Intelligent case division method and system for unhealthy assets

Abstract

The invention provides an intelligent case division method and system for unhealthy assets, and the method comprises the steps: carrying out the preprocessing of multi-source unhealthy asset data, constructing a case feature set of the unhealthy assets based on the preprocessed data, carrying out the classification and grading, predicting the risk score of the unhealthy assets through a pre-trained dynamic repayment capability prediction model, and obtaining the risk score of the unhealthy assets. Constructing a borrower risk portrait of the unhealthy assets, determining case types of the unhealthy assets based on the borrower risk portrait and preset features, determining initial case division strategies of the unhealthy assets based on the case types and a preset collection strategy library, and optimizing the initial case division strategies through a reinforcement learning mode, and on the basis of the optimized case division strategy, performing case division on the non-performing assets. The method can reduce the manual intervention of the case division link of the unhealthy assets, improves the disposal efficiency of the unhealthy assets, greatly shortens the disposal period of the unhealthy assets, and reduces the unnecessary disposal cost of the unhealthy assets.

Description

Technical Field

[0001] The present invention relates to the technical field of non-performing asset management, and in particular to a method and system for intelligently dividing non-performing assets. Background Art

[0002] Non-performing assets (NPLs) refer to assets such as loans and accounts receivable that are acquired by financial institutions or enterprises during the course of their operations and for which principal and interest cannot be recovered on time. Traditional NPL case allocation methods primarily utilize centralized litigation management platforms or basic case allocation systems. Centralized litigation management platforms enable batch filing and unified allocation of legal resources within a single court, but they can easily lead to case backlogs and prevent optimal allocation of judicial resources. Basic case allocation systems can allocate cases based on simple rules (such as territorial jurisdiction), but they lack dynamic adjustment capabilities and fail to account for the courts' real-time workload. Consequently, traditional NPL case allocation methods suffer from information fragmentation, low efficiency, and high labor costs. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a method and system for intelligent non-performing asset division to alleviate the above-mentioned problems existing in the existing non-performing asset division technology.

[0004] In the first aspect, an embodiment of the present invention provides a method for intelligent case classification of non-performing assets, comprising: obtaining case information and dynamic behavior data of non-performing assets from multiple different data sources, and pre-processing the case information and the dynamic behavior data; wherein, the case information includes basic information of the borrower, loan information, overdue information, document information, judicial data, credit data and asset data, and the dynamic behavior data includes repayment record, overdue history and collection response; constructing a case feature set of non-performing assets based on the pre-processed data, and classifying and grading the case feature set; wherein, the features in the case feature set include basic case information of the corresponding non-performing assets, debtor information, asset status, etc. The method comprises the following steps: first, predicting the classified and graded feature set through a pre-trained dynamic repayment ability prediction model to obtain a risk score for the non-performing assets, and constructing a risk profile of the borrower of the non-performing assets based on the external supplementary information of the non-performing assets and the risk score; second, constructing a risk profile of the borrower of the non-performing assets based on the external supplementary information of the non-performing assets and the risk score; third, determining the case type of the non-performing assets based on the borrower risk profile and preset features, and determining an initial case division strategy for the non-performing assets based on the case type and a preset collection strategy library; fourth, optimizing the initial case division strategy through reinforcement learning, and dividing the non-performing assets based on the optimized case division strategy.

[0005] In a second aspect, an embodiment of the present invention further provides an intelligent case classification system for non-performing assets, comprising: a pre-processing module for acquiring case information and dynamic behavior data of non-performing assets from a plurality of different data sources, and pre-processing the case information and the dynamic behavior data; wherein, the case information includes basic information of the borrower, loan information, overdue information, document information, judicial data, credit data and asset data, and the dynamic behavior data includes repayment record, overdue history and collection response; a feature construction module for constructing a case feature set of non-performing assets based on the pre-processed data, and classifying and grading the case feature set; wherein, the features in the case feature set include basic case information of non-performing assets, debtor information, asset status information and risk information. risk characteristic information; a risk assessment module, used to predict the classified and graded feature set through a pre-trained dynamic repayment ability prediction model to obtain a risk score for the non-performing assets, and to construct a risk profile of the borrower of the non-performing assets based on the external supplementary information of the non-performing assets and the risk score; wherein the external supplementary information includes at least one of judicial execution information, a credit blacklist, and social network data; a determination module, used to determine the case type of the non-performing assets based on the borrower risk profile and preset characteristics, and to determine the initial case allocation strategy for the non-performing assets based on the case type and a preset collection strategy library; an optimization case allocation module, used to optimize the initial case allocation strategy through reinforcement learning, and to allocate the non-performing assets based on the optimized case allocation strategy.

[0006] In a third aspect, an embodiment of the present invention further provides an electronic device comprising a processor and a memory, wherein the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the intelligent case division method for non-performing assets described in the first aspect above.

[0007] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the intelligent case division method for non-performing assets described in the first aspect above.

[0008] An embodiment of the present invention provides an intelligent case division method and system for non-performing assets. The method first pre-processes case information and dynamic behavior data of non-performing assets obtained from multiple different data sources, then constructs a case feature set of non-performing assets based on the pre-processed data and classifies and grades them, then predicts the classified and graded feature set through a pre-trained dynamic repayment ability prediction model to obtain a risk score for the non-performing assets, and constructs a risk profile of the borrower of the non-performing assets based on the external supplementary information and risk score of the non-performing assets. Then, the case type of the non-performing assets is determined based on the borrower risk profile and preset features, and the initial case division strategy for the non-performing assets is determined based on the case type and a preset collection strategy library. Then, the initial case division strategy is optimized through reinforcement learning, and the non-performing assets are divided based on the optimized case division strategy. By using the above technology, the predictive ability of the model can be used to integrate multi-source data to accurately assess the risks of non-performing assets, and then build a risk profile of the borrower to formulate an initial classification strategy for non-performing assets. This can reduce manual intervention in the classification of non-performing assets, improve the efficiency of non-performing asset disposal, and significantly shorten the non-performing asset disposal cycle; it can also be combined with reinforcement learning technology to optimize the initial classification strategy for non-performing assets, so that non-performing assets can be classified based on the optimized classification strategy, reducing unnecessary disposal costs of non-performing assets.

[0009] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or understood by practicing the present invention. The purposes and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0010] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0012] Figure 1 A flowchart of a method for intelligently classifying non-performing assets according to an embodiment of the present invention is shown;

[0013] Figure 2 This is a schematic diagram of the structure of an intelligent case division system for non-performing assets in an embodiment of the present invention;

[0014] Figure 3 Schematic diagram of the structure of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0016] Currently, traditional methods for classifying non-performing assets suffer from issues such as information fragmentation, low efficiency, and high labor costs. Based on this, the present invention provides an intelligent method and system for classifying non-performing assets, which can alleviate these issues in existing non-performing asset classifying technologies.

[0017] To facilitate understanding of this embodiment, firstly, a method for intelligent case division of non-performing assets disclosed in an embodiment of the present invention is described in detail. Figure 1 As shown, the method may include the following steps:

[0018] Step S102: Acquire case information and dynamic behavior data of non-performing assets from multiple different data sources, and pre-process the case information and dynamic behavior data.

[0019] Among them, case information may include borrower basic information, loan information, overdue information, document information, judicial data, credit data and asset data, etc., and dynamic behavior data may include repayment records, overdue history and collection response, etc.

[0020] Detailed information on nonperforming loan cases is collected from core systems, credit management systems, and collection systems of data sources such as banks and financial institutions. This includes basic borrower information (identity, contact information, occupation, income, etc.), loan information (loan amount, term, interest rate, repayment history, outstanding balance, etc.), overdue information (overdue time, overdue amount, collection history, etc.), and structured data from relevant contracts, agreements, and other documents. This data also integrates external data such as judicial data, credit reports, third-party credit reports, and asset data, as well as real-time data streams (such as repayment records, overdue history, and collection responses). Repayment records primarily include repayment amount, frequency, repayment method (full / partial), and repayment channel (active / withholding). Overdue history primarily includes the number of overdue payments, number of overdue days, and percentage of overdue amount. Collection responses primarily include the answer rate of collection calls, the rate of promised repayments, and the number of changes to repayment plans. External data primarily includes judicial freezes, records of defaulting debtors, and multiple loans.

[0021] Through API interfaces or data warehouses, data collected from different sources are cleaned, converted, and integrated, missing values, duplicate values, and outliers are processed, data formats (such as date, amount, address, etc.) are unified, and data is standardized (such as timestamp alignment and field mapping) to form a structured data set to ensure data quality.

[0022] Step S104: construct a case feature set of non-performing assets based on the pre-processed data, and classify and grade the case feature set.

[0023] Among them, the features in the case feature set may include basic case information, debtor information, asset status information and risk feature information of the corresponding non-performing assets.

[0024] Continuing with the previous example, feature extraction can be performed on the pre-processed data to analyze the key features of non-performing asset cases. Key features may include basic information of the borrower (such as age, occupation, income, debt ratio, etc.), loan features (loan amount, term, interest rate, etc.), historical repayment performance, overdue amount, number of days overdue, remaining principal, repayment ability, borrower's willingness to repay, collateral status, historical collection response rate, etc., so that the obtained key features can be combined into a feature set of non-performing asset cases. Specific feature extraction methods may include: extracting text features through a Transformer model (such as BERT), extracting image features through a convolutional neural network (CNN), analyzing time series data (such as repayment records, changes in financial indicators) through a recurrent neural network (RNN) to extract time series features, and constructing an asset association graph through a graph neural network (GNN) to analyze the complex relationships between enterprises, debtors, and guarantors to extract relationship features. Finally, text features, image features, time series features, and relationship features are integrated into a unified feature vector. When performing feature fusion, traditional financial indicators (such as debt-to-asset ratio, cash flow, etc.) can be combined with unstructured data (such as legal documents, news and public opinion, etc.) through multi-dimensional feature fusion to build a more comprehensive feature system. Feature weights can be dynamically adjusted according to market environment (such as economic cycles, policy changes) to improve model adaptability. Causal inference models (such as DoWhy, etc.) can be introduced to identify the causal relationship between features and asset disposal results to improve the accuracy of the fused features.

[0025] Non-performing asset cases can be classified and graded based on their case characteristics. For example, according to the overdue amount, they can be divided into small cases, medium cases, large cases, etc.; according to the overdue time, they can be divided into short-term cases, medium-term cases, long-term overdue cases, etc.; according to the complexity of the case, they can be divided into simple cases, complex cases, etc. Specifically, when classifying and grading, case characteristics should cover basic case information, debtor information, asset status information, risk characteristic information, etc., especially when classifying and grading based on the dimension of case complexity, it is necessary to focus on collecting characteristic information related to case complexity. The following is the key information that case characteristics must include when classifying and grading according to three dimensions (i.e., risk level, asset type, and case complexity):

[0026] (1) Basic case information: case number (a number that uniquely identifies the case, facilitating case management and tracking), case source (i.e., the source channel of the case, such as banks, courts, asset management companies, etc.), case type (e.g., loan default, guarantee compensation, bankruptcy liquidation, etc.), and case occurrence time (i.e., the time when the case occurred or was filed, used to assess timeliness and risk changes).

[0027] (2) Debtor information: basic information of the debtor (name, ID number / unified social credit code, contact information, etc.), financial status of the debtor (balance sheet, income and expenditure, cash flow, etc.), credit status of the debtor (credit record, historical defaults, litigation, etc.), willingness and ability of the debtor to repay (willingness and ability to repay assessed through communication, investigation, etc.).

[0028] (3) Asset status information: asset type (such as real estate, machinery and equipment, equity, debt, etc.), asset value (assessed value, market value, mortgage value, etc.), asset ownership (ownership, right of use, mortgage, etc.), asset liquidity (asset liquidity, disposability, etc.).

[0029] (4) Risk characteristic information (especially the dimension of case complexity): mainly includes the complexity of legal relationships, the complexity of asset status, the complexity of debtor status, the difficulty and cost of disposal, and external influencing factors.

[0030] Among them, the complexity of legal relationships: involving multiple parties (such as multiple creditors, debtors, and guarantors); the existence of complex legal relationships (such as chain guarantees and cross-defaults); involving cross-border legal issues or different legal jurisdictions. The complexity of asset status: diverse asset types, involving multiple asset forms; unclear asset ownership, and the existence of disputes or conflicts; difficulty in realizing assets, such as special assets (such as construction in progress and intellectual property rights). The complexity of debtor status: the debtor's business status is complex, such as involving multiple industries and multiple regions; the debtor has major matters such as bankruptcy, reorganization, and liquidation; the debtor is involved in major litigation or arbitration cases. Disposal difficulty and cost: obstacles and challenges that may be encountered during the disposal process (such as debtor resistance and asset seizure); the time, manpower, and material costs required for disposal; the possibility of successful disposal and the expected recovery rate. External influencing factors: the impact of the macroeconomic environment on case disposal (such as industry recession and policy changes); the impact of local protectionism or administrative intervention on case disposal; the pressure of social opinion or public attention on case disposal.

[0031] (5) Other relevant information: mainly including case progress, historical disposal records, and related case information.

[0032] Case Progress: The current stage of the case (e.g., case filing, trial, execution, etc.). Historical Disposition Records: The previous disposition methods and results of similar cases. Related Case Information: The relevance of the case to other cases, such as whether it is a series of cases or a serial case.

[0033] In step S106, the classified and graded feature set is predicted by a pre-trained dynamic repayment ability prediction model to obtain a risk score for the non-performing asset, and a risk profile of the borrower of the non-performing asset is constructed based on the external supplementary information of the non-performing asset and the risk score.

[0034] Among them, external supplementary information may include judicial execution information, credit blacklists, social network data, etc.

[0035] Step S108: Determine the case type of the non-performing asset based on the borrower risk profile and preset characteristics, and determine the initial case classification strategy for the non-performing asset based on the case type and the preset collection strategy library.

[0036] Step S110: Optimize the initial case division strategy through reinforcement learning, and divide the non-performing assets based on the optimized case division strategy.

[0037] An embodiment of the present invention provides an intelligent case division method for non-performing assets. First, case information and dynamic behavior data of non-performing assets obtained from multiple different data sources are pre-processed. Then, a case feature set of non-performing assets is constructed based on the pre-processed data and classified and graded. Then, the classified and graded feature set is predicted by a pre-trained dynamic repayment ability prediction model to obtain a risk score for the non-performing assets. A risk profile of the borrower of the non-performing assets is constructed based on the external supplementary information and risk score of the non-performing assets. Then, the case type of the non-performing assets is determined based on the borrower risk profile and preset features. Then, an initial case division strategy for the non-performing assets is determined based on the case type and a preset collection strategy library. Then, the initial case division strategy is optimized by reinforcement learning, and the non-performing assets are divided based on the optimized case division strategy. By using the above technology, the predictive ability of the model can be used to integrate multi-source data to accurately assess the risks of non-performing assets, and then build a risk profile of the borrower to formulate an initial classification strategy for non-performing assets. This can reduce manual intervention in the classification of non-performing assets, improve the efficiency of non-performing asset disposal, and significantly shorten the non-performing asset disposal cycle; it can also be combined with reinforcement learning technology to optimize the initial classification strategy for non-performing assets, so that non-performing assets can be classified based on the optimized classification strategy, reducing unnecessary disposal costs of non-performing assets.

[0038] As a possible implementation method, the above-mentioned feature set may include static features and dynamic features. Static features may include basic information of the borrower (such as age, occupation, income, debt ratio, etc.) and loan features (such as loan amount, term, interest rate, etc.). Dynamic features may include repayment behavior characteristics (such as repayment punctuality rate in the last three months, repayment amount volatility, etc.), overdue characteristics (such as the number of overdue payments in the last 6 months / 12 months and the average overdue days within 30 days, 30-60 days, and more than 60 days, etc.) and collection response characteristics (such as collection call answering rate, promised repayment fulfillment rate, etc.). Based on this, the above-mentioned step S106 uses the pre-trained dynamic repayment ability prediction model to predict the classified and graded feature set to obtain the risk score of the non-performing asset, which may include:

[0039] Step a1: input the feature set into the dynamic repayment ability prediction model.

[0040] A dynamic repayment ability prediction model can be constructed based on time series analysis, machine learning algorithms (such as logistic regression, random forest, XGBoost, etc.), and neural networks, and the dynamic repayment ability prediction model can be trained using historical data to optimize the parameters of the dynamic repayment ability prediction model.

[0041] The methods of constructing and training a dynamic repayment ability prediction model can mainly include: setting the goal of the dynamic repayment ability prediction model to predict the repayment ability of non-performing assets in the future (such as the probability of repayment on time, the probability of default, etc.); collecting historical data (including time series data, static features and macroeconomic data). These historical data come from the banking system, credit reporting agencies, public economic data, etc. Time series data can include historical repayment records (such as monthly repayment amount, overdue days), loan balance, interest rate changes, etc. Static features can include the borrower's age, occupation, income, and loan amount, term, collateral value, etc. Macroeconomic data can include GDP growth rate, unemployment rate, industry trends, etc.; preprocessing the collected historical data, and performing feature extraction based on the preprocessed historical data to extract static features and time series features (i.e. dynamic features). In the process of static feature extraction, Continuous variables (such as age and income) are converted into discrete intervals and interactive features reflecting repayment ability (such as income × loan amount / collateral value) are calculated. In the process of time series feature extraction, it is necessary to extract lag features (such as repayment records of the past N periods, etc.), rolling statistical features (such as the average repayment amount in the past period, the maximum number of overdue days, etc.), and trend features (such as the linear regression slope and volatility of the repayment amount, etc.); correlation analysis is performed on the extracted features to remove features with low correlation with the target variable (such as correlation below the preset correlation threshold), and random forest or XGBoost is used to calculate the importance score of each feature to remove features with low importance score (such as importance score below the preset importance score threshold); the training set (such as the 70% of data before the time sequence) and the test set (the 30% of data after the time sequence) are divided into two groups according to time sequence; the training set is used to train the model and the test set is used to test the model performance. You can use Grid Search or Bayesian optimization (such as Optuna) to tune hyperparameters, combine multiple models (such as Random Forest + XGBoost) to improve prediction stability, and adjust the weights of different features based on business needs (for example, more weight for recent repayment history). You can also deploy the model as an API service to receive real-time data and return prediction results, regularly evaluate model performance, and update the model to adapt to changes in data distribution (such as concept drift).

[0042] In the process of building and training a dynamic repayment capacity prediction model, it is necessary to ensure the integrity and consistency of time series data, the model results must be business interpretable (such as feature importance analysis), and relevant regulations (such as GDPR) must be complied with when processing sensitive data.

[0043] Through the above-mentioned dynamic repayment ability prediction model construction and training operation method, an efficient and stable dynamic repayment ability prediction model can be constructed. The dynamic repayment ability prediction model can not only capture dynamic changes through time series analysis, but also select appropriate machine learning models for calculation based on data characteristics.

[0044] In step a2, the dynamic repayment ability prediction model analyzes the static features in the feature set to obtain the initial risk level of the non-performing assets.

[0045] The initial risk level of non-performing assets can be quantitatively assessed by comprehensively analyzing the borrower's basic information and loan characteristics through a scoring card model or machine learning algorithm.

[0046] The core assessment dimensions and logic of the initial risk level can mainly include:

[0047] Based on the basic information of borrowers, risk stratification can be performed based on age to divide them into 25-40 years old (stable repayment ability), 40-55 years old (peak income period), and under 25 / over 55 years old (higher risk). For example, a 20-year-old borrower may have an increased risk of default due to unstable career or insufficient income; industry stability and position level can be defined based on occupation to define that civil servants, teachers and other professions have lower default rates than freelancers or entrepreneurs, and that executives have higher income stability than grassroots employees; risk stratification can be calculated based on income debt-to-income ratio (DTI) and income volatility to determine that the default risk increases significantly when the DTI is greater than 40%, and that sales and commission-based professions have higher risks than those with fixed salaries due to large income fluctuations; a debt ratio threshold can be set based on the debt ratio to determine that when the debt ratio is greater than 50%, the default risk of borrowers increases significantly due to possible cash flow pressure. By analyzing the debt structure, it can be determined that the risk increases with the proportion of high-interest debt (such as credit cards);

[0048] In view of the loan characteristics, the scale effect (i.e., the larger the loan amount, the more serious the default loss) and the diversification principle (i.e., micro loans can reduce overall risk through diversification) can be determined based on the loan amount; the probability of borrower default increases with the term (e.g., the probability of default of a 30-year mortgage is greater than that of a 5-year consumer loan) and the relationship between the repayment method and the default risk (equal installments of principal and interest are more likely to mask early default risks than equal installments of principal) can be defined based on the interest rate; the relationship between interest rate type and default risk (e.g., floating rate loans are greatly affected by market fluctuations and have a higher default risk than fixed rate loans) and the relationship between interest rate level and default risk (e.g., high-interest loans may force borrowers to "borrow to pay off loans") can be defined based on the interest rate.

[0049] The model for determining the initial risk level can mainly adopt a scoring card model or a machine learning model.

[0050] The variable weights for the scorecard model are: income (25%), debt ratio (20%), job stability (15%), loan amount (15%), term (10%), interest rate (10%), and age (5%). The scoring rules include: income greater than 10,000 yuan is assigned a +20 point bonus; debt ratio greater than 50% is assigned a -30 point bonus; and civil servant occupation is assigned a +15 point bonus. The initial risk classification system for the scorecard model is: a total score of 80 or more is considered low risk; a total score of 60 or more and less than 80 is considered medium risk; and a total score less than 60 is considered high risk.

[0051] Machine learning model algorithms include: logistic regression, random forest, and XGBoost. Input data for machine learning models can include historical default data and macroeconomic indicators (such as unemployment rate and GDP growth rate). Output data for machine learning models can include default probability (e.g., 0.15 indicates a 15% probability of default). The default probability output by the machine learning model can be mapped to a corresponding initial risk level: a default probability less than 0.2 is considered low risk; a default probability of at least 0.2 and less than 0.4 is considered medium risk; and a default probability of at least 0.4 is considered high risk.

[0052] In actual application, the initial risk level can be adjusted in a timely manner by dynamically monitoring relevant data and regularly updating the borrower's income, liabilities and other data.

[0053] In step a3, the dynamic repayment ability prediction model analyzes the dynamic features in the case feature set based on the initial risk level to obtain a risk score for the non-performing assets.

[0054] After determining the initial risk level of non-performing assets, the risk score of the non-performing assets can be calculated in real time by combining dynamic data such as repayment records, overdue history, and collection responses, significantly improving the timeliness and accuracy of risk assessments of non-performing assets. Methods for calculating risk scores in real time by combining dynamic data may include: converting non-numerical data of dynamic features (i.e., repayment behavior features, overdue features, and collection response features) into numerical values, and using a time decay function to weight historical data (i.e., historical data closer to the current moment is assigned a higher weight, such as historical data three months from the current moment has a higher weight than historical data six months from the current moment). Machine learning model algorithms include logistic regression, random forest (for handling non-linear relationships and suitable for complex data), XGBoost, and deep learning algorithms (such as LSTM). Among them, logistic regression is suitable for fast scoring and is suitable for small-scale data; random forest / XGBoost is suitable for handling non-linear relationships and is suitable for complex data; and LSTM is suitable for capturing time series features and is suitable for long-term behavioral analysis. The initial risk level and dynamic characteristics of non-performing assets can be input into the machine learning model to calculate the risk probability of the non-performing assets through the machine learning model and then map the risk probability into a risk score (for example, if the risk probability is calculated to be 0.8, it will be mapped to a risk score of 80 points).

[0055] When using machine learning models to calculate risk scores, tools like Kafka and Flink can be used to receive dynamic data in real time for streaming data processing (for example, immediately updating the most recent on-time payment rate after a user makes a repayment). Models can also be retrained regularly (e.g., daily / weekly) or weights can be dynamically adjusted using online learning algorithms (e.g., Vowpal Wabbit). Dynamic adjustment rules can also be introduced, for example, if the number of days overdue exceeds 60, the risk score can be directly increased to 90 points, and a risk score change threshold (e.g., ±5 points) can be set to prevent frequent fluctuations in the risk score.

[0056] As a possible implementation, constructing a risk profile of a borrower of a non-performing asset based on the external supplementary information and risk score of the non-performing asset in step S106 may include:

[0057] Step A1: Determine supplementary risk information of non-performing assets based on external supplementary information.

[0058] Exemplarily, the supplementary risk information may include at least one of the debt-paying ability information, credit risk information and implicit risk information of the non-performing assets; then the operation method of the above-mentioned step A1 (i.e., determining the supplementary risk information of the non-performing assets based on the external supplementary information) may include: if the external supplementary information includes judicial execution information, then determining the debt-paying ability information of the non-performing assets based on the judicial execution information in the external supplementary information; if the external supplementary information includes a credit blacklist, then determining the credit risk information of the non-performing assets based on the credit blacklist in the external supplementary information; if the external supplementary information includes social network data, then determining the implicit risk information of the non-performing assets based on the social network data in the external supplementary information.

[0059] Step A2: Associating the supplementary risk information and risk score of the same non-performing asset into a risk profile of the borrower of the same non-performing asset.

[0060] External data such as judicial execution information, credit blacklists, and social network data can be introduced to supplement risk assessment dimensions to significantly improve the comprehensiveness and accuracy of the assessment. Specific implementation methods for introducing external data to supplement risk assessment dimensions may include: accessing the National Court Debtor Information Inquiry System, the Judgment Documents Network, etc. to obtain judicial execution information to supplement risk dimensions. Supplementary risk dimensions include execution frequency (high-frequency execution records reflect the continued deterioration of the debtor's debt repayment ability), execution results (low success rate of asset auctions or failure to sell, indicating that the debtor's asset quality is poor), and related cases (cases involving private lending, contract disputes, etc., indicating the debtor's credit risk); accessing the central bank's credit reporting system and third-party credit reporting agencies to obtain credit blacklists to supplement risk dimensions. Supplementary risk dimensions include blacklist types (credit overdue, guarantee compensation, tax Different types of blacklists such as arrears correspond to different risk levels), time span (blacklist records in the past 1 year, 3 years or 5 years, reflecting the persistence of the default behavior), related information (whether the debtor's spouse or the legal representative of the company is on the blacklist, indicating related risks); legally obtain the debtor's public social account information (such as Weibo, WeChat Moments) as social network data to supplement the risk dimension. The supplementary risk dimensions include behavioral patterns (such as: abnormal interaction frequency, "debt", "bankruptcy" and other content keywords), relationship networks (such as the degree of correlation with high-risk groups such as dishonest enterprises), and geographical locations (such as frequent appearances in high-risk areas such as private lending gathering places).

[0061] By introducing external data to supplement risk assessment dimensions as mentioned above, external data can be used to effectively supplement the three dimensions of debt repayment capacity, credit risk and hidden risk in non-performing asset risk assessment, providing financial institutions with a more comprehensive risk profile of borrowers for non-performing assets.

[0062] As a possible implementation, the preset characteristics may include the asset value characteristics and disposal potential of the non-performing assets, the case types may include high-risk, high-value cases, low-risk cases, long-term overdue cases, etc., and the preset collection strategy library may include multiple collection strategies and matching relationships between case types and collection strategies. Based on this, determining the initial case division strategy for the non-performing assets based on the case type and the preset collection strategy library in step S108 may include: matching the case type with multiple collection strategies in the preset collection strategy library to determine a target collection strategy that matches the case type based on the matching relationship between the case type and the collection strategy; and determining the initial case division strategy for the non-performing assets based on the target collection strategy.

[0063] In order to formulate case assignment strategies based on the risk assessment results of non-performing assets (i.e., borrower risk profiles) so as to subsequently allocate non-performing assets to appropriate disposal channels or collection teams, a collection strategy library containing multiple different collection strategies (such as traditional telephone collection, judicial litigation, pre-trial preservation + judicial mediation, debt inclusion in judicial disposal, arbitration, and other collection methods) can be pre-built. The collection strategy library is configured with matching relationships between case types and collection strategies, collection strategy matching rules for the rule engine, and multiple case assignment strategy templates (such as those based on overdue stage, risk level, and disposal cost). The rule engine also sets case assignment rules based on factors such as asset risk level, borrower characteristics, and disposal channel capabilities. After determining the case type of the non-performing asset based on the dynamic repayment ability prediction results of the non-performing asset (i.e., borrower risk profile), asset value characteristics, and disposal potential, the rule engine can then match the optimal collection strategy for the non-performing asset from the collection strategy library based on the case type, thereby preliminarily determining the case assignment strategy for the non-performing asset. For example, the optimal collection strategy matched by the rule engine for high-risk, high-value cases is to assign them to a professional collection team or legal department. The optimal collection strategy matched by the rule engine for low-risk cases is to collect through low-cost traditional telephone collection channels such as SMS and IVR. The optimal collection strategy matched by the rule engine for long-term overdue cases is to transfer them to the judicial litigation process.

[0064] The definition of high-risk, high-value cases includes: borrowers with high risk, but assets or claims with high recovery value, which require professional means to maximize value. The classification of high-risk, high-value cases is based on:

[0065] (1.1) Risk profile of borrowers in high-risk, high-value cases: Borrowers with poor credit records, multiple overdue or default records; heavy debt burden, weak debt repayment ability, and may involve multiple loans or excessive debt; poor industry or operating conditions, with the risk of continued losses or bankruptcy.

[0066] (1.2) Asset Value Characteristics of High-Risk, High-Value Cases: The collateral or pledged assets are of high value, such as real estate, land, equipment, etc., and the ownership is clear; the debt itself has potential value, such as large corporate loans, high-quality project financing, etc.

[0067] (1.3) Disposal potential for high-risk, high-value cases: A high recovery rate is expected to be achieved through legal proceedings, asset restructuring, debt restructuring, etc.; more resources are required for due diligence, valuation, and disposal plan design.

[0068] For example, if a company's loan is overdue due to poor management, but the company owns many high-quality properties and the auction of the mortgaged properties can cover most of the company's debts, then the non-performing asset case is a high-risk and high-value case.

[0069] The definition of low-risk cases includes: low borrower risk, high probability of asset or debt recovery, and low disposal costs. The classification of low-risk cases is based on:

[0070] (2.1) Risk profile of borrowers in low-risk cases: The borrower has a good credit record and no major overdue or default behavior; has stable income or cash flow and strong debt repayment ability; and has a stable industry or business conditions and no major risk hazards.

[0071] (2.2) Asset value characteristics of low-risk cases: The value of the collateral or pledge is stable and has strong liquidity; the creditor's rights are clear and free of disputes or defects.

[0072] (2.3) Potential for handling low-risk cases: No complex handling methods are required, and the funds can be quickly recovered through negotiation, collection, etc.; the recovery rate is high and the handling cost is low.

[0073] For example, if a borrower's personal housing mortgage loan is overdue due to short-term cash flow problems, but the borrower has a stable income and sufficient property value, and the problem of overdue personal housing mortgage loan can be quickly resolved through negotiated repayment plan, then the non-performing asset case is a low-risk case.

[0074] The definition of a long-term overdue case includes: the borrower has not fulfilled its repayment obligations for a long time, the recovery of assets or claims is difficult, and long-term tracking and management are required. The classification of long-term overdue cases includes:

[0075] (3.1) Risk profile of borrowers in long-term overdue cases: The borrower has been out of contact for a long time or has refused to repay, and has very low willingness to repay; may be involved in malicious debt evasion, fraud, etc.; the debt burden is heavy and no longer able to repay.

[0076] (3.2) Asset value characteristics of long-term overdue cases: the value of the collateral or pledge may depreciate and become difficult to realize; the creditor's rights may have defects, such as exceeding the statute of limitations or ownership disputes.

[0077] (3.3) Disposal potential of long-term overdue cases: recovery requires legal action, compulsory execution and other means, with a long recovery period; the recovery rate is low and the disposal cost is high.

[0078] For example, if a corporate loan is overdue for more than five years, the company has gone bankrupt and liquidated, and the value of the collateral has shrunk significantly due to falling market conditions, and legal procedures are required to recover the remaining debt, then the non-performing asset case is considered a long-term overdue case.

[0079] By classifying non-performing asset case types according to the above definitions and classification criteria (i.e., high-risk and high-value cases, low-risk cases, and long-term overdue cases), it is convenient to formulate differentiated collection strategies for different case types and then divide the cases, which is conducive to optimizing the allocation of non-performing asset disposal resources and improving the overall recovery efficiency of non-performing assets.

[0080] As a possible implementation, before optimizing the initial case assignment strategy via reinforcement learning in step S110, the non-performing asset assignment action may be defined as a Markov decision process to construct a non-performing asset case assessment model. Based on this, optimizing the initial case assignment strategy via reinforcement learning in step S110 may include: inputting the initial case assignment strategy into the case assessment model to iteratively optimize the case assignment strategy via the case assessment model; wherein, each time the case assignment strategy is optimized, the case assessment model calculates and outputs a priority score for the current case assignment strategy to update the current case assignment strategy based on the priority score, with the current case assignment strategy serving as the initial case assignment strategy for the first optimization of the case assignment strategy.

[0081] Based on factors such as the urgency of the case, risk level, and disposal cost, combined with characteristics such as the borrower's region, occupation, and debt type, dynamic case division rules can be formulated to prioritize tasks and ensure that special cases are handled first. For example, high-risk customers are assigned to experienced collection teams first, while low-risk customers are subject to automated collection processes (such as text messages, AI voice, etc.). Based on real-time data feedback (such as collection results and changes in the market environment), the case division strategy is dynamically adjusted. Specifically, the case division action of non-performing assets can be defined as a Markov Decision Process (MDP) to iteratively optimize the case division strategy initially formulated for non-performing assets (i.e., the initial case division strategy) using reinforcement learning.

[0082] In order to optimize the initial case-splitting strategy using reinforcement learning, it is necessary to define the state-action value function Q(s,a) that satisfies Q(s,a) new =(1-α)Q(s,a) old +α[r+γmax a' Q(s',a')],Q(s,a) newand Q(s,a) old Represent the updated state-action value function value and the state-action value function value before the update, s and s' represent the state (i.e., case characteristics + disposal stage), a and a' represent the action (i.e., allocation of traditional electronic collection / judicial litigation / pre-litigation preservation + judicial mediation / debt joining judicial disposal / arbitration), α is the learning rate, γ is the decay value, and r is the reward (i.e., the ratio of the recovery amount to the recovery time, used to characterize the collection effect of the case division strategy). The process of optimizing the initial case division strategy using the reinforcement learning algorithm can include:

[0083] 1) Initialize the Q value table.

[0084] Specifically, a Q-value table can be randomly initialized for all possible states (s) and corresponding actions (a).

[0085] 2) Enter round-by-round iterative optimization.

[0086] Specifically, the following optimization rounds may be repeated several times:

[0087] At the beginning of each optimization round, reset the environment to its initial state(s);

[0088] Repeat the following operations in each step of each round of optimization: select an action (a) based on the current Q-value table, for example, using the ε-greedy strategy (random exploration with a certain probability, otherwise select the action with the largest current Q-value); execute action (a), immediately obtain the immediate reward (r) from the environment feedback, and transfer to the next state (s'); according to the formula Q(s,a) new =(1-α)Q(s,a) old +α[r+γmax a' Q(s',a')] updates the Q value of the current state-action pair; updates the current state (s) to the next state (s');

[0089] This round of optimization ends when the current state is the terminal state.

[0090] Before optimizing the initial case assignment strategy, it is necessary to build a case assessment model and design dynamic case assignment rules. The specific steps are as follows:

[0091] A) The evaluation dimensions of the case evaluation model include the case characteristic dimension and the borrower characteristic dimension.

[0092] Case characteristics include urgency, risk level, and handling cost. Urgency: A priority level determined by the number of days overdue and the promised repayment period (e.g., overdue within 30 days is low urgency, overdue over 60 days is high urgency). Risk level: A high / medium / low risk level based on the borrower's credit score, past default record, and repayment ability. Handling cost: The economic value of handling a case is determined by estimating manual collection costs, legal fees, and technological investment.

[0093] Borrower characteristic dimensions include region, occupation, and debt type. Region: Priority is set based on differences in regional economic level, judicial environment, etc. (for example, first-tier cities have abundant legal resources, so a higher priority can be set for non-performing asset cases in first-tier cities so that they can be handled first). Occupation: Priority is set based on the positive correlation between occupational stability and repayment ability (for example, borrowers with higher occupational stability, such as civil servants and state-owned enterprise employees, have stronger repayment ability, so a higher priority can be set for non-performing asset cases of borrowers with higher occupational stability so that they can be handled first). Debt type: The priority of non-performing asset cases for mortgage debts such as home loans and car loans is set higher than the priority of non-performing asset cases for unsecured debts such as credit cards and consumer loans.

[0094] B) Configure dynamic case division rules.

[0095] Assign corresponding weights to the priority of each dimension in A). For example, set the weights for urgency, risk level, resolution cost, and borrower characteristics to 40%, 30%, 20%, and 10%, respectively. Define the priority score formula as follows: Priority score = urgency priority × urgency weight + risk level priority × risk level weight + resolution cost priority × resolution cost weight + borrower characteristic priority × borrower characteristic weight.

[0096] Based on priority scores, cases are divided into high-priority cases (priority scores of 80 or more), medium-priority cases (priority scores greater than 50 and less than 80), and low-priority cases (priority scores of 50 or more). The rules engine's dynamic case assignment rules include: assigning high-priority cases to experienced collection teams; assigning medium-priority cases to regular collection teams; and utilizing automated collection processes (such as SMS and AI voice) for low-priority cases.

[0097] The dynamic case distribution rules configured in the rule engine for high-risk customers, low-risk customers, and regional differences include: for high-risk customers, they are given priority assignment to senior debt collectors to develop personalized repayment plans; for low-risk customers, automated collection is the main method, supplemented by manual intervention; and to address regional differences, mediation or negotiation is given priority in areas with scarce judicial resources.

[0098] In actual application, the priority scoring formula needs to be calibrated regularly to avoid weight imbalance, and differentiated case assignment strategies need to take into account efficiency and compliance (such as avoiding excessive collection) as well as maximizing collection rates, minimizing costs and improving customer experience.

[0099] As a possible implementation, the classification of non-performing assets based on the optimized classification strategy in step S110 may include:

[0100] Step B1: Generate a case list for non-performing assets based on the optimized case division strategy.

[0101] The case assignment list may include basic case information, collection resource allocation information, and case assignment time. Basic case information may include the case number, debtor's name, debt amount, and overdue duration. Collection resource allocation information includes the name and contact information of the assigned collection team or individual. Basic case information may also include case-related explanatory information (such as collection methods).

[0102] Step B2: Generate collection task instructions based on the case list.

[0103] The collection task instructions may include basic case information, collection targets, collection strategy recommendations, and collection task information; the collection task information may include the collection task deadline, and the matching relationship between the collection task progress and the reward.

[0104] Step B3: Send the collection task instruction to the target collection team and / or target collection individual corresponding to the collection resource allocation information, so that the collection task is performed by the target collection team and / or target collection individual.

[0105] In order to automatically assign non-performing assets to designated collection teams or individuals based on the optimized case assignment strategy, it is necessary to generate a case assignment list and task instructions.

[0106] Specifically, the real-time data of collection resources can be integrated first. The real-time data of collection resources can include: historical performance data of the collection team or individual collectors (such as collection rate, case closing rate), professional ability labels (such as legal litigation experience, collection experience in specific industries), current task saturation, geographic location and other information.

[0107] According to the business needs of financial institutions, the collection resource matching rules of the optimized case distribution strategy include asset attribute matching rules, collection ability matching rules, geographic location matching rules, task saturation balance rules, etc. Asset attribute matching rules: Assign debts to corresponding collection teams according to debt amount range, overdue stage, and asset type (personal loan / corporate loan). Collection ability matching rules: Assign high-risk cases to teams with legal litigation experience, and low-risk cases to regular collection teams. Geographic location matching rules: Assign debtors to local collection teams based on their location to improve communication efficiency. Task saturation balance rules: Dynamically monitor the task volume of each team to avoid oversaturation of tasks for a single collection resource.

[0108] Corresponding weights can also be assigned to the overdue duration, professional ability label and task saturation respectively (for example, weights of 30%, 40% and 30% are assigned to the overdue duration, professional ability of the collection team and task saturation respectively), so that the overdue duration, professional ability label and task saturation and the weights assigned to each of the three can be used to perform weighted calculations to obtain a comprehensive score for each collection team and / or each collection individual included in the collection resources, thereby facilitating the subsequent sorting of collection resources according to the comprehensive score and determining the target collection team and / or target collection individual based on the sorting results.

[0109] A visual interface is also provided for manual review and adjustment of case assignment strategies, allowing relevant personnel to manually review and adjust case assignment strategies based on actual business needs. For example, manual review and adjustment of case assignment strategies can include: special case marking (i.e., marking cases involving significant legal risks as "requiring legal review, temporarily suspended"), team adjustment (administrators manually adjust case assignment results, such as assigning a corporate loan case to a team with industry experience), etc.

[0110] After configuring the relevant rules for allocating collection resources, you can use the configured rules to generate a case list for non-performing assets based on the optimized case allocation strategy. Based on the case list, you can generate collection task instructions, which are then issued. Collection task instructions can be sent to target collection teams and / or individuals via system notifications, text messages, emails, and other methods. These instructions contain basic case information and collection targets, collection strategy recommendations (such as prioritizing telephone communication, requiring on-site asset verification, etc.), collection task deadlines, and performance-linked rules.

[0111] After the collection task instructions are issued, the non-performing asset case has been assigned to the target collection team and / or target collection individual. At this time, the case status of the non-performing asset case can be synchronously updated to the allocated collection resources.

[0112] In actual application, the collection progress, collection rate and other indicators of the collection team and / or individual collectors can be monitored in real time to evaluate the effectiveness of the optimized case division strategy, and the case division list of non-performing assets can be dynamically adjusted based on the effectiveness evaluation results of the optimized case division strategy.

[0113] The system also allows real-time monitoring of the collector's progress during the collection process. Monitoring metrics can include the number of calls, call duration, repayment commitments received, repayment amounts, and other performance indicators. Visual reports are generated to help relevant personnel identify problems in the collection process and take appropriate measures.

[0114] Furthermore, a multi-court collaborative scheduling interface can be set up to connect to different courts' case filing systems in real time. This allows for monitoring statutes of limitations, automatically calculating critical statutes of limitations for each case, generating early warnings, and dynamically adjusting case jurisdiction allocations. The time, method, results, and communication content of each collection action can also be recorded, forming a complete collection log to facilitate subsequent audits and adjustments to collection strategies.

[0115] By adopting the above-mentioned intelligent classification method for non-performing assets, efficient classification of non-performing assets can be achieved through steps such as multi-source data integration, risk assessment, intelligent analysis, classification strategy formulation, classification strategy optimization and execution. The use of artificial intelligence technology can improve the accuracy and efficiency of non-performing asset classification, and effectively solve the pain points of traditional non-performing asset classification methods such as information dispersion, low efficiency, reliance on experience, and lack of dynamic optimization, providing financial institutions with an intelligent, standardized and efficient non-performing asset classification method.

[0116] The beneficial effects of the above-mentioned intelligent non-performing asset classification method mainly include: compared with the traditional manual classification method of non-performing assets, the automation of multi-source data collection, model analysis, classification strategy formulation and other links can improve the efficiency of non-performing asset classification, reduce manual intervention in the non-performing asset classification process, improve the efficiency of non-performing asset disposal, and significantly shorten the non-performing asset disposal cycle; by optimizing the classification strategy, unnecessary disposal costs of non-performing assets can be reduced; through accurate risk assessment and targeted classification strategies, flexible disposal of non-performing assets can be achieved and the recovery rate of non-performing assets can be improved; by real-time monitoring of the progress of non-performing asset disposal and timely adjustment of the collection resource allocation strategy, the non-performing asset disposal effect can be optimized.

[0117] Based on the above-mentioned non-performing asset intelligent case division method, the embodiment of the present invention also provides a non-performing asset intelligent case division system, see Figure 2 As shown, the system may include the following modules:

[0118] The preprocessing module 202 is used to obtain case information and dynamic behavior data of non-performing assets from multiple different data sources, and preprocess the case information and the dynamic behavior data; wherein the case information includes basic information of the borrower, loan information, overdue information, document information, judicial data, credit data and asset data, and the dynamic behavior data includes repayment record, overdue history and collection response.

[0119] The feature construction module 204 is used to construct a case feature set of non-performing assets based on the pre-processed data, and classify and grade the case feature set; wherein the features in the case feature set include basic case information of the non-performing assets, debtor information, asset status information and risk feature information.

[0120] The risk assessment module 206 is used to predict the classified and graded feature set through a pre-trained dynamic repayment ability prediction model to obtain a risk score for the non-performing assets, and to construct a risk profile of the borrower of the non-performing assets based on the external supplementary information of the non-performing assets and the risk score; wherein the external supplementary information includes at least one of judicial execution information, credit blacklist, and social network data.

[0121] Determination module 208 is used to determine the case type of non-performing assets based on the borrower risk profile and preset characteristics, and to determine the initial case classification strategy for non-performing assets based on the case type and the preset collection strategy library; wherein the case types include high-risk and high-value cases, low-risk cases and long-term overdue cases.

[0122] The case allocation optimization module 210 is used to optimize the initial case allocation strategy through reinforcement learning, and allocate non-performing assets based on the optimized case allocation strategy.

[0123] By adopting the above-mentioned intelligent non-performing asset classification system, the model's predictive ability can be used to integrate multi-source data to accurately assess the risks of non-performing assets, and then build a risk profile of the borrower to formulate an initial classification strategy for non-performing assets. This can reduce manual intervention in the non-performing asset classification process, improve the efficiency of non-performing asset disposal, and significantly shorten the non-performing asset disposal cycle; it can also combine reinforcement learning technology to optimize the initial classification strategy for non-performing assets, so that non-performing assets can be classified based on the optimized classification strategy, reducing unnecessary disposal costs of non-performing assets.

[0124] The implementation principle and technical effects of the intelligent case division system for non-performing assets provided in the embodiment of the present invention are the same as those of the aforementioned embodiment of the intelligent case division method for non-performing assets. For the sake of brief description, for matters not mentioned in the embodiment of the intelligent case division system for non-performing assets, reference may be made to the corresponding contents in the aforementioned embodiment of the intelligent case division method for non-performing assets.

[0125] The embodiment of the present invention further provides an electronic device, such as Figure 3 As shown, it is a structural diagram of the electronic device, wherein the electronic device includes a processor 31 and a memory 30, the memory 30 stores computer executable instructions that can be executed by the processor 31, and the processor 31 executes the computer executable instructions to implement the above-mentioned non-performing asset intelligent case classification method.

[0126] exist Figure 3 In the illustrated embodiment, the electronic device further includes a bus 32 and a communication interface 33 , wherein the processor 31 , the communication interface 33 and the memory 30 are connected via the bus 32 .

[0127] Among them, the memory 30 may include a high-speed random access memory (RAM), and may also include a non-volatile memory (non-volatile memory), such as at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 33 (which can be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. can be used. The bus 32 can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus 32 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 3 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0128] The processor 31 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by hardware integrated logic circuits in the processor 31 or by software instructions. The processor 31 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of the present invention may be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor 31 reads the information in the memory and completes the steps of the intelligent classification method for non-performing assets in the above embodiment in combination with its hardware.

[0129] An embodiment of the present invention also provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the above-mentioned intelligent case division method for non-performing assets. The specific implementation can be found in the above-mentioned method embodiment, which will not be repeated here.

[0130] The computer program product of the method, device and electronic device for intelligent case division of non-performing assets provided in the embodiments of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the previous method embodiments. For specific implementation, please refer to the method embodiments and will not be repeated here.

[0131] Unless otherwise specifically stated, the relative steps, numerical expressions and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present invention.

[0132] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0133] In the description of the present invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate and simplify the description of the present invention. They are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation, and therefore should not be construed as limitations on the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0134] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A method for intelligent classification of non-performing assets, characterized by: include: Obtaining case information and dynamic behavior data on non-performing assets from multiple different data sources and pre-processing the case information and dynamic behavior data; wherein the case information includes basic borrower information, loan information, overdue information, documents, judicial data, credit data, and asset data; and the dynamic behavior data includes repayment records, overdue history, and collection responses; Constructing a case feature set of non-performing assets based on the pre-processed data, and classifying and grading the case feature set; wherein the features in the case feature set include basic case information, debtor information, asset status information, and risk feature information of the corresponding non-performing assets; Using a pre-trained dynamic repayment ability prediction model, the classified and graded feature set is predicted to obtain a risk score for the non-performing asset, and a risk profile of the borrower of the non-performing asset is constructed based on external supplementary information of the non-performing asset and the risk score; wherein the external supplementary information includes at least one of judicial execution information, credit blacklist, and social network data; Determine the case type of non-performing assets based on the borrower risk profile and preset characteristics, and determine the initial case classification strategy for non-performing assets based on the case type and the preset collection strategy library; The initial case division strategy is optimized through reinforcement learning, and the non-performing assets are divided based on the optimized case division strategy.

2. The intelligent classification method for non-performing assets according to claim 1, characterized in that: The feature set includes static features and dynamic features. The static features include basic information of the borrower and loan features, and the dynamic features include repayment behavior features, overdue features, and collection response features. The pre-trained dynamic repayment ability prediction model is used to predict the classified and graded feature sets to obtain the risk score of non-performing assets, including: Inputting the feature set into the dynamic repayment ability prediction model; The dynamic repayment ability prediction model analyzes the static features in the feature set to obtain an initial risk level of the non-performing assets; The dynamic repayment ability prediction model analyzes the dynamic features in the case feature set based on the initial risk level to obtain a risk score for the non-performing assets.

3. The intelligent classification method for non-performing assets according to claim 1, characterized in that: Constructing a risk profile of the borrower of the non-performing asset based on the external supplementary information of the non-performing asset and the risk score, including: Determining supplementary risk information of non-performing assets based on the external supplementary information; The supplementary risk information and risk score of the same non-performing asset are associated to form the risk profile of the borrower of the same non-performing asset.

4. The intelligent classification method for non-performing assets according to claim 3, characterized in that: The supplementary risk information includes at least one of the debt repayment ability information, credit risk information, and implicit risk information of the non-performing assets; the supplementary risk information of the non-performing assets is determined based on the external supplementary information, including: If the external supplementary information includes judicial execution information, determining the debt repayment ability information of the non-performing assets based on the judicial execution information in the external supplementary information; If the external supplementary information includes a credit blacklist, determining the credit risk information of the non-performing assets based on the credit blacklist in the external supplementary information; If the external supplementary information includes social network data, implicit risk information of the non-performing assets is determined based on the social network data in the external supplementary information.

5. The intelligent classification method for non-performing assets according to claim 1, characterized in that: The preset characteristics include the asset value characteristics and disposal potential of non-performing assets, the case types include high-risk and high-value cases, low-risk cases, and long-term overdue cases, and the preset collection strategy library includes multiple collection strategies and matching relationships between case types and collection strategies; Determine the initial classification strategy for non-performing assets based on the case type and the preset collection strategy library, including: Matching the case type with a plurality of collection strategies in the preset collection strategy library to determine a target collection strategy that matches the case type based on a matching relationship between the case type and the collection strategy; Determine the initial case allocation strategy for non-performing assets based on the target collection strategy.

6. The intelligent classification method for non-performing assets according to claim 1, characterized in that: NPLs are divided based on the optimized division strategy, including: Generate a case list for non-performing assets based on the optimized case allocation strategy; wherein the case list includes basic case information, collection resource allocation information, and case allocation time; the basic case information includes the case number, debtor name, debt amount, and overdue duration; the collection resource allocation information includes the name and contact information of the assigned collection team or individual; Generating a collection task instruction based on the case list; wherein the collection task instruction includes basic case information, collection goals, collection strategy recommendations, and collection task information; the collection task information includes a collection task deadline and a matching relationship between collection task progress and remuneration; The collection task instruction is sent to the target collection team and / or target collection individual corresponding to the collection resource allocation information, so that the collection task is performed by the target collection team and / or target collection individual.

7. The intelligent classification method for non-performing assets according to claim 1, characterized in that: Before optimizing the initial case allocation strategy by using reinforcement learning, the method further includes: defining the case allocation action of non-performing assets as a Markov decision process to construct a case evaluation model for non-performing assets; The initial case division strategy is optimized by reinforcement learning, including: The initial case assignment strategy is input into the case evaluation model to iteratively optimize the case assignment strategy through the case evaluation model; wherein, the case evaluation model calculates and outputs the priority score of the current case assignment strategy each time the case assignment strategy is optimized to update the current case assignment strategy based on the priority score, and the current case assignment strategy is the initial case assignment strategy for the first optimization of the case assignment strategy.

8. An intelligent case classification system for non-performing assets, characterized by: include: a preprocessing module for acquiring case information and dynamic behavior data of non-performing assets from multiple different data sources and preprocessing the case information and the dynamic behavior data; wherein the case information includes basic information of the borrower, loan information, overdue information, documents, judicial data, credit data, and asset data; and the dynamic behavior data includes repayment records, overdue history, and collection responses; A feature construction module is used to construct a case feature set of non-performing assets based on the pre-processed data, and to classify and grade the case feature set; wherein the features in the case feature set include basic case information of the non-performing assets, debtor information, asset status information, and risk feature information; A risk assessment module is configured to predict the classified and graded feature set using a pre-trained dynamic repayment capacity prediction model to obtain a risk score for the non-performing asset, and to construct a risk profile of the non-performing asset borrower based on external supplementary information about the non-performing asset and the risk score; wherein the external supplementary information includes at least one of judicial execution information, credit blacklists, and social network data; a determination module, configured to determine the case type of non-performing assets based on the borrower risk profile and preset characteristics, and to determine an initial case classification strategy for the non-performing assets based on the case type and a preset collection strategy library; The case allocation optimization module is used to optimize the initial case allocation strategy through reinforcement learning, and allocate non-performing assets based on the optimized case allocation strategy.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the method for intelligent case division of non-performing assets as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are called and executed by the processor, the computer-executable instructions prompt the processor to implement the intelligent case division method for non-performing assets as described in any one of claims 1 to 7.

Images