the environment and the reward obtained by the agent. Therefore, one of the core goals of reinforcement learning is for the agent to learn how to choose the best action based on the state. The agent's choice of action is usually based on a certain strategy, which can be deterministic or random. In the initial stage, the agent may randomly choose actions to explore the environment; as learning deepens, the agent will gradually form a strategy based on past experience to optimize long-term benefits. The action space defines all possible actions that the agent can take in each state. The design of the action space has a direct impact on the complexity of the reinforcement learning task and the search space for solutions.

[0076] Reward factor is a key element in reinforcement learning. It reflects the evaluation obtained by the agent after performing a specific action in a specific state. The purpose of reward is to guide the agent to learn the optimal strategy, that is, to maximize the long-term cumulative reward. Positive rewards encourage the agent to repeat beneficial behaviors, while negative rewards encourage it to avoid harmful behaviors. The reward function defines the immediate reward that the agent can obtain under a specific state and action. In many tasks, designing an effective reward function is the key to effective learning. The reward function must be closely related to the target task and accurately reflect the performance of the agent. Return is a core concept in reinforcement learning. It represents the cumulative sum of rewards from the current time step to all future time steps. The discount factor determines the current value of future rewards, that is, recent rewards are more valuable than distant rewards. This reflects the trade-off between immediate rewards and long-term rewards in reinforcement learning.

[0077] In summary, state factors, action factors, and reward factors play a vital role in reinforcement learning. Through accurate state description, wise action selection, and reasonable reward design, the agent can be effectively guided to learn how to make optimal decisions in complex environments. In practical applications, designers need to fully consider the relationship and influence of these three factors to achieve the best performance of the reinforcement learning model.

[0078] In this embodiment, in the initialization stage, the parameters θ of the policy network are set, which can be a multi-layer perceptron (MLP) model, which takes the characteristics of the case as input and outputs the decision of claim settlement or claim rejection. In a specific scenario, different decisions of the target enterprise can be divided into different discrete action categories, such as claim settlement and claim rejection, and the next action (claim settlement or claim rejection) is determined by each feature, and then each reward return is collected through a preset reward value calculation model to obtain a reward factor, and finally the state factor, action factor and reward factor are obtained to optimize the model.

[0079] In some embodiments, step S240 specifically includes:

[0080] Obtaining the action factor, and determining the type of the action factor;

[0081] When the action factor is a claim settlement action, a first reward value is output; when the action factor is a claim rejection action, a second reward value is output, wherein the first reward value is a negative number and the second reward value is a positive number.

[0082] In this embodiment, since the scenario applicable to the present invention is to discover fraud risks, the claim rejection action is a positive reward incentive, while the claim settlement action is a negative reward incentive. Therefore, the first reward value is a negative number, and the second reward value is a positive number. For example, if the prediction is a claim rejection, the reward return is increased by 15 points. If the prediction is a claim settlement, the reward return is reduced by 15 points.

[0083] In some embodiments, see Figure 4 , the step S300 specifically includes:

[0084] S310, based on the state factor and the action factor, calculating the probability that the initial strategy network selects the action factor;

[0085] S320, based on the probability, calculating the gradient of the logarithmic probability of the initial strategy network;

[0086] S330, calculating a policy gradient based on the gradient of the logarithmic probability of the initial policy network and the reward factor;

[0087] S340: Using policy gradients to update network parameters of the initial policy network to obtain an optimized policy network.

[0088] In this embodiment, each piece of data is a corporate case report. The agent first calculates the probability of the policy network choosing to settle a claim or rejecting a claim in the current state, which can be obtained by using the softmax function. Then, the agent calculates the reward based on the result of the case. The agent uses the policy gradient formula to calculate the gradient, which represents the impact of the current decision on the future reward.

[0089] In some embodiments, step S320 specifically includes:

[0090] Calculate the index value of each action based on the state factor and the action factor;

[0091] Based on the index value of each action, the probability of the initial policy network selecting the action factor is calculated.

[0092] In this embodiment, the softmax function is used to calculate the probability. The Softmax function is an activation function widely used in multi-classification problems in machine learning and deep learning. It can convert the original score into a probability distribution. Specifically, the Softmax function converts the score of each category into a corresponding probability through an exponential function and normalization, so that the output result is more suitable for the classification task. For the input vector z = [z1, z2, ..., zn], the output of the Softmax function is a probability distribution p = [p1, p2, ..., pn]. The Softmax function ensures that each element pi is between 0 and 1, and the sum of all elements is 1. This makes the output a valid probability distribution for representing the relative likelihood of different categories.

[0093] Specifically, the calculation formula of the probability is:

[0094]

[0095] Among them, π(ai|s) represents the probability of selecting action ai in state s, e zi Calculate the exponential value of the ith action for the process, e zj Compute the exponential value of the j-th action for the process.

[0096] In some embodiments, step S340 specifically includes:

[0097] Use the gradient ascent algorithm to maximize the policy gradient value;

[0098] Based on the maximized policy gradient value and the preset learning rate, the network parameters of the initial policy network are optimized until an optimized policy network is obtained.

[0099] In this embodiment, the gradient ascent method is an optimization algorithm used to find the local or global maximum of the objective function through iteration. Specifically, its core idea is to iteratively update along the positive direction of the function gradient, so that the objective function value gradually increases. The gradient ascent method is usually used in machine learning scenarios such as maximum likelihood estimation and constrained optimization. The gradient indicates that the directional derivative of a function at this point reaches the maximum value along this direction, that is, the function changes fastest along the gradient direction at this point. For example, for a univariate function f(x), the gradient at point x_0 is f'(x_0). The gradient ascent method starts from the selected initial parameters, calculates the gradient of the current point, and then updates the parameters in the positive direction of the gradient. The specific formula is Among them, α is the learning rate, is the gradient of the objective function at θ_i.

[0100] Exemplarily, the calculation formula of the policy gradient is:

[0101]

[0102] in, represents the policy gradient, is the gradient of the logarithmic probability of the strategy, S is the state, a is the action, and R is the reward. Represents the mathematical expectation of probability.

[0103] Optionally, after a training cycle, the agent uses the accumulated policy gradient to update the policy network parameters θ. This is done using the gradient ascent method, and the learning rate α controls the step size of the parameter update. The updated policy network will be more likely to take decisions that will yield high rewards in the long run. Specifically, the iterative update formula for the policy parameters is as follows:

[0104]

[0105] Among them, θ is the policy parameter, α is the learning rate, which is used to control the step size of parameter update, T represents the total iteration step size, t represents the current round, and a t represents the action factor of the current round, s t Represents the state factor of the current round, R t Represents the reward factor for the current round.

[0106] In some embodiments, the step S400 specifically includes:

[0107] Get the enterprise characteristic parameters of the current enterprise;

[0108] The enterprise characteristic parameters of the current enterprise are input into the optimized strategy network to generate a claim decision result.

[0109] In this embodiment, after obtaining the optimized strategy network, the optimized strategy network can be used to directly calculate the claim decision result. Specifically, first obtain the current enterprise characteristic parameters, such as insurance characteristics and claim characteristics, where the insurance parameters include insurance time, insured person, insured person address, legal person, enterprise credit code, employee list, etc.; claim characteristics include reporting time, reporter, investigation time, reporting reason and other data, and then directly input these data into the strategy network to obtain the claim decision result.

[0110] It is understandable that when there are multiple characteristic parameters for an enterprise, in order to achieve decision-making, the normalization method can be used to normalize the multiple characteristic parameters, thereby facilitating the calculation of subsequent models and further ensuring the accuracy of model calculations.

[0111] The technical solution provided by the present invention first obtains several historical enterprise claims data, builds a training set based on the several historical enterprise claims data, wherein the historical enterprise claims data includes enterprise characteristic parameters and claims results corresponding to the enterprise characteristic parameters, then establishes an initial policy network, builds a decision optimization factor based on the historical enterprise claims data and the initial policy network, wherein the decision optimization factor includes a state factor for characterizing the state, an action factor for characterizing the action, and a reward factor for characterizing the reward, and then based on the state factor, action factor and reward factor, uses a policy gradient to update the network parameters of the initial policy network to obtain an optimized policy network, and finally outputs the current enterprise's claims decision result based on the optimized policy network. The present invention can automatically optimize the decision model without manual intervention, can process data, is relatively robust to noise, and greatly improves the accuracy of decision-making. In addition, it can also process and give answers when the input information is incomplete, thereby enhancing the performance of the decision model.

[0112] Another embodiment of the present invention provides a claim decision-making device, see Figure 5 The claim decision-making device includes a data acquisition module 11, a decision optimization factor acquisition module 12, an optimization module 13 and a decision module 14.

[0113] The data acquisition module 11 is used to acquire a number of historical enterprise claims data and construct a training set based on the historical enterprise claims data, wherein the historical enterprise claims data includes enterprise characteristic parameters and claims results corresponding to the enterprise characteristic parameters.

[0114] The decision optimization factor acquisition module 12 is used to establish an initial strategy network, and construct a decision optimization factor based on the historical enterprise claims data and the initial strategy network, wherein the decision optimization factor includes a state factor for characterizing a state, an action factor for characterizing an action, and a reward factor for characterizing a reward.

[0115] The optimization module 13 is used to update the network parameters of the initial policy network using policy gradient based on the state factor, action factor and reward factor to obtain an optimized policy network.

[0116] The decision module 14 is used to output the current enterprise's claim decision result based on the optimized strategy network.

[0117] In some embodiments, the decision optimization factor acquisition module 12 includes:

[0118] An establishing unit, used for establishing an initial strategy network based on a multi-layer perceptron model, and presetting network parameters of the initial strategy network;

[0119] A status factor acquisition unit, used to use the enterprise characteristic parameters in the historical enterprise claims data as status factors;

[0120] An action factor acquisition unit, configured to use the initial strategy network to predict an action corresponding to the state factor based on the state factor and use the action as the action factor, wherein the action is a claim settlement action or a claim rejection action;

[0121] The reward factor acquisition unit is used to obtain a corresponding reward value based on the action factor using a preset reward value calculation model, and use the reward value as the reward factor.

[0122] In some embodiments, the reward factor acquisition unit is specifically used to:

[0123] Obtaining the action factor, and determining the type of the action factor;

[0124] When the action factor is a claim settlement action, a first reward value is output; when the action factor is a claim rejection action, a second reward value is output, wherein the first reward value is a negative number and the second reward value is a positive number.

[0125] In some embodiments, the optimization module 13 includes:

[0126] A probability calculation unit, used to calculate the probability of the initial strategy network selecting the action factor based on the state factor and the action factor;

[0127] A logarithmic probability gradient calculation unit, used to calculate the gradient of the logarithmic probability of the initial policy network based on the probability;

[0128] A policy gradient calculation unit, configured to calculate a policy gradient based on the gradient of the logarithmic probability of the initial policy network and the reward factor;

[0129] An updating unit is used to update the network parameters of the initial policy network using a policy gradient to obtain an optimized policy network.

[0130] In some embodiments, the probability calculation unit is specifically used to:

[0131] Calculate the index value of each action based on the state factor and the action factor;

[0132] Based on the index value of each action, the probability of the initial policy network selecting the action factor is calculated.

[0133] In some embodiments, the updating unit is specifically configured to:

[0134] Use the gradient ascent algorithm to maximize the policy gradient value;

[0135] Based on the maximized policy gradient value and the preset learning rate, the network parameters of the initial policy network are optimized until an optimized policy network is obtained.

[0136] In some embodiments, the decision module 14 includes:

[0137] A parameter acquisition unit, used to acquire enterprise characteristic parameters of the current enterprise;

[0138] The decision result output unit is used to input the enterprise characteristic parameters of the current enterprise into the optimized strategy network to generate a claim decision result.

[0139] In this embodiment, firstly, a number of historical enterprise claims data are obtained, and a training set is constructed based on the several historical enterprise claims data, wherein the historical enterprise claims data include enterprise characteristic parameters and claims results corresponding to the enterprise characteristic parameters, and then an initial policy network is established, and a decision optimization factor is constructed based on the historical enterprise claims data and the initial policy network, wherein the decision optimization factor includes a state factor for characterizing the state, an action factor for characterizing the action, and a reward factor for characterizing the reward, and then based on the state factor, action factor, and reward factor, the network parameters of the initial policy network are updated using a policy gradient to obtain an optimized policy network, and finally based on the optimized policy network, the current enterprise's claims decision result is output. The present invention can automatically optimize the decision model without manual intervention, can be processed with data, is relatively robust to noise, and greatly improves the accuracy of the decision. In addition, it can also process and give answers when the input information is incomplete, thereby enhancing the performance of the decision model.

[0140] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0141] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0142] To solve the above technical problems, the embodiment of the present invention also provides a computer device. Figure 6 , Figure 6 This is a basic structural block diagram of the computer device in this embodiment.

[0143] The computer device 8 includes a memory 8a, a processor 8b, and a network interface 8c that are interconnected through a system bus. It should be noted that the figure only shows a computer device 8 having components 8a-8c, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), digital processors (DSPs), embedded devices, etc.

[0144] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0145] The memory 8a includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 8a can be an internal storage unit of the computer device 8, such as a hard disk or memory of the computer device 8. In other embodiments, the memory 8a can also be an external storage device of the computer device 8, such as a plug-in hard disk equipped on the computer device 8, a smart memory card (Smart Med ia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. Of course, the memory 8a can also include both the internal storage unit of the computer device 8 and its external storage device. In this embodiment, the memory 8a is generally used to store the operating system and various application software installed on the computer device 8, such as a computer-readable instruction of a claim decision method, etc. In addition, the memory 8a can also be used to temporarily store various types of data that have been output or are to be output.

[0146] The processor 8b may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other claims processing chips in some embodiments. The processor 8b is generally used to control the overall operation of the computer device 8. In this embodiment, the processor 8b is used to run computer-readable instructions or process data stored in the memory 8a, such as computer-readable instructions for running the claims processing method.

[0147] The network interface 8c may include a wireless network interface or a wired network interface. The network interface 8c is generally used to establish a communication connection between the computer device 8 and other electronic devices.

[0148] The computer device proposed in this embodiment belongs to the field of financial technology and is applied to the insurance claims process in financial scenarios, such as corporate insurance claims. The present invention first obtains a number of historical corporate claims data, and constructs a training set based on the several historical corporate claims data, wherein the historical corporate claims data includes corporate characteristic parameters and claims results corresponding to the corporate characteristic parameters, and then establishes an initial policy network, and constructs a decision optimization factor based on the historical corporate claims data and the initial policy network, wherein the decision optimization factor includes a state factor for characterizing the state, an action factor for characterizing the action, and a reward factor for characterizing the reward, and then based on the state factor, action factor and reward factor, the network parameters of the initial policy network are updated using a policy gradient to obtain an optimized policy network, and finally based on the optimized policy network, the current corporate claims decision result is output. The present invention can automatically optimize the decision model without manual intervention, can be processed with data, is relatively robust to noise, and the accuracy of the decision is greatly improved. In addition, the answer can be given even when the input information is incomplete, which enhances the performance of the decision model.

[0149] The present invention also provides another embodiment, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by a processor to enable the processor to perform the steps of the claim decision method as described above.

[0150] The computer-readable storage medium proposed in this embodiment belongs to the field of financial technology and is applied to the insurance claims process in financial scenarios, such as corporate insurance claims. The present invention first obtains a number of historical corporate claims data, and constructs a training set based on the several historical corporate claims data, wherein the historical corporate claims data includes corporate characteristic parameters and claims results corresponding to the corporate characteristic parameters, and then establishes an initial policy network, and constructs a decision optimization factor based on the historical corporate claims data and the initial policy network, wherein the decision optimization factor includes a state factor for characterizing the state, an action factor for characterizing the action, and a reward factor for characterizing the reward, and then based on the state factor, action factor and reward factor, the network parameters of the initial policy network are updated using a policy gradient to obtain an optimized policy network, and finally based on the optimized policy network, the current corporate claims decision result is output. The present invention can automatically optimize the decision model without manual intervention, can be processed with data, is relatively robust to noise, and the accuracy of the decision is greatly improved. In addition, the answer can be given even when the input information is incomplete, which enhances the performance of the decision model.

[0151] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0152] In summary, the claims decision method, device, equipment and storage medium provided by the present invention first obtain several historical enterprise claims data, build a training set based on the several historical enterprise claims data, wherein the historical enterprise claims data include enterprise characteristic parameters and claims results corresponding to the enterprise characteristic parameters, then establish an initial policy network, build a decision optimization factor based on the historical enterprise claims data and the initial policy network, wherein the decision optimization factor includes a state factor for characterizing the state, an action factor for characterizing the action and a reward factor for characterizing the reward, then based on the state factor, action factor and reward factor, use the policy gradient to update the network parameters of the initial policy network to obtain an optimized policy network, and finally output the current enterprise's claims decision result based on the optimized policy network. The present invention can automatically optimize the decision model without manual intervention, can process data, is relatively robust to noise, and greatly improves the accuracy of decision making. In addition, it can also process and give answers when the input information is incomplete, thereby enhancing the performance of the decision model.

[0153] It should be noted that if software tools or components other than those of the Company appear in the embodiments of the present application, they are merely used for illustration and do not represent actual use. The specific implementation of the present invention described above does not constitute a limitation on the protection scope of the present invention. Any other corresponding changes and modifications made based on the technical concept of the present invention shall be included in the protection scope of the claims of the present invention.

Claims

1. A claim decision-making method, characterized in that: The steps include: Acquire a number of historical enterprise claims data, and construct a training set based on the number of historical enterprise claims data, wherein the historical enterprise claims data includes enterprise characteristic parameters and claims results corresponding to the enterprise characteristic parameters; Establishing an initial strategy network, and constructing a decision optimization factor based on the historical enterprise claims data and the initial strategy network, wherein the decision optimization factor includes a state factor for characterizing a state, an action factor for characterizing an action, and a reward factor for characterizing a reward; Based on the state factor, action factor and reward factor, the network parameters of the initial policy network are updated using policy gradient to obtain an optimized policy network; Based on the optimized strategy network, the current enterprise's claims decision result is output.

2. The claim decision-making method according to claim 1, characterized in that: The establishing of the initial strategy network, constructing a decision optimization factor based on the historical enterprise claims data and the initial strategy network, includes: Establishing an initial strategy network based on a multi-layer perceptron model, and presetting network parameters of the initial strategy network; Using the enterprise characteristic parameters in the historical enterprise claims data as state factors; Based on the state factor, the initial strategy network is used to predict the action corresponding to the state factor and used as the action factor, wherein the action is a claim settlement action or a claim rejection action; Based on the action factor, a preset reward value calculation model is adopted to obtain a corresponding reward value, and the reward value is used as the reward factor.

3. The claim decision-making method according to claim 2, characterized in that: The preset reward value calculation model is used to obtain the corresponding reward value, including: Obtaining the action factor, and determining the type of the action factor; When the action factor is a claim settlement action, a first reward value is output; when the action factor is a claim rejection action, a second reward value is output, wherein the first reward value is a negative number and the second reward value is a positive number.

4. The claim decision-making method according to claim 1, characterized in that: The method of updating the network parameters of the initial policy network by using policy gradient based on the state factor, action factor and reward factor to obtain an optimized policy network includes: Based on the state factor and the action factor, calculating the probability that the initial strategy network selects the action factor; Based on the probability, calculating the gradient of the log probability of the initial policy network; Calculating a policy gradient based on the gradient of the log probability of the initial policy network and the reward factor; The network parameters of the initial policy network are updated using policy gradient to obtain an optimized policy network.

5. The claim decision-making method according to claim 4, characterized in that: The calculating, based on the state factor and the action factor, the probability that the initial strategy network selects the action factor comprises: Calculate the index value of each action based on the state factor and the action factor; Based on the index value of each action, the probability of the initial policy network selecting the action factor is calculated.

6. The claim decision-making method according to claim 1, characterized in that: The adopting the policy gradient to update the network parameters of the initial policy network to obtain an optimized policy network includes: Use the gradient ascent algorithm to maximize the policy gradient value; Based on the maximized policy gradient value and the preset learning rate, the network parameters of the initial policy network are optimized until an optimized policy network is obtained.

7. The claim decision-making method according to claim 1, characterized in that: The output of the current enterprise's claim decision result based on the optimized strategy network includes: Get the enterprise characteristic parameters of the current enterprise; The enterprise characteristic parameters of the current enterprise are input into the optimized strategy network to generate a claim decision result.

8. A claim decision-making device, characterized in that: include: A data acquisition module, used to acquire a number of historical enterprise claims data, and construct a training set based on the historical enterprise claims data, wherein the historical enterprise claims data includes enterprise characteristic parameters and claims results corresponding to the enterprise characteristic parameters; A decision optimization factor acquisition module, used to establish an initial strategy network, and construct a decision optimization factor based on the historical enterprise claims data and the initial strategy network, wherein the decision optimization factor includes a state factor for representing a state, an action factor for representing an action, and a reward factor for representing a reward; An optimization module, configured to update the network parameters of the initial policy network using a policy gradient based on the state factor, the action factor, and the reward factor to obtain an optimized policy network; The decision module is used to output the current enterprise's claim decision result based on the optimized strategy network.

9. A computer device, characterized in that: It comprises a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the claim decision-making method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the claim decision-making method according to any one of claims 1 to 7.

Images