Risk identification factor mining method and device, computer equipment and storage medium

Abstract

The invention belongs to the field of artificial intelligence and finance, and relates to a risk identification factor mining method, which comprises the following steps of: setting different types of operators according to an insurance risk identification target through acquired data of different data types, constructing a genetic programming model according to the different types of operators, and obtaining a risk identification factor by initializing a formula population. And iteratively calculating the risk interpretation degree, continuously optimizing the formula tree through a genetic algorithm to obtain a final formula tree species group, selecting the formula tree with the highest risk interpretation degree from the final formula tree species group as a risk identification factor, and performing test optimization on the risk identification factor to obtain an optimal risk identification model. The invention further provides a risk identification factor mining device, computer equipment and a storage medium. In addition, the invention also relates to a block chain technology, and the text data and the numeric data can be stored in a block chain. According to the invention, the automatic identification of the insurance risk can be better realized, and the accuracy of risk identification is improved.

Description

Technical Field

[0001] The present application relates to the fields of artificial intelligence and financial technology, and in particular to a method, device, computer equipment and storage medium for mining risk identification factors. Background Art

[0002] Insurance fraud, as an illegal act, has long plagued the insurance industry. Fraudsters attempt to defraud excess insurance amounts by fabricating insurance objects, deliberately creating insurance accidents, fabricating the causes of accidents, or exaggerating the extent of losses. This not only causes insurance companies to pay more insurance money than they should, but also seriously undermines the fairness and stability of the insurance market. Traditional risk identification methods mainly rely on manual review by insurance salesmen. Although this method can identify some fraudulent behaviors to a certain extent, it is limited by factors such as manpower, time, and experience, and its efficiency and accuracy are insufficient.

[0003] In recent years, with the rapid development of machine learning and artificial intelligence technologies, insurance companies have begun to try to introduce these advanced technologies to strengthen risk fraud management. By building a risk rule library, insurance companies can use big data and algorithm models to conduct more comprehensive and in-depth analysis of insurance cases, thereby more effectively identifying potential fraud. However, although this method has achieved certain results, it still faces a series of problems that need to be solved in actual application.

[0004] First, the diversity of data types and the high number of dimensions are a major challenge in current risk fraud management. In order to more accurately portray case information and the user portrait of the policyholder, insurance companies need to collect and process a large amount of multi-dimensional data. These data include both numerical amount variables and time variables, as well as long text information in categorical variables, such as the course of the accident, diagnosis results, etc. When extracting the characteristic variables of these data, it is difficult to extract due to the diversity and complexity of the data types. In particular, for long text information in categorical variables, although the dimensionality can be reduced by extracting keywords, it often leads to the problem of dimensionality explosion, further increasing the difficulty of data processing.

[0005] Secondly, the high requirements for factor combinations in insurance-specific scenarios are also one of the problems that need to be solved urgently. In the special scenario of using policy data for fraud detection, if numerical variables such as time and amount are set as factors separately, their prediction effect is often poor. For example, if the two variables of reporting time and policy start time are used separately, their ability to identify fraud is limited. However, if these two variables are combined, such as subtracting each other, to generate new identification factors such as the difference between the reporting time and the policy start time, fraud may be more effectively identified. In addition, composite factors such as the ratio of the actual premium amount to the insured amount of the claim are also important indicators commonly used in fraud detection. However, the construction and mining of these factors require high professional knowledge and experience, and often rely on the subjective judgment of business experts, with a low degree of automation.

[0006] Finally, the low degree of automation in the construction and mining of risk identification factors is also a major problem currently faced. At present, risk identification factors are often synthesized and explained by business experts with rich experience. These factors are relatively simple in composition, and usually have logic first and then factors. Although this method can explain the rationality of risk factors to a certain extent, it is difficult to mine deep-level risk combinations that are more explanatory of fraud factors. Summary of the invention

[0007] The purpose of the embodiments of the present application is to propose a method, device, computer equipment and storage medium for mining risk identification factors, so as to solve the technical problems in the related technology that the data types are diverse and complex, resulting in increased difficulty in data processing, and the risk identification factor combination has high requirements but simple structure, making it difficult to mine deep risk factor combinations, resulting in low risk identification accuracy.

[0008] In order to solve the above technical problems, the present application embodiment provides a method for mining risk identification factors, which adopts the following technical solution:

[0009] Acquire unstructured text data and numerical data in an insurance business database, perform word segmentation processing on the text data to obtain word segmentation results, extract keywords from the word segmentation results to obtain keyword data;

[0010] A first operation formula is obtained based on the logical operator set for the keyword data, a second operation formula is obtained based on the basic operator set for the numerical data, and a genetic programming model is constructed through the first operation formula and the second operation formula according to an insurance risk identification target;

[0011] Generate a formula tree population with a preset number of formula trees by random initialization as an initial formula tree population corresponding to the genetic programming model;

[0012] Acquire the risk label data annotated in the insurance business database, divide the data into a training set and a test set according to a preset ratio, and calculate the risk interpretation degree of each formula tree in the initial formula tree population through the training set;

[0013] According to the risk interpretation degree, the initial formula tree population is iteratively optimized using a genetic algorithm until a preset termination condition is reached to obtain a final formula tree population;

[0014] Selecting a first preset number of formula trees with the highest risk explanation degree from the final formula tree population as risk identification factors;

[0015] Using the test set, calculating the risk identification accuracy of the risk identification factor, and when the risk identification accuracy is greater than or equal to a preset threshold, outputting the genetic programming model corresponding to the risk identification factor as a risk identification model;

[0016] The insurance business data to be identified is obtained, and the insurance business data to be identified is input into the risk identification model to obtain a risk identification result.

[0017] In order to solve the above technical problems, the embodiment of the present application also provides a risk identification factor mining device, which adopts the following technical solution:

[0018] An acquisition module is used to acquire unstructured text data and numerical data in an insurance business database, perform word segmentation processing on the text data to obtain word segmentation results, extract keywords from the word segmentation results to obtain keyword data;

[0019] A construction module, configured to obtain a first operation formula based on the logical operator set for the keyword data, obtain a second operation formula based on the basic operator set for the numerical data, and construct a genetic programming model through the first operation formula and the second operation formula according to an insurance risk identification target;

[0020] An initialization module, used for generating a formula tree population with a preset number of formula trees by random initialization as an initial formula tree population corresponding to the genetic programming model;

[0021] A calculation module, used to obtain the risk label data marked in the insurance business database, divide the training set and the test set according to a preset ratio, and calculate the risk interpretation degree of each formula tree in the initial formula tree population through the training set;

[0022] An iteration module, used to iteratively optimize the initial formula tree population using a genetic algorithm according to the risk interpretation degree until a preset termination condition is reached to obtain a final formula tree population;

[0023] A selection module, used for selecting a first preset number of formula trees with the highest risk explanation degree from the final formula tree population as risk identification factors;

[0024] A testing module, used to calculate the risk identification accuracy of the risk identification factor using the test set, and when the risk identification accuracy is greater than or equal to a preset threshold, output the genetic programming model corresponding to the risk identification factor as a risk identification model;

[0025] The identification module is used to obtain the insurance business data to be identified, input the insurance business data to be identified into the risk identification model, and obtain the risk identification result.

[0026] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0027] The computer device includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the above-mentioned risk identification factor mining method when executing the computer-readable instructions.

[0028] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0029] The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the method for mining risk identification factors as described above are implemented.

[0030] Compared with the prior art, this application has the following beneficial effects:

[0031] The present application provides a method for mining risk identification factors. By acquiring data of different data types, different types of operators are set according to insurance risk identification goals, and operation formulas for different data are obtained according to different types of operators. A genetic programming model is constructed based on the operation formulas. The formula population is initialized, the risk interpretation degree is iteratively calculated, and the formula tree is continuously optimized through a genetic algorithm to obtain a final formula tree population. The formula tree with the highest risk interpretation degree is selected from the final formula tree population as the risk identification factor, and the risk identification factor is tested and optimized to obtain an optimal risk identification model. The present application applies a genetic programming algorithm to the field of insurance risk identification, which can automatically discover complex nonlinear risk identification factors and improve the accuracy and efficiency of risk identification. At the same time, by modeling different data types separately, the multidimensional characteristics of insurance data are fully utilized, and the automatic identification of insurance risks can be better realized, further improving the accuracy and interpretability of risk identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0033] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0034] Figure 2 is a flow chart of an embodiment of a method for mining risk identification factors according to the present application;

[0035] Figure 3 It is a structural schematic diagram of a specific implementation method of the formula tree connection provided by the present application;

[0036] Figure 4 It is a structural schematic diagram of another specific implementation method of the formula tree connection provided by the present application;

[0037] Figure 5 It is a structural schematic diagram of a specific implementation method of the cross operation according to the present application;

[0038] Figure 6 is a schematic structural diagram of an embodiment of an excavation device for risk identification factors according to the present application;

[0039] Figure 7 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0040] 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 application; 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 application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application 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 application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0041] 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 application. The appearance of the phrase in various locations 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.

[0042] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0043] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables.

[0044] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0045] The terminal device 101 can be any electronic device with a display screen and supporting web browsing. In addition to a laptop computer 1011, a tablet computer 1012 or a mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV), a laptop computer, a desktop computer, etc.

[0046] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0047] It should be noted that the risk identification factor mining method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the risk identification factor mining device is generally set in the server / terminal device.

[0048] 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 implementation requirements.

[0049] Continue to refer Figure 2 , shows a flow chart of an embodiment of a method for mining risk identification factors according to the present application, comprising the following steps:

[0050] Step S201, obtaining unstructured text data and numerical data in the insurance business database, performing word segmentation processing on the text data to obtain word segmentation results, extracting keywords from the word segmentation results to obtain keyword data.

[0051] An insurance business database refers to a database specifically used to store, manage and process data related to the business activities of an insurance company. It contains a large amount of insurance data, including but not limited to customer information, product information, order information, financial data, supply chain data and inventory data, involving different data types, such as numeric type and text type. The numeric type includes the insurance amount and insurance time, and the text type includes text information such as the accident process and diagnosis results.

[0052] In this embodiment, the risk identification factor mining method is executed on the electronic device (eg Figure 1 The server / terminal device shown in the figure can obtain the unstructured text data and numerical data in the insurance business database through a wired connection or a wireless connection. It should be noted that the above wireless connection method may include but is not limited to 3G / 4G / 5G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or to be developed in the future.

[0053] After obtaining the insurance data in the insurance business database, extract the unstructured text data and numerical data in the insurance data, and perform word segmentation on the text data. The word segmentation process can use word segmentation tools, BERT word segmenter, etc. to obtain word segmentation results, and extract keywords from the word segmentation results.

[0054] In this embodiment, before the word segmentation process, the text data may be preprocessed, including removing irrelevant spaces, special characters, punctuation marks, and stop words.

[0055] It should be emphasized that in order to further ensure the privacy and security of text data and numerical data, the above text data and numerical data can also be stored in a node of a blockchain.

[0056] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0057] In some optional implementations of this embodiment, the step of extracting keywords from the word segmentation results to obtain keyword data includes:

[0058] According to the word segmentation results, the number of occurrences of each word in the text data is calculated by the word frequency statistics method to obtain the word frequency of each word;

[0059] Using the inverse document frequency algorithm, calculate the distribution of each word in the text data to obtain the inverse document frequency;

[0060] Multiply the word frequency by the inverse document frequency, calculate the weight value of each word segment, and get the word weight list;

[0061] Obtain the preset word frequency threshold and document frequency threshold, and filter out the words in the word weight list that satisfy both the word frequency greater than or equal to the word frequency threshold and the document frequency less than or equal to the document frequency threshold as candidate keywords;

[0062] The candidate keywords are sorted from high to low according to the weight values, and the candidate keywords with a preset number of words ranked first are selected as keyword data.

[0063] In this embodiment, the larger the weight value, the more important the keyword. The final keyword data is selected through word frequency, inverse document frequency and weight value to improve the efficiency of keyword extraction. At the same time, non-keywords with lower weights can be eliminated, so that the extracted keywords have stronger explanatory power in risk identification, thereby improving the accuracy of risk identification.

[0064] Step S202, obtain a first operation formula based on the logical operator set based on the keyword data, obtain a second operation formula based on the basic operator set based on the numerical data, and construct a genetic programming model through the first operation formula and the second operation formula according to the insurance risk identification target.

[0065] Among them, genetic programming is a heuristic supervised learning algorithm of the violent iteration type inspired by Darwin's theory of natural evolution. The elements in genetic programming are represented by a tree graph structure. The tree is a formula composed of functions or operators acting on variables. The internal nodes are functions or operators, and the leaf nodes are the screened variables.

[0066] In this embodiment, different operators are set for data of different data types, and operation formulas can be obtained based on the operators. The operation formulas are components of the genetic programming formula tree. Specifically, logical operators are set for keyword data, and the logical operators include or, and, and not; basic operators are set for numerical data, and the basic operators include addition, subtraction, multiplication, and division. A formula containing a logical operator is used as the first operation formula, and a formula containing basic operators of addition, subtraction, multiplication, and division is used as the second operation formula. According to the insurance risk identification target, a genetic programming model is constructed through the first operation formula and the second operation formula.

[0067] In some optional implementations, different data types construct different genetic programming models. For example, three data types of amount, time and keyword in insurance data are obtained. For the amount data, which is numerical data, a formula containing basic operators of addition, subtraction, multiplication and division is used as an operation formula for constructing a first genetic programming model; for the time data, which is also numerical data, a formula containing basic operators of addition, subtraction, multiplication and division is used as an operation formula for constructing a second genetic programming model; for the keyword data, a formula containing logical operators is used as an operation formula for constructing a third genetic programming model; the first genetic programming model, the second genetic programming model and the third genetic programming model are combined to obtain a total genetic programming model.

[0068] The public tables of genetic planning are shown in Table 1 and Table 2. Table 1 is the calculation formula for numerical data, and Table 2 is the calculation formula for keywords:

[0069] Table 1

[0070]

[0071]

[0072] Table 2

[0073] Formula Name Number of parameters explain ∨ n Or Operator

[0074] The time / amount formula tree example is shown in Figure 3 As shown, the formula tree is:

[0075]

[0076] This formula explains that if the insured has a high daily hospitalization fee and pays a high premium, it is very likely a case of insurance fraud in the form of injury and recuperation.

[0077] For example, see the keyword formula tree Figure 4 , the formula tree is:

[0078] tree_level = fall ∨ fall ∨ accident ∨ slip,

[0079] This formula indicates that if the keywords "fall" or "tumble" or "accident" or "slip" appear in the insurance policy data, there is a risk of fraud. The explanation of this risk factor is some risk cases of the insured that are easier to forge.

[0080] Step S203, generating a preset number of formula tree populations by random initialization as the initial formula tree population corresponding to the genetic programming model.

[0081] In this embodiment, a preset number of formula tree populations are generated by random initialization as the initial formula tree population corresponding to each genetic programming model. Each formula tree in the formula tree population is formed by operators acting on variables, wherein the variables are numerical data such as amount variables and time variables, and keyword data such as text classification variables such as the process of the accident and the diagnosis results. Specifically, according to the complexity of the business and the limitation of computing resources, the number of tree structures to be generated is determined, that is, the size of the initial population is determined, and then a random number generator is used to generate the nodes and connection relationships of each formula tree structure, and the nodes include operators and variables.

[0082] Step S204, obtaining the risk label data that has been annotated in the insurance business database, dividing the data into a training set and a test set according to a preset ratio, and calculating the risk explanation degree of each formula tree in the initial formula tree population through the training set.

[0083] In this embodiment, the risk label data includes but is not limited to claims risk, credit risk, insurance fraud risk, capital risk, misinsurance risk, etc. The risk label data is randomly divided into a training set and a test set according to a preset ratio, such as a ratio of 8:2. For the divided training set, a genetic programming model is used for training, and the risk identification accuracy of the variable logistic regression model is used as the risk interpretation degree of the genetic programming model. Among them, the risk interpretation degree indicates the ability of the formula to explain the risk. The higher the risk interpretation degree, the better the fit between the risk factor and the risk label, and the more accurate the risk identification.

[0084] In this embodiment, each formula tree in the initial formula tree population represents a potential risk identification factor. A variable logistic regression model is used to fit the risk factor represented by each formula tree to the risk label, and the risk interpretation degree of each formula tree is obtained by calculating the ratio of the regression mean square to the error mean square.

[0085] Logistic regression is a simple machine learning method for solving binary classification problems. It outputs the probability that data belongs to a certain category based on the given independent variable data. Since the value range of this probability is between 0 and 1, the value range of this probability can be mapped to the entire real number range through the logit function:

[0086]

[0087] In the above formula, pi represents the probability of an event occurring, Logit() represents the logit function, Beta_0…Beta_k represent the regression coefficients, X_1…X_k represent the values ​​of the independent variables, and the coefficients of the above formula are usually estimated by maximum likelihood estimation (MLE).

[0088] In this embodiment, for a single risk identification factor, the F value of the univariate Logistic regression can be selected as the risk explanation degree of the factor.

[0089]

[0090] In the above formula, MSR represents the mean square of regression, MSE represents the mean square of error, SSR represents the regression sum of squares, SSE represents the residual sum of squares, and n represents the number of samples. Through variance decomposition, the total variance SST sum of squares can be decomposed into the regression sum of squares SSR and the residual sum of squares SSE, and divided by their respective degrees of freedom to obtain the regression mean square MER and the error mean square MSE. The ratio of the two measures the degree of fit of the logistic function to the dependent variable. The larger the F value, the better the degree of fit to the dependent variable, that is, the higher the risk interpretation of the risk identification factor, and the more likely it is to become the parent of the new population.

[0091] Calculating the risk interpretation of the formula tree through the variable logistic regression model can improve the efficiency and accuracy of the calculation, so as to obtain the optimal formula tree population better and faster.

[0092] Step S205, according to the risk interpretation degree, adopting a genetic algorithm to iteratively optimize the initial formula tree population until a preset termination condition is reached to obtain a final formula tree population.

[0093] Specifically, according to the calculated risk interpretation degree, a population selection algorithm is used to select the second preset number of formula trees with the highest risk interpretation degree as the parent formula tree; the parent formula tree is copied to the new population; through the preset crossover probability, some parent formula trees are randomly selected for pairing, and subtrees of the parent formula tree are randomly selected for exchange and recombination to generate new crossover child formula trees; according to the preset mutation probability, the parent formula tree is mutated to generate new mutant child formula trees; the crossover child formula tree and the mutant child formula tree are added to the new population, and iterative evolution is continuously performed until the preset termination condition is reached. Among them, the preset termination condition is to reach the preset number of iterations or the population converges, and the population convergence refers to the change in the risk interpretation degree of the optimal formula for multiple consecutive generations is less than the preset threshold.

[0094] In this embodiment, the population selection algorithm is a roulette selection method, a tournament selection method or a gene selection method, and the roulette selection method, the tournament selection method or the gene selection method is used to select a formula tree with a risk interpretation degree higher than a preset interpretation degree threshold from the current formula tree population as a parent formula tree. Among them, the principle of the roulette selection method is to select a candidate formula tree according to the probability distribution, and the probability of the candidate being selected is determined by its performance relative to the entire formula population. This operation is similar to the survival of the fittest in nature. Candidates with better performance are more likely to be selected to produce offspring; the principle of the tournament selection method is to select two or more candidate formulas from the population at a time, and then select candidates with better performance from them. Compared with the global method of roulette, this method reserves the opportunity for candidates with local potential to enter the next round of offspring generation; the principle of the gene selection method is to sort all formula trees according to the risk interpretation degree, and then the formula tree with a larger risk interpretation degree will replace the formula tree with poor performance. For example, the initial formula tree population is sorted according to the risk interpretation degree, and the formula tree with a risk interpretation degree higher than the preset interpretation degree threshold is selected as the parent formula tree.

[0095] After selecting the parent formula tree, replication, crossover, and mutation operations are performed in sequence to optimize the combination of risk identification factors to increase population diversity and thus improve risk interpretation.

[0096] In some optional implementations of this embodiment, the step of copying the parent formula tree to the new population includes:

[0097] Get the tree structure data of the parent formula tree;

[0098] A depth-first search algorithm is used to traverse all nodes of the parent formula tree, and the parent formula tree is converted into a string form through a serialization operation to obtain a parent formula tree string;

[0099] Copy the parent formula tree string to a new buffer and allocate memory space of the same size as the parent formula tree in the memory of the new population;

[0100] Deserialize the parent formula tree string into a tree structure and rebuild the parent formula tree;

[0101] Insert the reconstructed parent formula tree into the memory address corresponding to the new population;

[0102] Determine whether all selected parent formula trees have been copied. If so, the copy operation is completed. Otherwise, return to continue selecting the next parent formula tree for copying until all parent formula trees are copied.

[0103] Through the copying operation, the formula tree with higher risk explanation degree is copied from the parent population to the new population, maintaining population diversity and accelerating the evolution process. At the same time, excellent genetic information is retained, providing a basis for subsequent crossover and mutation operations.

[0104] In some optional implementations, the step of randomly selecting part of the parent formula tree for pairing according to a preset crossover probability, and randomly selecting subtrees of the parent formula tree for exchange and recombination to generate a new crossover sub-generation formula tree includes:

[0105] According to the preset crossover probability, determine whether to perform a crossover operation on the copied offspring formula tree in the new population;

[0106] If so, two formula trees are randomly selected from the new population as the first and second parents;

[0107] Randomly selecting a first subtree and a second subtree in the first parent generation and the second parent generation respectively;

[0108] Determine whether the root node type of the first subtree matches the root node type of the second subtree;

[0109] If they match, the positions of the first subtree and the second subtree in their respective parent generations are exchanged to obtain a first cross-child formula tree and a cross-child formula tree.

[0110] Specifically, the crossover operation is to randomly pair the copied formula trees in the new population, randomly select subtrees for recombination through a pre-set crossover probability, and the obtained new formula trees enter the new offspring population.

[0111] The basic process of the crossover step is: first select the formula tree with the highest risk explanation as the parent of the population, and randomly select a subtree from the formula tree and replace it with a random subtree in the formula tree with the second highest explanation to generate a new offspring. For example, assuming that factors 1, 2, 3, and 4 represent the actual premium paid, the insured amount, the final case filing amount, and the number of hospitalization days, the crossover evolution process of the formula tree is as follows: Figure 5 shown.

[0112] The root node type of the first subtree matches the root node type of the second subtree. Specifically, the two root node types match, that is, both are function nodes or both are terminal nodes.

[0113] If the root node type of the first subtree does not match the root node type of the second subtree, a subtree is reselected until the root node type matching condition is met.

[0114] In this embodiment, the above crossover operation is repeated until a preset number of crossover operations is reached.

[0115] Through crossover operations, while introducing new risk identification factor combinations, excellent genetic information is retained, thereby accelerating the evolution rate.

[0116] In some optional implementations, the step of performing a mutation operation on the parent generation formula tree according to a preset mutation probability to generate a new mutant child generation formula tree includes:

[0117] Determine whether to perform a subtree mutation operation on the parent formula tree. If so, randomly select a subtree of the parent formula tree and replace the subtree with a randomly generated new subtree to obtain the first formula tree after the subtree mutation.

[0118] If the parent formula tree is not subjected to the subtree mutation operation, it is determined whether the parent formula tree is subjected to the point mutation operation. If so, a node of the parent formula tree is randomly selected and replaced with a new node with the same number of variables to obtain the second formula tree after the point mutation.

[0119] If the parent formula tree is not subjected to the point mutation operation, it is determined whether the parent tree is subjected to the Hoist mutation operation. If so, an initial subtree of the parent formula tree is randomly selected, and a replacement subtree is randomly selected from the initial subtree. The replacement subtree replaces the initial subtree to obtain the third formula tree after the Hoist mutation.

[0120] The first formula tree, the second formula tree and the third formula tree are used as variant child formula trees.

[0121] Among them, mutation: through the set mutation probability, part of the structure of the formula tree is mutated, and the new formula tree obtained enters the new offspring population. The mutation in genetic planning can be roughly divided into three categories: subtree mutation, Hoist mutation and subtree mutation. Subtree mutation is a random subtree that replaces the parent's random subtree by a randomly generated subtree. First, the formula tree with the highest risk interpretation is selected as the parent, and the subtree structure is randomly selected from it, and the subtree is also randomly generated to replace it to generate a new tree structure offspring; point mutation ensures the diversity of formulas in the population, so that the formulas eliminated in the selection step have the opportunity to return to the new population. Specifically, the formula tree with the highest risk interpretation is first selected as the parent, and nodes and leaves are randomly selected from it. Then, other formulas with the same number of variables are randomly selected from the old population to replace them to form a new offspring; Hoist mutation simplifies the formula tree by pruning the already generated formula tree structure to reduce overfitting. Specifically, Hoist mutation randomly selects a subtree structure from the parent formula tree with high risk interpretation and raises the random subtree coverage in the subtree to the position of the original subtree to generate a new offspring.

[0122] By mutating the parent formula tree, we can increase population diversity, improve the local search capability of the algorithm, and help to deeply explore risk identification factors.

[0123] In this embodiment, the formula tree population is evolved through genetic operations such as selection, replication, crossover, and mutation to obtain the final formula tree population, which can automatically mine deep risk identification factors that cannot be obtained from experience, enhance risk prediction capabilities, and improve risk assessment efficiency.

[0124] Step S206: Select a first preset number of formula trees with the highest risk explanation degree from the final formula tree population as risk identification factors.

[0125] From the high-quality final formula tree population obtained by the final evolution, several formula trees with the highest risk explanation are extracted as candidate risk identification factors. The candidate risk identification factors are subjected to correlation analysis, and repeated factors with correlation greater than the preset correlation threshold are eliminated, and redundant variables are excluded to obtain the final risk identification factor set. The risk identification factor set is used as a feature variable for the subsequent establishment of risk identification models such as decision trees and logistic regression, thereby realizing automated and precise insurance risk identification.

[0126] Step S207, using the test set, calculating the risk identification accuracy of the risk identification factor, and when the risk identification accuracy is greater than or equal to a preset threshold, outputting the genetic programming model corresponding to the risk identification factor as a risk identification model.

[0127] In this embodiment, a test set is used, and the risk identification factor value of each insurance data in the test set is substituted into the formula for calculation. When the formula calculation result is True, it is predicted that the insurance data has a risk, otherwise it is predicted that there is no risk. Based on this, a risk identification label is generated, and compared with the existing risk labels in the test set, and the risk identification accuracy of the optimal risk interpretation formula on the test set is calculated. When the risk identification accuracy is greater than or equal to the preset threshold, it indicates that the formula has a strong risk interpretation ability and can be used to guide the risk management practice of insurance business. The genetic programming model corresponding to the risk identification factor is output as the risk identification model.

[0128] If the risk identification accuracy is less than the preset threshold, it means that the model performance has not met expectations. In this case, the parameter settings of the genetic programming model, such as population size, crossover probability, mutation probability, etc., should be adjusted, and retraining and optimization should be performed.

[0129] Step S208, obtaining the insurance business data to be identified, inputting the insurance business data to be identified into the risk identification model, and obtaining the risk identification result.

[0130] Acquire new insurance business data and pre-process the new insurance business data, including removing noise data, filling missing values, standardizing data formats, etc., to obtain new insurance business data to be identified, extract numerical features, including insurance amount, premium, age, etc., extract keyword features, including occupation, medical history, family medical history, etc. Input the extracted numerical features and keyword features into the risk identification model to obtain risk identification results. Write the risk identification results, including risk scores and risk levels, into the insurance business database and store them in association with the original new insurance business data. Make business decisions based on the risk identification results in the insurance business database, conduct key audits on high-risk businesses, and quickly approve normal risk businesses to improve business processing efficiency.

[0131] This application applies genetic programming algorithms to the field of insurance risk identification, which can automatically discover complex nonlinear risk identification factors and improve the accuracy and efficiency of risk identification. At the same time, by modeling different data types separately, it fully utilizes the multidimensional characteristics of insurance data, which can better realize the automatic identification of insurance risks and further improve the accuracy and interpretability of risk identification.

[0132] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0133] The basic technologies of artificial intelligence generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning. A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed 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 may include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium may be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0134] 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.

[0135] Further references Figure 6 , as a response to the above Figure 2 In order to realize the method shown in the figure, the present application provides an embodiment of a device for mining risk identification factors, and the device embodiment is Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0136] like Figure 6 As shown, the risk identification factor mining device 600 described in this embodiment includes: an acquisition module 601, a construction module 602, an initialization module 603, a calculation module 604, an iteration module 605, a selection module 606, a test module 607 and an identification module 608. Among them:

[0137] The acquisition module 601 is used to acquire unstructured text data and numerical data in the insurance business database, perform word segmentation processing on the text data to obtain word segmentation results, extract keywords from the word segmentation results to obtain keyword data;

[0138] The construction module 602 is used to obtain a first operation formula based on the logical operator set by the keyword data, obtain a second operation formula based on the basic operator set by the numerical data, and construct a genetic programming model through the first operation formula and the second operation formula according to the insurance risk identification target;

[0139] The initialization module 603 is used to generate a formula tree population with a preset number of formula trees by random initialization as an initial formula tree population corresponding to the genetic programming model;

[0140] The calculation module 604 is used to obtain the risk label data marked in the insurance business database, divide the training set and the test set according to a preset ratio, and calculate the risk interpretation degree of each formula tree in the initial formula tree population through the training set;

[0141] The iteration module 605 is used to iteratively optimize the initial formula tree population using a genetic algorithm according to the risk interpretation degree until a preset termination condition is reached to obtain a final formula tree population;

[0142] The selection module 606 is used to select a first preset number of formula trees with the highest risk explanation degree from the final formula tree population as risk identification factors;

[0143] The testing module 607 is used to calculate the risk identification accuracy of the risk identification factor using the test set, and when the risk identification accuracy is greater than or equal to a preset threshold, output the genetic programming model corresponding to the risk identification factor as a risk identification model;

[0144] The identification module 608 is used to obtain the insurance business data to be identified, input the insurance business data to be identified into the risk identification model, and obtain the risk identification result.

[0145] It should be emphasized that in order to further ensure the privacy and security of text data and numerical data, the above text data and numerical data can also be stored in a node of a blockchain.

[0146] The mining device 600 based on the above-mentioned risk identification factors can automatically discover complex nonlinear risk identification factors by applying genetic programming algorithms to the field of insurance risk identification, thereby improving the accuracy and efficiency of risk identification. At the same time, by modeling different data types separately, it fully utilizes the multidimensional characteristics of insurance data, and can better realize the automatic identification of insurance risks, further improving the accuracy and interpretability of risk identification.

[0147] In this embodiment, the acquisition module 601 includes a word segmentation submodule, which is used to:

[0148] According to the word segmentation results, the number of occurrences of each word segment in the text data is calculated by a word frequency statistics method to obtain the word frequency of each word segment;

[0149] Using an inverse document frequency algorithm, the distribution of each word segment in the text data is calculated to obtain an inverse document frequency;

[0150] Multiply the word frequency by the inverse document frequency, calculate the weight value of each word segment, and obtain a word weight list;

[0151] Obtaining a preset word frequency threshold and a document frequency threshold, and filtering out the words in the word weight list that simultaneously satisfy the word frequency greater than or equal to the word frequency threshold and the document frequency less than or equal to the document frequency threshold as candidate keywords;

[0152] The candidate keywords are sorted from high to low according to the weight values, and the candidate keywords with a preset number of words ranked first are selected as keyword data.

[0153] The final keyword data is selected through word frequency, inverse document frequency and weight value to improve the efficiency of keyword extraction. At the same time, non-keywords with low weights can be eliminated, so that the extracted keywords have stronger explanatory power in risk identification, thereby improving the accuracy of risk identification.

[0154] In some optional implementations of this embodiment, the calculation module 604 is further used to:

[0155] A variable logistic regression model is used to fit the risk factors and risk labels represented by each of the formula trees, and the risk interpretation degree of each of the formula trees is obtained by calculating the ratio of the regression mean square to the error mean square.

[0156] Calculating the risk interpretation of the formula tree through the variable logistic regression model can improve the efficiency and accuracy of the calculation, so as to obtain the optimal formula tree population better and faster.

[0157] In some optional implementations, the iteration module 605 includes:

[0158] A selection submodule is used to select a second preset number of formula trees with the highest risk interpretation degree as parent formula trees by using a population selection algorithm according to the calculated risk interpretation degree;

[0159] A copy submodule, used for copying the parent formula tree to a new population;

[0160] A crossover submodule, used to randomly select part of the parent formula tree for pairing according to a preset crossover probability, and randomly select subtrees of the parent formula tree for exchange and recombination to generate a new crossover sub-generation formula tree;

[0161] A mutation submodule, used to perform a mutation operation on the parent generation formula tree according to a preset mutation probability to generate a new mutation child generation formula tree;

[0162] The iterative submodule is used to add the crossover child formula tree and the variant child formula tree into the new population, and continuously iterate and evolve until the preset termination condition is reached.

[0163] Through genetic operations such as selection, replication, crossover, and mutation, the formula tree population is evolved to obtain the final formula tree population, which can automatically mine deep risk identification factors that cannot be obtained from experience, enhance risk prediction capabilities, and improve risk assessment efficiency.

[0164] In some optional implementations of this embodiment, the replication submodule is further used for:

[0165] Obtaining tree structure data of the parent formula tree;

[0166] Using a depth-first search algorithm to traverse all nodes of the parent formula tree, converting the parent formula tree into a string form through a serialization operation to obtain a parent formula tree string;

[0167] Copy the parent formula tree string to a new buffer, and apply for a memory space of the same size as the parent formula tree in the memory of the new population;

[0168] Deserialize the parent formula tree string into a tree structure, and rebuild the parent formula tree;

[0169] Inserting the reconstructed parent formula tree into the memory address corresponding to the new population;

[0170] Determine whether all selected parent formula trees have been copied. If so, the copy operation is completed. Otherwise, return to continue selecting the next parent formula tree for copying until all parent formula trees are copied.

[0171] Through the copying operation, the formula tree with higher risk explanation degree is copied from the parent population to the new population, maintaining population diversity and accelerating the evolution process. At the same time, excellent genetic information is retained, providing a basis for subsequent crossover and mutation operations.

[0172] In some optional implementations of this embodiment, the cross-connect submodule is further used for:

[0173] According to a preset crossover probability, determining whether to perform a crossover operation on the offspring formula tree copied in the new population;

[0174] If yes, two formula trees are randomly selected from the new population as the first parent and the second parent;

[0175] Randomly selecting a first subtree and a second subtree in the first parent generation and the second parent generation respectively;

[0176] Determine whether the root node type of the first subtree matches the root node type of the second subtree;

[0177] If they match, the positions of the first subtree and the second subtree in their respective parent generations are exchanged to obtain a first cross-child formula tree and a cross-child formula tree.

[0178] Through crossover operations, while introducing new risk identification factor combinations, excellent genetic information is retained, thereby accelerating the evolution rate.

[0179] In some optional implementations of this embodiment, the mutation submodule is further used to:

[0180] Determine whether to perform a subtree mutation operation on the parent formula tree. If yes, randomly select a subtree of the parent formula tree, replace the subtree with a randomly generated new subtree, and obtain a first formula tree after the subtree mutation;

[0181] If the subtree mutation operation is not performed on the parent formula tree, it is determined whether the point mutation operation is performed on the parent formula tree. If so, a node of the parent formula tree is randomly selected and replaced with a new node with the same number of variables to obtain a second formula tree after the point mutation.

[0182] If the parent formula tree is not subjected to the point mutation operation, it is determined whether the parent tree is subjected to the Hoist mutation operation. If so, an initial subtree of the parent formula tree is randomly selected, a replacement subtree is randomly selected from the initial subtree, and the replacement subtree replaces the initial subtree to obtain a third formula tree after the Hoist mutation;

[0183] The first formula tree, the second formula tree and the third formula tree are used as variant child formula trees.

[0184] By mutating the parent formula tree, we can increase population diversity, improve the local search capability of the algorithm, and help to deeply explore risk identification factors.

[0185] To solve the above technical problems, the present application also provides a computer device. Figure 7 , Figure 7 This is a basic structural block diagram of the computer device in this embodiment.

[0186] The computer device 7 includes a memory 71, a processor 72, and a network interface 73 that are interconnected through a system bus. It should be noted that the figure only shows a computer device 7 with a memory 71, a processor 72, and a network interface 73, 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 (ASIC), programmable gate arrays (FPGA), digital signal processors (DSP), embedded devices, etc.

[0187] 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.

[0188] The memory 71 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 71 can be an internal storage unit of the computer device 7, such as a hard disk or memory of the computer device 7. In other embodiments, the memory 71 can also be an external storage device of the computer device 7, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. equipped on the computer device 7. Of course, the memory 71 can also include both the internal storage unit of the computer device 7 and its external storage device. In this embodiment, the memory 71 is generally used to store the operating system and various application software installed on the computer device 7, such as computer-readable instructions of the mining method of risk identification factors, etc. In addition, the memory 71 can also be used to temporarily store various types of data that have been output or are to be output.

[0189] The processor 72 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor 72 is generally used to control the overall operation of the computer device 7. In this embodiment, the processor 72 is used to run computer-readable instructions stored in the memory 71 or process data, such as computer-readable instructions for running the mining method of the risk identification factor.

[0190] The network interface 73 may include a wireless network interface or a wired network interface. The network interface 73 is generally used to establish a communication connection between the computer device 7 and other electronic devices.

[0191] By applying genetic programming algorithms to the field of insurance risk identification, complex nonlinear risk identification factors can be automatically discovered, thereby improving the accuracy and efficiency of risk identification. At the same time, by modeling different data types separately, the multidimensional characteristics of insurance data are fully utilized, which can better realize the automatic identification of insurance risks and further improve the accuracy and interpretability of risk identification.

[0192] The present application also provides another embodiment, namely, providing a computer-readable storage medium, which stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the risk identification factor mining method as described above.

[0193] By applying genetic programming algorithms to the field of insurance risk identification, complex nonlinear risk identification factors can be automatically discovered, thereby improving the accuracy and efficiency of risk identification. At the same time, by modeling different data types separately, the multidimensional characteristics of insurance data are fully utilized, which can better realize the automatic identification of insurance risks and further improve the accuracy and interpretability of risk identification.

[0194] 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 application, 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, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0195] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A method for mining risk identification factors, characterized in that: The steps include: Acquire unstructured text data and numerical data in an insurance business database, perform word segmentation processing on the text data to obtain word segmentation results, extract keywords from the word segmentation results to obtain keyword data; A first operation formula is obtained based on the logical operator set for the keyword data, a second operation formula is obtained based on the basic operator set for the numerical data, and a genetic programming model is constructed through the first operation formula and the second operation formula according to an insurance risk identification target; Generate a formula tree population with a preset number of formula trees by random initialization as an initial formula tree population corresponding to the genetic programming model; Acquire the risk label data annotated in the insurance business database, divide the data into a training set and a test set according to a preset ratio, and calculate the risk interpretation degree of each formula tree in the initial formula tree population through the training set; According to the risk interpretation degree, the initial formula tree population is iteratively optimized using a genetic algorithm until a preset termination condition is reached to obtain a final formula tree population; Selecting a first preset number of formula trees with the highest risk explanation degree from the final formula tree population as risk identification factors; Using the test set, calculating the risk identification accuracy of the risk identification factor, and when the risk identification accuracy is greater than or equal to a preset threshold, outputting the genetic programming model corresponding to the risk identification factor as a risk identification model; The insurance business data to be identified is obtained, and the insurance business data to be identified is input into the risk identification model to obtain a risk identification result.

2. The method for mining risk identification factors according to claim 1, characterized in that: The step of extracting keywords from the word segmentation results to obtain keyword data comprises: According to the word segmentation results, the number of occurrences of each word segment in the text data is calculated by a word frequency statistics method to obtain the word frequency of each word segment; Using an inverse document frequency algorithm, the distribution of each word segment in the text data is calculated to obtain an inverse document frequency; Multiply the word frequency by the inverse document frequency, calculate the weight value of each word segment, and obtain a word weight list; Obtaining a preset word frequency threshold and a document frequency threshold, and filtering out the words in the word weight list that simultaneously satisfy the word frequency greater than or equal to the word frequency threshold and the document frequency less than or equal to the document frequency threshold as candidate keywords; The candidate keywords are sorted from high to low according to the weight values, and the candidate keywords with a preset number of words ranked first are selected as keyword data.

3. The method for mining risk identification factors according to claim 1, characterized in that: The step of calculating the risk interpretation degree of each formula tree in the initial formula tree population through the training set comprises: A variable logistic regression model is used to fit the risk factors and risk labels represented by each of the formula trees, and the risk interpretation degree of each of the formula trees is obtained by calculating the ratio of the regression mean square to the error mean square.

4. The method for mining risk identification factors according to claim 1, characterized in that: The step of iteratively optimizing the initial formula tree population using a genetic algorithm according to the risk interpretation degree until a preset termination condition is reached includes: According to the calculated risk interpretation degree, a population selection algorithm is used to select a second preset number of formula trees with the highest risk interpretation degree as parent formula trees; Copying the parent generation formula tree to a new population; By using a preset crossover probability, a part of the parent generation formula tree is randomly selected for pairing, and a subtree of the parent generation formula tree is randomly selected for exchange and recombination to generate a new crossover child generation formula tree; According to a preset mutation probability, the parent generation formula tree is mutated to generate a new mutated child generation formula tree; The crossover offspring formula tree and the variant offspring formula tree are added to the new population, and the evolution is iterated continuously until the preset termination condition is reached.

5. The method for mining risk identification factors according to claim 4, characterized in that: The step of copying the parent formula tree to the new population includes: Obtaining tree structure data of the parent formula tree; Using a depth-first search algorithm to traverse all nodes of the parent formula tree, converting the parent formula tree into a string form through a serialization operation to obtain a parent formula tree string; Copy the parent formula tree string to a new buffer, and apply for a memory space of the same size as the parent formula tree in the memory of the new population; Deserialize the parent formula tree string into a tree structure, and rebuild the parent formula tree; Inserting the reconstructed parent formula tree into the memory address corresponding to the new population; Determine whether all selected parent formula trees have been copied. If so, the copy operation is completed. Otherwise, return to continue selecting the next parent formula tree for copying until all parent formula trees are copied.

6. The method for mining risk identification factors according to claim 5, characterized in that: The step of randomly selecting part of the parent formula tree for pairing by a preset crossover probability, and randomly selecting a subtree of the parent formula tree for exchange and recombination to generate a new crossover child formula tree comprises: According to a preset crossover probability, determining whether to perform a crossover operation on the offspring formula tree copied in the new population; If yes, two formula trees are randomly selected from the new population as the first parent and the second parent; Randomly selecting a first subtree and a second subtree in the first parent generation and the second parent generation respectively; Determine whether the root node type of the first subtree matches the root node type of the second subtree; If they match, the positions of the first subtree and the second subtree in their respective parent generations are exchanged to obtain a first cross-child formula tree and a cross-child formula tree.

7. The method for mining risk identification factors according to claim 4, characterized in that: The step of performing a mutation operation on the parent generation formula tree according to a preset mutation probability to generate a new mutant child generation formula tree comprises: Determine whether to perform a subtree mutation operation on the parent formula tree. If yes, randomly select a subtree of the parent formula tree, replace the subtree with a randomly generated new subtree, and obtain a first formula tree after the subtree mutation; If the subtree mutation operation is not performed on the parent formula tree, it is determined whether the point mutation operation is performed on the parent formula tree. If so, a node of the parent formula tree is randomly selected and replaced with a new node with the same number of variables to obtain a second formula tree after the point mutation; If the parent formula tree is not subjected to the point mutation operation, it is determined whether the parent tree is subjected to the Hoist mutation operation. If so, an initial subtree of the parent formula tree is randomly selected, a replacement subtree is randomly selected from the initial subtree, and the replacement subtree replaces the initial subtree to obtain a third formula tree after the Hoist mutation; The first formula tree, the second formula tree and the third formula tree are used as variant child formula trees.

8. A device for mining risk identification factors, characterized in that: include: An acquisition module is used to acquire unstructured text data and numerical data in an insurance business database, perform word segmentation processing on the text data to obtain word segmentation results, extract keywords from the word segmentation results to obtain keyword data; A construction module, configured to obtain a first operation formula based on the logical operator set for the keyword data, obtain a second operation formula based on the basic operator set for the numerical data, and construct a genetic programming model through the first operation formula and the second operation formula according to an insurance risk identification target; An initialization module, used for generating a formula tree population with a preset number of formula trees by random initialization as an initial formula tree population corresponding to the genetic programming model; A calculation module, used to obtain the risk label data marked in the insurance business database, divide the training set and the test set according to a preset ratio, and calculate the risk interpretation degree of each formula tree in the initial formula tree population through the training set; An iteration module, used to iteratively optimize the initial formula tree population using a genetic algorithm according to the risk interpretation degree until a preset termination condition is reached to obtain a final formula tree population; A selection module, used for selecting a first preset number of formula trees with the highest risk explanation degree from the final formula tree population as risk identification factors; A testing module, used to calculate the risk identification accuracy of the risk identification factor using the test set, and when the risk identification accuracy is greater than or equal to a preset threshold, output the genetic programming model corresponding to the risk identification factor as a risk identification model; The identification module is used to obtain the insurance business data to be identified, input the insurance business data to be identified into the risk identification model, and obtain the risk identification result.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the risk identification factor mining method according to any one of claims 1 to 7 when executing the computer-readable instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the method for mining risk identification factors according to any one of claims 1 to 7 are implemented.

Images