Risk level analysis method, device, equipment and medium

Abstract

The application relates to the technical field of artificial intelligence, and discloses a risk level analysis method, device, equipment and medium, which comprises the following steps: obtaining historical feature data of a plurality of customers, preprocessing the historical feature data to obtain a training data set; performing feature screening on the training data set based on a pre-constructed machine learning model to obtain core feature data in the training data set; dividing the customers into groups based on the core feature data, generating group risk features of each customer group based on a pre-established large language model; obtaining feature data of a to-be-evaluated customer, generating a preliminary risk evaluation level of the to-be-evaluated customer based on the pre-established large language model; and determining a business risk level of the to-be-evaluated customer based on the preliminary risk evaluation level and a preset scoring function. The application can be applied to the insurance risk evaluation scene of financial technology, and solves the technical problems that the insurance risk evaluation in the prior art lacks comprehensiveness, adaptability and interpretability.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a risk level analysis method, apparatus, equipment, and medium. Background Technology

[0002] Currently, risk assessment and customer segmentation in the insurance sector primarily rely on statistical models based on historical data, expert rule systems, or a simple combination of both. However, when faced with real-world scenarios such as business expansion and the influx of new customers, existing technological systems exhibit significant limitations in terms of comprehensiveness, adaptability, and interpretability, specifically as follows: 1. Insufficient risk assessment capability for new customers with high growth but low information: Traditional risk assessment models heavily rely on historical customer data (such as past claims records and policy duration) for modeling and prediction. When insurance products lower the entry threshold and a large number of new customers lacking historical data enter the market, these models face the contradiction of "high feature dimensionality and few effective samples," which can easily lead to overfitting or assessment failure, making it impossible to make reliable risk judgments for "cold start" customers.

[0003] 2. Lack of in-depth analysis and integration of group risk characteristics: Existing methods typically focus on individual customer characteristics, neglecting the common risk patterns inherent in their respective groups (such as those in the same region or with similar farming scale). Even when introducing group statistical indicators (such as regional average payout rates), these are limited to simple numerical indicators and cannot extract textual, interpretable descriptions of group behavior and common risk characteristics from multi-dimensional data, resulting in a lack of dimensionality in risk assessment information utilization.

[0004] 3. Poor model interpretability and disconnect between business rules and data models: Many risk control systems based on machine learning (such as deep learning and complex ensemble models) have certain predictive performance, but their decision-making process is often a "black box," making it difficult to explain the scoring criteria to business personnel. While pure expert rule systems are interpretable, they rely on subjective experience, making it difficult to adapt to dynamically changing data environments, and rule iteration is lagging. Existing technologies lack a mechanism for organically integrating data-driven results with business knowledge and achieving two-way calibration.

[0005] 4. Failure to effectively utilize the deep reasoning and feature generation capabilities of Large Language Models (LLM): In the current risk control field, LLM is mostly used for text sentiment analysis or simple classification tasks, and its potential in integrating multi-source information, performing common-sense reasoning, and generating generalized text features has not been fully realized. Especially when new customer groups lack quantitative labels, existing technologies lack a systematic approach to using LLM for qualitative-to-quantitative risk conversion.

[0006] In summary, existing methods for assessing the risk level of insurance customers suffer from a lack of comprehensiveness, adaptability, and interpretability. Summary of the Invention

[0007] In view of the shortcomings of the prior art, the present invention provides a risk level analysis method, apparatus, computer equipment and medium to solve the technical problems of lack of comprehensiveness, adaptability and interpretability in insurance risk assessment in the prior art.

[0008] Firstly, a risk level analysis method is provided, including: Obtain historical feature data of several customers, preprocess the historical feature data to obtain a training dataset; Based on a pre-built machine learning model, feature filtering is performed on the training dataset to obtain core feature data from the training dataset; Based on the core feature data, customers are segmented into groups, and based on a pre-established large language model, group risk characteristics of each customer group are generated. Obtain the characteristic data of the customer to be evaluated, and generate a preliminary risk assessment level for the customer to be evaluated based on the pre-established large language model; Based on the preliminary risk assessment level and the preset scoring function, the business risk level of the customer to be assessed is determined.

[0009] Secondly, a risk level analysis device is provided, comprising: The data acquisition module is used to acquire historical feature data of several customers, preprocess the historical feature data, and obtain a training dataset. The core feature filtering module is used to filter features in the training dataset based on a pre-built machine learning model to obtain core feature data in the training dataset. The group risk feature acquisition module is used to divide customers into groups based on the core feature data and generate group risk features for each customer group based on a pre-established large language model. The preliminary risk assessment module is used to acquire the characteristic data of the customer to be assessed and generate the preliminary risk assessment level of the customer to be assessed based on the pre-established large language model. The risk level determination module is used to determine the business risk level of the customer to be assessed based on the preliminary risk assessment level and a preset scoring function.

[0010] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the risk level analysis method described above.

[0011] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the aforementioned risk level analysis method.

[0012] The above-mentioned risk level analysis method, apparatus, computer equipment, and storage medium can acquire historical characteristic data of several customers, preprocess the historical characteristic data to obtain a training dataset; based on a pre-built machine learning model, feature filtering is performed on the training dataset to obtain core characteristic data; based on the core characteristic data, customers are grouped, and based on a pre-established large language model, group risk characteristics of each customer group are generated; characteristic data of the customer to be evaluated is acquired, and based on the pre-established large language model, a preliminary risk assessment level of the customer to be evaluated is generated; based on the preliminary risk assessment level and a preset scoring function, the business risk level of the customer to be evaluated is determined. This invention innovatively constructs a multi-level risk assessment system of "data filtering—group analysis—large language model reasoning—rule fusion". Unlike traditional methods that rely solely on historical data or static rules, this invention first utilizes machine learning to objectively filter core individual characteristics from massive datasets. Second, it introduces a large language model for the first time to perform deep textual feature mining of customer groups, transforming difficult-to-quantify group risk patterns into usable textual features. Third, it leverages the comprehensive reasoning capabilities of the large language model to conduct preliminary, fine-grained level assessments of customers. Finally, through a transparent scoring function that combines statistical weights and business expert experience, all the above information is integrated to output the final business level. This invention achieves a progressive analysis from micro-level individual data to macro-level group insights, and then to intelligent comprehensive judgment, significantly improving the comprehensiveness, adaptability, and interpretability of risk assessment. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a schematic diagram of an application environment for a risk level analysis method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a risk level analysis method in one embodiment of the present invention; Figure 3 yes Figure 1 A schematic diagram of a specific implementation method for step S10; Figure 4 yes Figure 1 A schematic diagram of a specific implementation method for step S20; Figure 5 yes Figure 1 A schematic diagram of a specific implementation method for step S30; Figure 6 yes Figure 1 A schematic diagram of a specific implementation of step S40; Figure 7 yes Figure 1 A schematic diagram of a specific implementation method for step S50; Figure 8 This is a schematic diagram of a risk level analysis device in one embodiment of the present invention; Figure 9 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention; Figure 10 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] The risk level analysis method provided in this invention can be applied to, for example... Figure 1 In this application environment, the client communicates with the server via a network. The server can obtain historical feature data of several clients through the client, preprocess the historical feature data, and obtain a training dataset. Based on a pre-built machine learning model, the training dataset is feature-filtered to obtain core feature data. Based on this core feature data, customers are grouped, and based on a pre-established large language model, group risk characteristics for each customer group are generated. Feature data of the customer to be evaluated is obtained, and based on the pre-established large language model, a preliminary risk assessment level is generated for the customer to be evaluated. Based on the preliminary risk assessment level and a preset scoring function, the business risk level of the customer to be evaluated is determined. This invention innovatively constructs a multi-level risk assessment system of "data filtering—group analysis—large language model reasoning—rule fusion." Unlike traditional methods that rely solely on historical data or static rules, this invention first uses machine learning to objectively filter core individual characteristics from massive features; secondly, it introduces a large language model for the first time to perform deep textual feature mining of customer groups, transforming difficult-to-quantify group risk patterns into usable textual features; thirdly, it utilizes the comprehensive reasoning capabilities of the large language model to conduct a preliminary, fine-grained level assessment of customers; finally, through a transparent scoring function combining statistical weights and business expert experience, all the above information is fused to output the final business level. This invention achieves a progressive analysis, moving from micro-level individual data to macro-level group insights, and finally to intelligent comprehensive judgment, significantly improving the comprehensiveness, adaptability, and interpretability of risk assessment. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0017] Please see Figure 2 As shown, Figure 2 A flowchart illustrating the risk level analysis method provided in this embodiment of the invention includes the following steps: S10: Obtain historical feature data of several customers, preprocess the historical feature data to obtain a training dataset.

[0018] In this embodiment, historical feature data of several customers can be obtained from the insurance business system, customer-filled information, historical claims records and external data sources. Then, by preprocessing the historical feature data, a training dataset is obtained to ensure that the training dataset can achieve a good training effect.

[0019] For example, approximately 100 initial characteristic factors can be collected from insurance business systems, customer-reported information, historical claims records, and external data sources. These initial characteristic factors may include data such as historical risk indicators, business scale and structure, breeding conditions and environment, regional and industry factors, technology and experience, etc. Among them, historical risk indicators may include data such as the previous year's loss ratio, historical accident frequency, and average claim amount for the customer (or region); business scale and structure may include data such as the number of insured fattening pigs (i.e., insurance scale), customer type (individual / enterprise), pig farm type (self-breeding and self-raising, contract farming, etc.), and whether the person in charge of the enterprise personally participates in daily management; breeding conditions and environment may include data such as whether a backup generator is equipped, the degree of automation of the feeding and drinking water system, whether slatted floors are used, breeding density, ventilation conditions, and disease prevention facilities; regional and industry factors may include data such as the location of the farm, the historical average loss ratio of the region, climate characteristics, and the history of epidemics in the surrounding area; and technology and experience may include data such as the average years of experience of breeding personnel, whether standardized breeding techniques are used, and whether there is cooperation with large-scale breeding enterprises.

[0020] After the raw data collection is complete, the collected raw data undergoes preprocessing. This includes, for example,... Figure 3 As shown, step S10, which involves acquiring historical feature data from several customers and preprocessing the historical feature data to obtain a training dataset, includes the following steps: S11: Obtain historical characteristic data for several customers.

[0021] S12: Label the historical feature data to quantify the historical feature data.

[0022] S13: Perform Z-score standardization on the historical feature data after numerical processing, and use the standardized dataset as the training dataset.

[0023] For steps S11-S13, after obtaining the original data, label encoding is applied to each discrete feature factor of the original data to convert it into numerical form to meet the requirements of model input. For example, each discrete feature factor can be assigned an independent integer, and a fully numerical feature matrix can be output, making the dataset conform to the numerical input specifications of models such as linear regression, XGBoost, LR, and SVM. Then, Z-score normalization is performed on the feature factors, performing a linear transformation based on the feature mean and standard deviation to eliminate interference caused by inconsistencies in the dimensions and numerical scales of the features, unifying the feature distribution to a standard normal distribution. After standardization, the features have horizontal comparability, which can effectively accelerate the convergence of gradient-based models, improve the accuracy of distance-based algorithms, and optimize the overall model fitting and generalization performance.

[0024] S20: Based on the pre-built machine learning model, perform feature filtering on the training dataset to obtain the core feature data in the training dataset.

[0025] After obtaining the preprocessed training data, in order to address the problem of model overfitting or insufficient training caused by "a large number of initial features (about 100) and relatively limited historical customer sample data", this application adopts the XGBoost machine learning model to evaluate and screen the importance of features, and then objectively screen core personal features from the massive features.

[0026] For example, such as Figure 4 As shown, in step S20, which involves feature filtering of the training dataset based on a pre-built machine learning model to select the core feature data in the training dataset, the following steps are included: S21: Train the pre-built machine learning model based on the training dataset.

[0027] S22: Based on the trained machine learning model, obtain feature importance scores for various types of historical feature data.

[0028] S23: Sort the importance scores of each feature and select the core feature data in the training dataset based on the sorting results.

[0029] In this embodiment, model training and importance assessment are performed first. Specifically, preprocessed historical customer data can be used as the training set, with approximately 100 feature factors as input features (X) and customer payout ratio as the prediction target (y), to train an XGBoost regression model. After training, the feature importance scores output by the model are extracted. Next, feature ranking and selection are performed, specifically, all features can be sorted in descending order of their importance scores. Based on a balance between business needs and model performance, the top K features (e.g., K=12) are selected as the initial core feature factor set. These K factors constitute the basic features for the individual dimensions of subsequent analysis.

[0030] S30: Based on the core feature data, customers are divided into groups, and based on the pre-established large language model, group risk characteristics of each customer group are generated.

[0031] In this embodiment, a Large Language Model (LLM) is introduced to perform deep textual feature mining on customer groups, transforming difficult-to-quantify group risk patterns into usable textual features. This embodiment addresses the challenge in the insurance industry of scarce historical data on new customer groups and the inadequacy of individual characteristics to fully reflect the common risk characteristics of their respective groups. It creatively proposes using a Large Language Model to generate "textualized group risk features." This method transforms structured data of customer groups segmented according to core dimensions into natural language descriptions, inputting them into the Large Language Model. The Model is then instructed to extract and summarize the core risk characteristics, common behavioral patterns, and potential risk points of the group, generating "soft features" that are difficult to obtain directly using traditional quantitative models. These features are then solidified as new assessment dimensions after expert screening. This fully leverages the convenience of LLM and solves the problem of insufficient risk representation dimensions in small sample or cold-start scenarios.

[0032] In order to obtain the risk characteristics of a group, such as Figure 5 As shown, a specific scheme for obtaining group risk characteristics is provided. In S30, that is, after dividing customers into groups based on the core feature data, the group risk characteristics of each customer group are generated based on a pre-established large language model, including the following steps: S31: Based on the core feature data, customers are divided into groups.

[0033] S32: Generate a semi-structured data description based on the historical characteristic data of customers in each customer group.

[0034] S33: Using the semi-structured data description as prompt words, input the prompt words into the pre-established large language model to generate the group risk characteristics of the customer group.

[0035] For example, when acquiring group risk, the first step is to segment the customer group. Specifically, this can be done by segmenting historical customers based on key dimensions in the "core characteristic factor set" (such as "insured size," "region," and "customer type"). For instance, "insured size" can be divided into four levels: small (e.g., <3000 heads), relatively small (3000-6000 heads), relatively large (6000-10000 heads), and large (>10000 heads). Customers within each level constitute a group.

[0036] After the customer groups are segmented, for each segmented customer group, all initial characteristic data (approximately 100 items) of all customers within that group are summarized and organized to form a semi-structured data description. This description is then input as a prompt into a large language model (e.g., the qwen3-235b large language model). The large language model then summarizes and extracts a concise text description based on the group data. This text should summarize the group's core risk characteristics, common behavioral patterns, or potential risk points. For example, after inputting data for "small-scale insured groups," the large language model might output: "This group is mainly composed of individual farmers, with generally low levels of automation in their farming facilities, weak resilience to extreme weather, and historical claims data showing significant management volatility."

[0037] After obtaining the text descriptions, the group characteristics can be manually screened and solidified. Specifically, the numerous group text descriptions generated by the large language model (each group corresponds to one paragraph) can be analyzed and manually screened by insurance experts or senior underwriters. Based on the distinguishability of the text information, its relevance to business risks, and the clarity of expression, M of the most representative text descriptions (M=3 in this embodiment) are finally selected as the newly added "group risk characteristics." These characteristics are presented as text strings, alongside the customer's core personal characteristics.

[0038] S40: Obtain the characteristic data of the customer to be evaluated, and generate a preliminary risk assessment level for the customer to be evaluated based on the pre-established large language model.

[0039] In this embodiment, the comprehensive reasoning capabilities of a large language model are utilized to conduct a preliminary, fine-grained risk assessment of customers. Specifically, multi-source data, such as the core characteristic data of the customer to be assessed and the group risk characteristics, are first input into the large language model. Then, the large language model scores each characteristic data individually, outputting a score corresponding to each characteristic data. Afterward, the scores corresponding to each characteristic data are weighted by preset weight parameters, thereby outputting a score data that can preliminarily assess the customer's risk level.

[0040] like Figure 6 As shown, a specific preliminary risk assessment scheme is provided. In S40, namely, obtaining the characteristic data of the customer to be assessed and generating the preliminary risk assessment level of the customer to be assessed based on the pre-established large language model, the following steps are included: S41: Obtain the characteristic data of the customer to be evaluated and the group risk characteristics of the group corresponding to the customer to be evaluated.

[0041] S42: Generate the core feature data of the customer to be evaluated based on the feature data.

[0042] S43: Input the core characteristic data and group risk characteristics of the customer to be evaluated into the large language model, and output the preliminary risk assessment level of the customer to be evaluated.

[0043] In this embodiment, the customer feature vector is first constructed. For example, for a customer to be assessed (including new customers), their 12 core personal characteristic factors and 3 textualized group risk characteristics corresponding to their group are organized into a complete feature description. After the feature vector is constructed, a preliminary risk level assessment is performed using a large language model. For example, the aforementioned feature description information is input into the large language model, and explicit instructions are designed to enable the large language model to autonomously conduct insurance risk assessment. Based on all the provided feature information, the model assesses the customer's overall risk level. The assessment output is an integer risk level, for example, divided into ten levels from 1 to 10, where a higher level value represents a higher risk level for the customer. The final output risk assessment level can serve as a new, comprehensive derived feature for the customer, i.e., the preliminary large language model risk level, which can provide a reliable basis for the subsequent final risk rating.

[0044] S50: Based on the preliminary risk assessment level and the preset scoring function, determine the business risk level of the customer to be assessed.

[0045] In this embodiment, data-driven quantitative indicators are deeply integrated with insurance industry expertise to construct a final risk scoring and grading system that is objective, transparent, and business-interpretable. By integrating the core characteristics of the customer to be assessed with the preliminary risk assessment level output by the big data model, a business risk level that can directly guide underwriting decisions is transformed.

[0046] like Figure 7 As shown, a specific business risk level determination scheme is provided. In S50, that is, determining the business risk level of the customer to be assessed based on the preliminary risk assessment level and the preset scoring function, includes the following steps: S51: Input the core characteristic data of the customer to be evaluated and the preliminary risk assessment level into the scoring function to obtain the risk score of the customer to be evaluated.

[0047] S52: Based on the risk score of the customer to be evaluated, determine the scoring threshold range to which the customer to be evaluated belongs.

[0048] S53: Based on the mapping relationship between the scoring threshold range and the business risk level, determine the business risk level of the customer to be evaluated.

[0049] In some embodiments, the construction of the scoring function includes: Based on the core feature data in the training dataset, determine the statistical weights of the core feature data; Based on the statistical weights of the core feature data, the statistical weights of the preliminary risk assessment level are determined; The scoring function is constructed based on the statistical weights of the core feature data and the statistical weights of the preliminary risk assessment level.

[0050] Specifically, the features ultimately used to construct the scoring function consist of two parts: the K core personal characteristic factors selected above, and the preliminary risk assessment level generated by the large language model. The scoring function can then be constructed using a linearly weighted or piecewise weighted function. For example, a specific scoring function could be: ∑ i=1 12 (w i x i )+W R, where x i It is the i-th core feature data; w i This refers to the weight corresponding to the core feature data. The weight can be determined jointly based on the statistical analysis of historical data and the experience of business experts. R is the preliminary risk assessment level, and W is the weight of the preliminary risk assessment level.

[0051] Among them, the weight (w) i The determination of weights W and W employs a dual mechanism of "data statistical calibration + business expert verification." For example, based on historical customer data, regression analysis and correlation analysis with business objectives (such as loss ratio) can be used to calculate the initial statistical weights of each feature. Subsequently, insurance underwriting experts and aquaculture industry experts review and adjust the statistical weights based on their experience. Optionally, to ensure the crucial role of the comprehensive reasoning conclusions of the large language model, the weight W can be set to a value significantly higher than that of most other weights W. i The higher value of W (preferably 35% of the sum of all absolute weights in this embodiment) gives the fusion judgment of the large language model a higher decision weight in the evaluation of new customer groups with limited data.

[0052] Furthermore, when determining the final business risk level, a threshold range can be determined first, and then a fixed mapping relationship can be established between the risk score of the customer to be assessed and the business risk level, thereby enabling the automatic lookup of the corresponding business risk level based on the risk score.

[0053] Exemplarily, when determining the threshold intervals, a certain number (e.g., the past three years) of "known customer" samples with risk statuses that have been verified by the market can be selected, and the risk score S thereof is calculated using the scoring function. Afterwards, the threshold intervals are divided in combination with the business, for example, the number of final risk等级 levels can be jointly determined by business experts. In this embodiment, the business risk levels are divided into 5 levels, where level 1 represents the lowest risk and level 5 represents the highest risk. The scoring threshold intervals (T1, T2, T3, T4) for dividing different risk levels are determined according to the proportion of customers at each level required by the business (e.g., it is desired to control the highest-risk customers within 5%), and with reference to the distribution (e.g., quantiles) of the historical sample scores S. When S<T1, the customer is rated as "level 1" (high-quality customer); when T1≤S<T2, the customer is rated as "level 2" (standard customer); when T2≤S<T3, the customer is rated as "level 3" (attention-required customer); when T3≤S<T4, the customer is rated as "level 4" (high-risk customer); when S≥T4, the customer is rated as "level 5" (extremely high-risk / cautious underwriting customer).

[0054] When performing the mapping relationship solidification process, the corresponding relationship between the determined scoring threshold intervals (T1, T2, T3, T4) and the business risk levels (level 1 to level 5) is stored in the system rule base as a solidified mapping rule. For any new customer subsequently, the system will automatically complete the level determination according to this rule.

[0055] In this embodiment, in the final rating stage, a weight determination and scoring mapping mechanism that takes into account both objective data and subjective experience is designed. First, the feature weights are not purely determined by data statistics, but a dual mechanism of "data-driven initial determination + business expert verification and correction" is adopted, which ensures the consistency between the model result and industry cognition. Second, the "preliminary risk level" feature generated by the large language model is specifically assigned a higher decision weight (e.g., accounting for 35% of the total weight), placing the comprehensive reasoning capability of the large language model at the core of decision-making, which highlights the priority reliance on intelligent judgment under the condition of limited data. Third, the scoring thresholds are delineated based on the historical sample distribution and clear business objectives (e.g., the proportion of each level), so that the final risk level mapping has clear business meanings and a controllable risk distribution. The finally constructed scoring function not only has stronger prediction performance, but also features a transparent process, interpretable results, and key parameters that can be intervened by the business.

[0056] By integrating XGBoost feature screening and group text feature mining of the large language model, the problem of lack of information and easy failure of the model during the "cold start" of high-increment new customers is effectively solved; at the same time, the LLM comprehensive reasoning is introduced to generate a preliminary risk level, and combined with the transparent and interpretable scoring function of "statistical weight + business rule", the discrimination, stability and business adoption rate of risk assessment are significantly improved.

[0057] Tests showed that on the real test set, the KS value for new customers (cold start) improved from 0.21 to 0.63, representing a 200% increase in discriminative ability. Furthermore, after removing the group text features generated by LLM, the KS value plummeted from 0.63 to 0.42, demonstrating the crucial role of the group feature mining module. Within the statistical time period, the actual payout rate for high-risk (level 4-5) customers showed a significant downward trend.

[0058] As can be seen, in the above scheme, historical feature data of several customers is acquired, preprocessed, and used to obtain a training dataset. Based on a pre-built machine learning model, feature filtering is performed on the training dataset to identify core feature data. Based on this core feature data, customers are grouped, and based on a pre-established large language model, group risk characteristics are generated for each customer group to obtain feature data for the customer to be evaluated. Based on the pre-established large language model, a preliminary risk assessment level is generated for the customer to be evaluated. Based on the preliminary risk assessment level and a preset scoring function, the business risk level of the customer to be evaluated is determined. This innovatively constructs a multi-layered risk assessment system of "data filtering—group analysis—large language model inference—rule fusion". Unlike traditional methods that rely solely on historical data or static rules, this approach first utilizes machine learning to objectively filter core individual characteristics from massive datasets. Second, it introduces a large language model for the first time to perform deep textual feature mining of customer groups, transforming difficult-to-quantify group risk patterns into usable textual features. Third, it leverages the comprehensive reasoning capabilities of the large language model to conduct preliminary, fine-grained risk assessments. Finally, a transparent scoring function combining statistical weights and business expert experience integrates all the above information to output the final business rating. This achieves a progressive analysis from micro-level individual data to macro-level group insights, and then to intelligent comprehensive judgment, significantly improving the comprehensiveness, adaptability, and interpretability of risk assessment.

[0059] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0060] In one embodiment, a risk level analysis device is provided, which should... Figure 8 As shown, the risk level analysis device includes a data acquisition module 101, a core feature screening module 102, a group risk feature acquisition module 103, a preliminary risk assessment module 104, and a risk level determination module 105. Detailed descriptions of each functional module are as follows: The data acquisition module 101 is used to acquire historical feature data of several customers, preprocess the historical feature data, and obtain a training dataset.

[0061] The core feature filtering module 102 is used to filter features in the training dataset based on a pre-built machine learning model to obtain core feature data in the training dataset.

[0062] The group risk feature acquisition module 103 is used to divide customers into groups based on the core feature data and generate group risk features for each customer group based on a pre-established large language model.

[0063] The preliminary risk assessment module 104 is used to acquire the characteristic data of the customer to be assessed and generate the preliminary risk assessment level of the customer to be assessed based on the pre-established large language model.

[0064] The risk level determination module 105 is used to determine the business risk level of the customer to be assessed based on the preliminary risk assessment level and the preset scoring function.

[0065] In one embodiment, the data acquisition module 101 is specifically used for: Obtain historical characteristic data of several customers; The historical feature data is tagged and encoded to quantify the historical feature data; The historical feature data after numerical processing is standardized by Z-score, and the standardized dataset is used as the training dataset.

[0066] In one embodiment, the core feature filtering module 102 is specifically used for: Based on the training dataset, the pre-built machine learning model is trained; Based on the trained machine learning model, feature importance scores are obtained for various types of historical feature data. The importance scores of each feature are ranked, and the core feature data in the training dataset are selected based on the ranking results.

[0067] In one embodiment, the group risk characteristic acquisition module 103 is specifically used for: Based on the core feature data, customers are segmented into several customer groups. Based on the historical characteristic data of customers in each customer group, a semi-structured data description is generated; Using the semi-structured data description as prompt words, the prompt words are input into the pre-established large language model to generate the group risk characteristics of the customer group.

[0068] In one embodiment, the preliminary risk assessment module 104 is specifically used for: Obtain the characteristic data of the customer to be evaluated and the group risk characteristics of the group corresponding to the customer to be evaluated; Generate the core characteristic data of the customer to be evaluated based on the aforementioned characteristic data; The core characteristic data and group risk characteristics of the customer to be evaluated are input into the large language model, and the preliminary risk assessment level of the customer to be evaluated is output.

[0069] In one embodiment, the preliminary risk assessment module 104 is further configured to: The core characteristic data of the customer to be evaluated and the preliminary risk assessment level are input into the scoring function to obtain the risk score of the customer to be evaluated; Based on the risk score of the customer to be evaluated, determine the scoring threshold range to which the customer to be evaluated belongs; Based on the mapping relationship between the scoring threshold range and the business risk level, the business risk level of the customer to be evaluated is determined.

[0070] In one embodiment, the preliminary risk assessment module 104 is further configured to: Based on the core feature data in the training dataset, determine the statistical weights of the core feature data; Based on the statistical weights of the core feature data, the statistical weights of the preliminary risk assessment level are determined; The scoring function is constructed based on the statistical weights of the core feature data and the statistical weights of the preliminary risk assessment level.

[0071] This invention provides a risk level analysis device. It acquires historical characteristic data from several customers, preprocesses the historical characteristic data to obtain a training dataset, and then filters the training dataset based on a pre-built machine learning model to obtain core characteristic data. Based on the core characteristic data, customers are grouped, and based on a pre-established large language model, group risk characteristics of each customer group are generated. The device then acquires characteristic data of the customer to be assessed, and based on the pre-established large language model, generates a preliminary risk assessment level for the customer to be assessed. Based on the preliminary risk assessment level and a preset scoring function, the business risk level of the customer to be assessed is determined. This innovatively constructs a multi-level risk assessment system of "data filtering—group analysis—large language model reasoning—rule fusion". Unlike traditional methods that rely solely on historical data or static rules, this invention first utilizes machine learning to objectively filter core individual characteristics from massive datasets. Second, it introduces a large language model for the first time to perform deep textual feature mining of customer groups, transforming difficult-to-quantify group risk patterns into usable textual features. Third, it leverages the comprehensive reasoning capabilities of the large language model to conduct preliminary, fine-grained level assessments of customers. Finally, through a transparent scoring function that combines statistical weights and business expert experience, all the above information is integrated to output the final business level. This achieves a progressive analysis from micro-level individual data to macro-level group insights, and then to intelligent comprehensive judgment, significantly improving the comprehensiveness, adaptability, and interpretability of risk assessment.

[0072] Specific limitations regarding the risk level analysis device can be found in the limitations of the risk level analysis method described above, and will not be repeated here. Each module in the aforementioned risk level analysis device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0073] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a risk level analysis method on the server side.

[0074] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 10 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a risk level analysis method on the client side.

[0075] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Obtain historical feature data of several customers, preprocess the historical feature data to obtain a training dataset; Based on a pre-built machine learning model, feature filtering is performed on the training dataset to obtain core feature data from the training dataset; Based on the core feature data, customers are segmented into groups, and based on a pre-established large language model, group risk characteristics of each customer group are generated. Obtain the characteristic data of the customer to be evaluated, and generate a preliminary risk assessment level for the customer to be evaluated based on the pre-established large language model; Based on the preliminary risk assessment level and the preset scoring function, the business risk level of the customer to be assessed is determined.

[0076] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Obtain historical feature data of several customers, preprocess the historical feature data to obtain a training dataset; Based on a pre-built machine learning model, feature filtering is performed on the training dataset to obtain core feature data from the training dataset; Based on the core feature data, customers are segmented into groups, and based on a pre-established large language model, group risk characteristics of each customer group are generated. Obtain the characteristic data of the customer to be evaluated, and generate a preliminary risk assessment level for the customer to be evaluated based on the pre-established large language model; Based on the preliminary risk assessment level and the preset scoring function, the business risk level of the customer to be assessed is determined.

[0077] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0078] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0079] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0080] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0081] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A risk level analysis method, characterized in that, include: Obtain historical feature data of several customers, preprocess the historical feature data to obtain a training dataset; Based on a pre-built machine learning model, feature filtering is performed on the training dataset to obtain core feature data from the training dataset; Based on the core feature data, customers are segmented into groups, and based on a pre-established large language model, group risk characteristics of each customer group are generated. Obtain the characteristic data of the customer to be evaluated, and generate a preliminary risk assessment level for the customer to be evaluated based on the pre-established large language model; Based on the preliminary risk assessment level and the preset scoring function, the business risk level of the customer to be assessed is determined.

2. The risk level analysis method as described in claim 1, characterized in that, The step of acquiring historical feature data from several customers, preprocessing the historical feature data to obtain a training dataset, includes: Obtain historical characteristic data of several customers; The historical feature data is tagged and encoded to quantify the historical feature data; The historical feature data after numerical processing is standardized by Z-score, and the standardized dataset is used as the training dataset.

3. The risk level analysis method as described in claim 1, characterized in that, The pre-built machine learning model performs feature filtering on the training dataset to select core feature data from the training dataset, including: Based on the training dataset, the pre-built machine learning model is trained; Based on the trained machine learning model, feature importance scores are obtained for various types of historical feature data. The importance scores of each feature are ranked, and the core feature data in the training dataset are selected based on the ranking results.

4. The risk level analysis method as described in claim 1, characterized in that, After segmenting customers based on the core feature data, the group risk characteristics of each customer group are generated based on a pre-established large language model, including: Based on the core feature data, customers are segmented into several customer groups. Based on the historical characteristic data of customers in each customer group, a semi-structured data description is generated; Using the semi-structured data description as prompt words, the prompt words are input into the pre-established large language model to generate the group risk characteristics of the customer group.

5. The risk level analysis method as described in claim 1, characterized in that, The process of acquiring the characteristic data of the customer to be assessed and generating a preliminary risk assessment level for the customer based on the pre-established large language model includes: Obtain the characteristic data of the customer to be evaluated and the group risk characteristics of the group corresponding to the customer to be evaluated; Generate the core characteristic data of the customer to be evaluated based on the aforementioned characteristic data; The core characteristic data and group risk characteristics of the customer to be evaluated are input into the large language model, and the preliminary risk assessment level of the customer to be evaluated is output.

6. The risk level analysis method as described in claim 5, characterized in that, The process of determining the business risk level of the customer to be assessed based on the preliminary risk assessment level and a preset scoring function includes: The core characteristic data of the customer to be evaluated and the preliminary risk assessment level are input into the scoring function to obtain the risk score of the customer to be evaluated; Based on the risk score of the customer to be evaluated, determine the scoring threshold range to which the customer to be evaluated belongs; Based on the mapping relationship between the scoring threshold range and the business risk level, the business risk level of the customer to be evaluated is determined.

7. The risk level analysis method as described in claim 6, characterized in that, The construction of the scoring function includes: Based on the core feature data in the training dataset, determine the statistical weights of the core feature data; Based on the statistical weights of the core feature data, the statistical weights of the preliminary risk assessment level are determined; The scoring function is constructed based on the statistical weights of the core feature data and the statistical weights of the preliminary risk assessment level.

8. A risk level analysis device, characterized in that, include: The data acquisition module is used to acquire historical feature data of several customers, preprocess the historical feature data, and obtain a training dataset. The core feature filtering module is used to filter features in the training dataset based on a pre-built machine learning model to obtain core feature data in the training dataset. The group risk feature acquisition module is used to divide customers into groups based on the core feature data and generate group risk features for each customer group based on a pre-established large language model. The preliminary risk assessment module is used to acquire the characteristic data of the customer to be assessed and generate the preliminary risk assessment level of the customer to be assessed based on the pre-established large language model. The risk level determination module is used to determine the business risk level of the customer to be assessed based on the preliminary risk assessment level and a preset scoring function.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the risk level analysis method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the risk level analysis method as described in any one of claims 1 to 7.

Images