Intelligent processing method for unhealthy assets based on value evaluation and related equipment

Abstract

The invention relates to the technical field of unhealthy asset processing, and provides an intelligent unhealthy asset processing method and related equipment based on value evaluation, and the method comprises the steps: obtaining the basic information and multi-modal dynamic data of unhealthy assets; based on the debtor information and the creditor's right information of the unhealthy assets, obtaining unhealthy asset association information; obtaining a non-performing asset value evaluation value and a disposal risk coefficient based on debtor information, creditor's right information, multi-modal dynamic data, non-performing asset association information and a non-performing asset amount through the trained non-performing asset value evaluation model; obtaining a non-performing asset processing strategy set based on the non-performing asset value evaluation value and the processing risk coefficient; and processing the unhealthy assets based on the unhealthy asset processing strategy set. According to the invention, the accuracy, objectivity and reliability of the processing of the unhealthy assets are effectively improved, the influence of personal subjective tendency is reduced, and the intelligent processing of the unhealthy assets is realized.

Description

Technical Field

[0001] This application relates to the field of non-performing asset processing technology, and in particular to an intelligent processing method and related equipment for non-performing assets based on value assessment. Background Technology

[0002] Non-performing assets mainly refer to claims that are difficult to recover principal and interest on time and in full, as well as equity or physical assets whose value continues to depreciate and are difficult to generate income.

[0003] Currently, the assessment and handling of non-performing assets mainly rely on single-dimensional data and depend on the experience and professional knowledge of the personnel involved. This leads to significant subjective biases and judgment errors, resulting in low accuracy in the valuation of non-performing assets and consequently affecting the intelligent handling of these assets. Summary of the Invention

[0004] To address the above technical issues, this application proposes an intelligent processing method and related equipment for non-performing assets based on value assessment.

[0005] The first aspect of this application provides an intelligent processing method for non-performing assets based on value assessment, the method comprising: Obtain basic information and multimodal dynamic data on non-performing assets; Based on the debtor and creditor information of non-performing assets, obtain related information about non-performing assets; The non-performing asset valuation model, after training, yields the valuation value and disposal risk coefficient of non-performing assets based on the debtor information, the creditor information, the multimodal dynamic data, the non-performing asset association information, and the amount of non-performing assets. Based on the assessed value of the non-performing assets and the disposal risk coefficient, a set of non-performing asset disposal strategies is obtained; The non-performing assets are processed based on the aforementioned set of non-performing asset disposal strategies.

[0006] In an optional embodiment, the processing of non-performing assets based on the set of non-performing asset processing strategies includes: Based on expert experience, the set of non-performing asset disposal strategies is screened to obtain candidate non-performing asset disposal strategies; The candidate non-performing asset disposal strategies are decomposed into tasks and structured disposal solutions are generated. The non-performing assets are processed based on the aforementioned structured disposal scheme.

[0007] In an optional embodiment, obtaining non-performing asset-related information based on debtor and creditor information includes: Based on the debtor information, a correlation information graph is constructed, and the business status information and potential correlation risk information are obtained by searching the correlation information graph links. Based on the aforementioned debt information, target transaction case information and industry disposal benchmark parameters are matched from the market transaction case database; Non-performing asset association information is obtained based on the aforementioned operating status information, potential associated risk information, target transaction case information, and industry disposal benchmark parameters.

[0008] In an optional embodiment, the non-performing asset valuation model includes a first sub-model and a second sub-model. The non-performing asset valuation model, after training, derives the non-performing asset valuation value and disposal risk coefficient based on the debtor information, the creditor information, the multimodal dynamic data, the non-performing asset correlation information, and the non-performing asset amount, including: Time-series data and static data are obtained from the multimodal dynamic data and the non-performing asset association information; The time series data is input into the first sub-model, and the time series feature evaluation vector is output through the first sub-model. The debtor information, the creditor information, the static data, and the amount of non-performing assets are input into the second sub-model, and the non-linear static feature evaluation vector is output through the second sub-model. Obtain the first target weight of the first sub-model, and obtain the second target weight of the second sub-model; A fusion evaluation vector is obtained based on the first target weight and the temporal feature evaluation vector, the second target weight and the nonlinear static feature evaluation vector; The fusion evaluation vector is input into the fully connected layer, and the fully connected layer outputs the non-performing asset value assessment value and disposal risk coefficient.

[0009] In an optional embodiment, obtaining the first target weight of the first sub-model includes: The first initial weight is determined based on the asset type of the non-performing assets; A timeliness score is calculated based on the acquisition time and update frequency of the multimodal dynamic data. The first correction factor is determined based on the timeliness score; The first initial weight is corrected using the first correction factor to obtain the first target weight.

[0010] In an optional embodiment, obtaining the second target weights of the second sub-model includes: The second initial weight is determined based on the asset type of the non-performing assets; A completeness score is calculated based on the number of missing items and key items in the debtor information and the claim information. A second correction factor is determined based on the integrity score; The second initial weight is corrected using the second correction factor to obtain the second target weight.

[0011] In an optional embodiment, obtaining the fused evaluation vector based on the first target weight and the temporal feature evaluation vector, the second target weight and the nonlinear static feature evaluation vector includes: Determine whether the sum of the first target weight and the second target weight is 1; When the sum is not 1, the weight deviation is calculated based on the first target weight and the second target weight; The feature contribution ratio is determined based on the time-series feature evaluation vector and the nonlinear static feature evaluation vector. The weight balance factor is obtained based on the degree of weight deviation and the proportion of feature contribution. The first target weight and the second target weight are corrected based on the weight balancing factor to obtain a first corrected weight and a second corrected weight, and the sum of the first corrected weight and the second corrected weight is 1. The fusion evaluation vector is obtained based on the first modified weight and the time-series feature evaluation vector, the second modified weight and the nonlinear static feature evaluation vector.

[0012] In an optional embodiment, the method further includes: After the disposal of non-performing assets is completed, a disposal effectiveness evaluation report is generated based on the actual recovered value of the non-performing assets and the assessed value of the non-performing assets.

[0013] A second aspect of this application provides an electronic device comprising 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 an intelligent processing method for non-performing assets based on value assessment.

[0014] A third aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of a method for intelligent processing of non-performing assets based on value assessment.

[0015] Beneficial effects: This application provides a method and related equipment for intelligent processing of non-performing assets based on value assessment, which effectively improves the accuracy, objectivity, and reliability of non-performing asset processing and realizes intelligent processing of non-performing assets. Specifically, it obtains related information of non-performing assets based on debtor and creditor information, which not only improves the comprehensiveness of information and serves as the information basis for making processing strategies, but also enhances the relevance of information, effectively reducing information noise and thus improving the accuracy of processing strategies. Through a trained non-performing asset value assessment model, based on relatively comprehensive information, it obtains the value assessment value and disposal risk coefficient of non-performing assets. By intuitively representing the value and disposal risk of non-performing assets in a digital form, it fully considers the existing value of non-performing assets and the potential risks during disposal, largely eliminating the influence of subjective bias, improving the objectivity and reliability of non-performing asset processing strategies, and automatically generating a set of non-performing asset processing strategies, thereby realizing intelligent processing of non-performing assets. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating the intelligent processing method for non-performing assets based on value assessment provided in this application embodiment; Figure 2 A schematic diagram of the login interface of the non-performing asset valuation management platform provided in this application embodiment; Figure 3 This is a schematic diagram of the interface for basic information on non-performing assets provided in an embodiment of this application; Figure 4 A schematic diagram of the data analysis interface provided for embodiments of this application; Figure 5 A functional block diagram of the intelligent non-performing asset processing device based on value assessment provided in the embodiments of this application; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] Various embodiments of this disclosure will be described more fully below. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0019] Reference Figure 1 As shown in the embodiments of this application, a method for intelligent processing of non-performing assets based on value assessment is proposed. The method for intelligent processing of non-performing assets based on value assessment includes: S100 obtains basic information on non-performing assets and multimodal dynamic data.

[0020] Users access the non-performing asset valuation and management platform via a browser using the platform's default domain name / IP address, which automatically displays the platform's login interface. Figure 2 As shown; after the user enters their account and password and completes secondary verification (SMS / key), the system verifies the permissions and enters the platform's interactive interface, as shown. Figure 3 As shown in the image. On the platform's user interface, users can input basic information about non-performing assets.

[0021] Basic information on non-performing assets may include: unique asset identifier, asset category and type, asset value, associated asset number, asset name, current asset status, date the asset information was obtained, and supplementary information.

[0022] After obtaining the basic information of non-performing assets input by the user, the intelligent non-performing asset processing platform can connect to databases such as Oracle and MySQL through the configured JDBC / ODBC driver. It uses the basic information of non-performing assets as the retrieval primary key to associate multi-source data tables and extract creditor information and debtor information. It also obtains on-site image data of assets, public opinion data of debtors, real-time data of macroeconomic prosperity and industry policy changes, and integrates them to form multimodal data of non-performing assets.

[0023] S200 obtains related information about non-performing assets based on debtor and creditor information.

[0024] Based on the creditor and debtor information of non-performing assets, relevant information is obtained through big data correlation retrieval to improve the source of assessment data.

[0025] By acquiring debtor and creditor information for non-performing assets (NPAs), and thus related information, a more comprehensive analysis of the current status of NPAs can be conducted. This allows for a deeper understanding of the causes, liquidity, and potential risks of NPAs, providing ample information support for decision-making and facilitating risk control and management. The acquisition of NPA-related information is guided by creditor and debtor information, improving the correlation and effectiveness between NPA-related information and specific NPAs. Furthermore, the integration of multimodal dynamic data reduces data lag, thereby enhancing the scientific rigor, objectivity, and accuracy of decision-making.

[0026] S300, through a trained non-performing asset valuation model, obtains the valuation value of non-performing assets and the disposal risk coefficient based on debtor information, creditor information, multimodal dynamic data, non-performing asset correlation information, and non-performing asset amount.

[0027] In some embodiments of this application, the non-performing asset valuation model may adopt a dual-model fusion architecture that combines an improved LSTM with a gradient boosting tree.

[0028] The training of the non-performing asset valuation model includes: obtaining a sample set comprising multiple seven-tuples. Each seven-tuple includes: debtor information, creditor information, multimodal dynamic data, related information, amount, assessed value, and disposal risk coefficient. The assessed value can be determined by experts based on historical transaction records and actual judicial auction data, while the disposal risk coefficient can be labeled by senior appraisers and other experts based on the creditor information, debtor information, multimodal dynamic data, related information, and amount. The sample set is then used to train the dual-model fusion architecture to obtain the non-performing asset valuation model.

[0029] S400 is a set of non-performing asset disposal strategies based on the assessed value of non-performing assets and the disposal risk coefficient.

[0030] Based on preset processing objectives, processing paths and specific solutions are generated. These preset objectives include one or more of the following: loss mitigation and stop-loss, balanced recovery targets and failures, maximizing profits, and rapid capital recovery. Processing paths may include one or more of the following: collection, litigation, debt restructuring, holding and operating, and auction / transfer.

[0031] The system supplements the processing path with details to obtain processing solutions, and predicts the processing cycle, processing costs, and difficulty list under different processing paths or solutions. Based on the difficulties listed in the difficulty list, corresponding contingency plans are generated. The predicted revenue is simulated based on the asset value assessment, processing cycle, and processing costs. The risk probability is simulated by combining the disposal risk coefficient, processing cycle, processing costs, and the number of difficulties in the difficulty list.

[0032] Understandably, different processing paths or solutions vary in terms of costs, difficulties, and contingency plans. For example, excluding mediation and negotiation, litigation typically takes more than two years from filing to enforcement. Similarly, auction transfer presents difficulties for assets with unclear ownership, and corresponding contingency plans could include judicial confirmation of ownership.

[0033] S500 is a set of non-performing asset disposal strategies for processing non-performing assets.

[0034] The set of non-performing asset disposal strategies includes one or more non-performing asset disposal strategies. Each disposal strategy includes one or more of the following: disposal path, disposal plan corresponding to the disposal path, disposal cycle, disposal cost, predicted return, risk probability, list of difficulties, and contingency plans for each difficulty.

[0035] This application obtains related information about non-performing assets based on the creditor and debtor information in the basic information of non-performing assets, realizing targeted expansion of non-performing asset information, improving the comprehensiveness of decision-making information, and providing sufficient data support for decision-making; through the trained non-performing asset valuation model, the value and risk of specific non-performing assets are measured in a data-driven form, which is an intuitive reflection of the value attributes of non-performing assets and reduces the influence of subjective factors; and based on the valuation value and disposal risk coefficient of non-performing assets, a set of non-performing asset disposal strategies is automatically generated, realizing intelligent decision-making and disposal of non-performing assets.

[0036] In an optional embodiment, processing non-performing assets based on a set of non-performing asset disposal strategies includes: Based on expert experience, a set of non-performing asset disposal strategies was screened to obtain candidate non-performing asset disposal strategies; The candidate non-performing asset disposal strategies are broken down into tasks and structured disposal plans are generated. Non-performing assets are handled based on structured disposal solutions.

[0037] The non-performing asset valuation and management platform has a pre-set expert experience database, which can be a structured experience database storing the strategy selection experience of experts in the field of non-performing asset disposal. This strategy selection experience specifically includes a triplet of experience data: asset characteristics - disposal scenario - optimal strategy. Furthermore, the expert experience database can be configured with experience matching and weight adjustment rules. When preset conditions are met, such as when the actual recovery rate of a new disposal case is higher than the predicted recovery rate of the strategy recommended in the experience database, the platform automatically adds the new case to the experience database and updates the corresponding experience matching weight for subsequent strategy selection optimization.

[0038] The system can access an expert experience database to screen a set of non-performing asset disposal strategies. The screening criteria can be set as high disposal feasibility, high disposal returns, high disposal speed, etc., and candidate non-performing asset disposal strategies can be obtained based on the screening criteria.

[0039] The processing path or solution within the candidate asset processing strategy is broken down into multiple sub-tasks and corresponding task objectives for each sub-task, and the key time nodes of the tasks are determined. The personnel information table stored in the database is called to match the personnel suitability of different task nodes with the personnel information table, and one or more of the responsible entities, processing locations and processing times are planned. For the list of difficulties and contingency plans in the non-performing asset processing strategy, the implementation personnel and implementation methods are pre-assigned to form a complete structured disposal plan.

[0040] The above-mentioned optional embodiments, by screening the collection of non-performing asset disposal strategies based on expert experience to obtain candidate non-performing asset disposal strategies, can significantly improve the objectivity of decision-making. The collection of non-performing asset disposal strategies is obtained based on the scientifically quantified value assessment of non-performing assets and the disposal risk coefficient, thus possessing scientific rigor and objectivity. The candidate non-performing asset disposal strategies are selected from the collection of non-performing asset disposal strategies based on expert experience, combining practical operational experience, which can avoid personal subjective bias and improve the feasibility and flexibility of the disposal strategies.

[0041] By breaking down candidate non-performing asset disposal strategies into task-based solutions and generating structured disposal plans, complex processing procedures can be broken down, dependencies can be clearly identified, multi-line collaborative execution can be carried out, the overall disposal cycle can be shortened, disposal efficiency can be improved, and the controllability of the disposal process and results can be enhanced, thus realizing intelligent disposal of non-performing assets.

[0042] In an optional embodiment, obtaining information related to non-performing assets based on debtor and creditor information includes: Based on debtor information, a relational information graph is constructed, and operational status information and potential related risk information are obtained by searching along the links in the relational information graph; Based on debt information, target transaction case information and industry disposal benchmark parameters are matched from the market transaction case database; Information on non-performing assets is obtained based on operating status information, potential associated risk information, target transaction case information, and industry disposal benchmark parameters.

[0043] Using basic information on non-performing assets as the primary key, and after associating multiple data tables, the core objects of each piece of pending information in the multiple data tables are extracted. The core objects can mainly include two categories: debtors and creditors. The core objects are standardized according to preset attributes. The preset attributes of debtors can include ID card number, social security code, and address, while the preset attributes of creditors can include project name, contract number, and asset location. The extracted core objects are checked for relevance to the basic information on non-performing assets. If they are related, the information is retained as relevant information. Based on the same core objects, the subordinate relationship between each piece of relevant information and the core objects is established. According to the different core objects, the relevant information is divided into debtor information and creditor information.

[0044] Debtor information may include: debtor nature, ID number, unified social credit code, actual controller, equity structure, and business registration status; debt information may include: contract number, debt nature, cause of debt, asset amount and accrued interest, debt formation time, asset type, guarantee information, processing records, and repayment intention. Processing records can be generated based on processing paths or plans; for example, in debt collection, this may include collection methods, number of collections, collection time, and repayment records.

[0045] In constructing the information graph, based on the obtained debtor information, further multi-source data collection and expansion are carried out on the debtor information according to the preset target information. The preset target information may include business and financial status, asset and property information, litigation and legal information, creditor-debtor relationships and related party information, etc. The collected target information is cleaned, standardized and transformed. According to the preset pattern, the processed target information is transformed into specific nodes and relationships in the information graph. For example, the core objects in the debtor information are used as core nodes, and the entities in the target information are used as related nodes. Based on the content of the target information, the connections between core nodes and related nodes, and between related nodes are established. Core nodes and related nodes, and between related nodes are connected by edges. The edges are used to define the semantic connections between core nodes and related nodes, such as "Debtor XX - Actual Controller - XX Company".

[0046] Entities and relationships can be extracted from unstructured or semi-structured data using rules, natural language processing techniques, or large language models.

[0047] Operating status information and potential related risk information are obtained by searching the links in the related information graph. Specifically, the link search of the related information graph can be achieved by traversing the graph. Starting from the core node, the edges are used as paths, and all nodes are explored according to the traversal strategy. Information related to operating status and potential related risks is marked or extracted. The marked or extracted information is then integrated to obtain the operating status information and potential related risk information. The traversal strategy can be a depth-first search strategy or a breadth-first search strategy. For related nodes and terminal nodes, the related information graph can be further constructed as needed and connected to the original related information graph to complete the iterative update of the related information graph.

[0048] Understandably, the constructed relational information graph can be used to obtain information on operational status and potential related risks. It can eliminate obstacles such as multi-layered shareholding and complex guarantees, enabling comprehensive and accurate identification of the debtor's actual situation. Based on the transmission path of risk information, and through the layered association of risk signals, the traceability of information is enhanced, making it easier to determine the authenticity of information. Furthermore, it enhances the visibility of information clues, transforming complex relationships into visual graphics, which helps to achieve accurate decision-making. In addition, the construction of the relational network graph also helps to uncover implicit relationships between different entities. For example, graph mining algorithms can be used to accurately identify the actual controller in cases of complex equity structures or cross-shareholdings.

[0049] The market transaction case database stores transaction information, asset information, and relevant documents. Transaction information specifically includes the transaction price, appraised value, and transaction date. Asset information specifically includes the industry, asset type, asset ownership, asset status, and guarantee information. Relevant documents specifically include transaction contracts and appraisal reports.

[0050] It can dynamically detect major data sources, such as publicly available information from the People's Court Litigation Assets Network, local property rights exchanges, financial asset exchanges, and financial news, and update the market transaction case database based on this publicly available information. The industry disposal benchmark parameters can include valuation benchmarks, risk benchmarks, and transaction environment benchmarks. Valuation benchmarks can include one or more of the following: market value ratio, capitalization rate, premium rate, or discount rate. Risk benchmarks can include the overall number of lawsuits and the probability of loss. Transaction environment benchmarks can include transaction activity and realization cycle.

[0051] The methods for obtaining industry disposal benchmark parameters include: obtaining transaction case information about the specific non-performing asset industry from the market transaction case database; comprehensively analyzing the transaction case information to obtain the initial industry disposal benchmark parameters; and using multimodal dynamic data to fine-tune the initial industry disposal benchmark parameters to obtain the industry disposal benchmark parameters.

[0052] Fine-tuning the initial baseline parameters for industry disposal using multimodal dynamic data includes: analyzing the impact of multimodal dynamic data on these parameters; for example, the emergence of macroeconomic policy demands or heightened trading sentiment may shorten disposal cycles and increase premium rates. Appropriate influence weights are then assigned based on the impact of the multimodal dynamic data to fine-tune the initial baseline parameters. This fine-tuning compensates for data latency, improving the effectiveness and accuracy of the baseline parameters, thereby contributing to more precise decision-making.

[0053] The process involves matching target transaction case information from the market transaction case database, including: performing preliminary screening of transaction cases based on preset fields to obtain initial transaction cases, such as industry, region, and transaction time; calculating the similarity of the initial transaction cases based on the debt information of non-performing assets, sorting them in descending order of similarity, and selecting a preset number of initial transaction cases as target transaction cases; and obtaining the corresponding target transaction case information based on the target transaction cases.

[0054] The similarity score can be calculated by converting the features of the initial transaction case into vectors and measuring the degree of similarity with a specific non-performing asset by calculating the spatial distance between the vectors; or by setting a similarity weight value for each dimension of the non-performing asset's claim information, and the similarity weight value can be set and adjusted based on experience or expert opinions; the similarity score of each initial transaction case is calculated based on the similarity weight value of each dimension, and the similarity score is used to measure the similarity.

[0055] Information on non-performing assets is obtained based on operating status information, potential associated risk information, target transaction case information, and industry disposal benchmark parameters.

[0056] Understandably, based on basic information about non-performing assets, related information about non-performing assets, including information on operating conditions, potential associated risks, target transaction cases, and industry disposal benchmark parameters, is obtained. A standardized process for information acquisition has been established, which effectively extends information about non-performing assets, improves the ability to search for and integrate non-performing asset information, provides comprehensive information for decision-making, improves the accuracy and effectiveness of information, reduces information processing load, reduces waste of computing resources, and improves information processing efficiency.

[0057] In an optional embodiment, the non-performing asset valuation model includes a first sub-model and a second sub-model. The first sub-model can be an improved LSTM, and the second sub-model can be a gradient boosting tree.

[0058] In an optional embodiment, the non-performing asset valuation model, after training, obtains the non-performing asset valuation value and disposal risk coefficient based on debtor information, creditor information, multimodal dynamic data, non-performing asset correlation information, and non-performing asset amount, including: Extract time-series and static data from multimodal dynamic data and non-performing asset correlation information; Input the time series data into the first sub-model, and output the time series feature evaluation vector through the first sub-model; Input debtor information, creditor information, static data, and non-performing asset amount into the second sub-model, and output a non-linear static feature evaluation vector through the second sub-model; Obtain the first objective weight of the first sub-model and the second objective weight of the second sub-model; The fusion evaluation vector is obtained based on the first objective weight and the time-series feature evaluation vector, the second objective weight, and the nonlinear static feature evaluation vector. The fusion assessment vector is input into the fully connected layer, and the fully connected layer outputs the non-performing asset value assessment value and disposal risk coefficient.

[0059] Time-series data refers to dynamic data with time-series attributes, such as business status information, potential related risk information, industry processing benchmark parameters, debtor public opinion data, and macroeconomic prosperity data; while static data mainly refers to data without time attributes that does not change easily, such as basic data of collateral and transaction case information.

[0060] More specifically, before being input into the first sub-model, time-series data can be categorized into a time-series feature set and subjected to standardization, missing value imputation, and sequence length unification. Before being input into the second sub-model, debtor information, creditor information, static data, and non-performing asset amounts can be categorized into a static feature set and subjected to one-hot encoding, feature normalization, and outlier removal.

[0061] The first sub-model uses an improved LSTM model and introduces a forget gate optimization mechanism. By setting a reasonable time step and the number of neurons in the hidden layer, it can extract long-short-term dependencies from the temporal feature set and output a temporal feature evaluation vector. The second sub-model uses a gradient boosting tree model based on the XGBoost architecture. By constructing multiple decision trees and iteratively training based on the residuals of the previous tree, it can uncover complex mapping relationships between nonlinear static features and finally output a nonlinear static feature evaluation vector.

[0062] The dimension of the time series feature evaluation vector is 1*N, where N represents the preset time series feature evaluation dimension; the dimension of the nonlinear static feature evaluation vector is consistent with the dimension of the time series feature evaluation vector.

[0063] The temporal feature evaluation vector is used to reflect the importance of different time points and different dimensions in the temporal feature, while the nonlinear static feature evaluation vector is used to reflect the importance of different dimensions in the static feature.

[0064] The fusion evaluation vector is obtained by fusing the weighted calculation of the time-series feature evaluation vector based on the first objective weight and the weighted calculation of the nonlinear static feature evaluation vector based on the second objective weight. It is a unified representation of time-series data and static data.

[0065] The fully connected layer of the non-performing asset valuation model is configured with a weight matrix and a bias vector. The hidden layer in the connected layer can perform a linear transformation on the input fusion valuation vector to support further calculation and processing of the fully connected layer. The output layer is configured with two output nodes to output the real-time valuation value and the disposal risk coefficient. The weight matrix and bias vector of the output layer can be set independently of the hidden layer to calculate and output the non-performing asset valuation value and the disposal risk coefficient.

[0066] Understandably, information on non-performing assets is categorized into time-series data and non-linear static data, and processed according to the data type. Specifically, a first sub-model is used to process time-series data, and a second sub-model is used to process static data. This approach fully considers the inherent characteristics and information value of different data types. The time-series feature assessment vector reflects changes in the value of non-performing assets, predicting trends or revealing hidden risks. The non-linear static feature assessment vector reflects the basic condition of non-performing assets and is the foundation for value assessment. By working collaboratively with the first and second sub-models, the accuracy and effectiveness of the processing are improved, the data processing speed is accelerated, and the characteristics and value of different data types are fully utilized.

[0067] Specifically, the improved LSTM model effectively captures long-term dependencies and dynamic patterns in sequences, while the XGBoost architecture's gradient boosting tree model excels at handling structured tabular data, automatically and efficiently capturing complex nonlinear relationships and feature interactions. The architectures of the first and second sub-models are adapted to the corresponding data types being processed, thereby improving the accuracy, effectiveness, and efficiency of feature evaluation vector computation.

[0068] Different types of non-performing assets (NPLs) exhibit significant differences in their value drivers and risk sources. Similarly, the importance and information content of time-series and static data vary across different asset types. The first and second initial weights can be derived from historical data during the training of the NPL valuation model, and different first and second initial weights can be trained for different types of NPLs to accommodate the varying importance or value of time-series and static data for different types of NPLs. For example, in NPLs involving commercial real estate mortgage loans, asset value is highly dependent on time-series indicators such as market rent and vacancy rates; therefore, the first initial weight will be set relatively higher than for other types of NPLs. Conversely, in NPLs involving unsecured consumer loans, the recovery prospects depend more on the debtor's occupation, income, and historical credit history; therefore, the second initial weight will be set relatively higher than for other types of NPLs.

[0069] In an optional embodiment, obtaining the first target weights of the first sub-model includes: The first initial weight is determined based on the asset type of the non-performing assets; A timeliness score is calculated based on the acquisition time and update frequency of multimodal dynamic data; The first correction factor is determined based on the timeliness score; The first initial weight is corrected using the first correction factor to obtain the first target weight.

[0070] Based on the different types of core objects, non-performing assets can be divided into non-performing assets of legal persons and non-performing assets of natural persons. A first initial weight and a second initial weight are set for different types of non-performing assets. For example, the first initial weight of non-performing assets of legal persons can be set to 0.4 and the second initial weight can be set to 0.6; the first initial weight of non-performing assets of natural persons can be set to 0.3 and the second initial weight can be set to 0.7.

[0071] The average update frequency for each multimodal dynamic data point corresponding to a specific data subtype is obtained. A value decay curve is constructed by combining the average update frequency and the specific data subtype for each multimodal dynamic data point. The value decay curve can be an exponential decay curve, a stepped decay curve, or a linear decay curve. The value decay curve includes both a time dimension and a value dimension, and the value quantity of each value dimension within the corresponding value decay curve is obtained based on the collection time of each multimodal dynamic data point. Finally, a timeliness score is obtained by averaging or weighted averaging the value quantities of all multimodal dynamic data. Both the value quantity of the value dimension and the timeliness score can be expressed on a percentage scale. Each piece of multimodal dynamic data corresponds to a specific data subtype, such as debtor sentiment data, macroeconomic prosperity data, and industry processing benchmark parameters. Debtor sentiment data is suitable for exponential decay curves, macroeconomic prosperity data, and industry processing benchmark parameters are suitable for step decay curves; while asset site image data is suitable for linear decay curves. The parameters of the curve, such as the slope, are generally set with reference to the data update frequency. However, if the weighted average calculation method is used, the timeliness of the data type needs to be considered, and higher weights should be assigned to specific data subtypes with stronger timeliness.

[0072] A first correction factor is determined based on the timeliness score; the first initial weight is then adjusted using the first correction factor to obtain a first target weight. For example, the first correction factor is determined based on the ratio of the timeliness score to the full timeliness score. If the timeliness score is 90 points and the full score is 100 points, then the first correction factor is 0.9, the first initial weight is 0.4, and the first initial weight is adjusted using the first initial weight to obtain a first target weight of 0.36.

[0073] Understandably, on the one hand, the timeliness score is calculated based on the collection time and update frequency of multimodal dynamic data, and the first correction factor is determined. Furthermore, the first initial weight is adjusted based on this first correction factor, enhancing the processing flexibility of the first sub-model. It can dynamically adjust the first target weight according to the timeliness of time-series data, reflecting the significance of the data's timeliness and thus improving the accuracy and reliability of predictions. On the other hand, the first initial weight is determined based on the type of non-performing assets, while the first target weight is set considering timeliness based on the first initial weight. This achieves a dual dynamic adjustment of the first target weight, fully demonstrating the importance of time-series data to the type of non-performing assets, and providing more reasonable and scientific data support for decision-making.

[0074] In an optional embodiment, obtaining the second objective weights of the second sub-model includes: The second initial weight is determined based on the asset type of the non-performing assets; A completeness score is calculated based on the number of missing items and key items in debtor and creditor information. The second correction factor is determined based on the integrity score; The second initial weights are corrected using a second correction factor to obtain the second target weights.

[0075] The process involves acquiring a pre-defined complete set of debtor and creditor information, traversing the debtor and creditor information to identify missing and key items, and statistically analyzing the number of missing and key items. Based on the acquired complete set of debtor and creditor information and the number of missing items, a missing rate is calculated. A penalty value is also calculated based on the acquired complete set of debtor and creditor information and the number of key items. Finally, a completeness score is calculated based on the missing rate and the penalty value. Specifically, the missing rate = number of missing items / number of items in the complete set of debtor and creditor information; the completeness score = full score - (base missing rate × full score + penalty coefficient). The penalty coefficient is derived from the number of key items or the number of missing key items; the more missing items, the higher the penalty coefficient becomes, and the specific step-by-step values ​​can be adjusted and set according to actual needs. Missing items refer to missing information, while key items refer to crucial information, such as amount, whether there is collateral, and the current status of assets.

[0076] A second correction factor is determined based on the integrity score; the second initial weight is then adjusted using the second correction factor to obtain the second target weight. For example, the second correction factor is determined based on the ratio of the integrity score to the full integrity score. If the integrity score is 95 points and the full score is 100 points, then the second correction factor is 0.95, the second initial weight is 0.6, and the second initial weight is adjusted using the second correction factor to obtain a second target weight of 0.57.

[0077] Understandably, on the one hand, a completeness score is calculated based on the missing and key items of debtor and creditor information of non-performing assets, and a second correction factor is determined. The second initial weights are then adjusted based on this second correction factor, enhancing the flexibility of the second sub-model. The second target weights can be dynamically adjusted according to the completeness of the debtor and creditor information of non-performing assets, reflecting the criticality and sufficiency of the data, thereby improving the accuracy and reliability of the model's predictions. On the other hand, the second initial weights are also determined based on the type of non-performing assets. The second target weights are set based on the second initial weights, further considering completeness and criticality, achieving a dual dynamic adjustment of the second target weights. This also provides more reasonable and scientific data support for decision-making. Furthermore, by specifically adjusting the first and second initial weights, the accuracy of the adjustment is effectively improved, enhancing data processing quality. The two work synergistically, further improving the reliability and scientific nature of the decision-making process.

[0078] In an optional embodiment, the fused evaluation vector is obtained based on the first target weight and the time-series feature evaluation vector, the second target weight, and the nonlinear static feature evaluation vector, including: Determine whether the sum of the first target weight and the second target weight is 1; When the sum is not 1, the weight deviation is calculated based on the first target weight and the second target weight. The contribution ratio of features is determined based on time-series feature evaluation vectors and nonlinear static feature evaluation vectors. The weight balance factor is obtained based on the degree of weight deviation and the proportion of feature contribution. The first target weight and the second target weight are corrected based on the weight balancing factor to obtain the first corrected weight and the second corrected weight. The sum of the first corrected weight and the second corrected weight is 1. The fusion evaluation vector is obtained based on the first modified weight and the time series feature evaluation vector, the second modified weight, and the nonlinear static feature evaluation vector.

[0079] As in the example above, if the weight of the first objective is 0.36 and the weight of the second objective is 0.57, then the sum of the weights of the first and second objectives is clearly not 1.

[0080] The formula for calculating the weight deviation based on the first target weight and the second target weight is β = |W1 +W2 - 1|, where β ∈ [0, +∞). Where W1 represents the first objective weight, W2 equals the second objective weight, and β represents the degree of weight deviation. The larger the β value, the more serious the deviation of the weight from the normalized state.

[0081] The feature contribution ratio is determined based on the temporal feature evaluation vector and the nonlinear static feature evaluation vector. For example, the temporal feature evaluation vector and the nonlinear static feature evaluation vector are obtained separately. The absolute value of each dimension in the temporal feature evaluation vector is extracted, and the sum of the absolute values ​​of each dimension in the temporal feature evaluation vector is used to obtain the first value. Similarly, the absolute value of each dimension in the nonlinear static feature evaluation vector is extracted, and the sum of the absolute values ​​of each dimension in the nonlinear static feature evaluation vector is used to obtain the second value. The ratio of the first value to the second value is used to obtain the feature contribution ratio of the temporal feature evaluation vector and the nonlinear static feature evaluation vector.

[0082] Optionally, the sum of the absolute values ​​of each dimension in the time-series feature evaluation vector and the sum of the absolute values ​​of each dimension in the nonlinear static feature evaluation vector can be calculated using the Manhattan distance or the first norm formula, respectively.

[0083] Optionally, the feature contribution percentage α = S1 / (S1 + S2), α∈ (0,1); Where S1 represents the first value, S2 represents the second value, α represents the contribution ratio of the time-series feature evaluation vector, 1-α represents the contribution ratio of the nonlinear static feature evaluation vector, and α and 1-α reflect the actual influence weight of the two types of features on the evaluation results.

[0084] The weight balance factor λ = α × (1 + β) is obtained based on the degree of weight deviation and the proportion of feature contribution. λ is used to balance the weight deviation and the actual contribution of features.

[0085] The first target weight and the second target weight are corrected based on the weight balancing factor to obtain the first corrected weight and the second corrected weight. The sum of the first corrected weight and the second corrected weight is 1. The formula for calculating the first corrected weight W1' is W1' = W1 × λ / (W1 × λ + W2 × (1 -λ)), and the formula for calculating the second corrected weight W2' is W2' = W2 × (1 - λ) / (W1 × λ + W2 × (1 -λ)). This ensures that the first and second corrected weights meet the normalization requirements and retain the matching between the relative proportions of the original weights and the feature contribution.

[0086] The fused evaluation vector is obtained based on the first corrected weight and the time-series feature evaluation vector, the second corrected weight, and the nonlinear static feature evaluation vector. Specifically, the time-series feature evaluation vector is weighted using the first corrected weight, and the nonlinear static feature evaluation vector is weighted using the second corrected weight. The weighted values ​​of the time-series feature evaluation vector and the nonlinear static feature evaluation vector are then added together to obtain the fused evaluation vector.

[0087] Understandably, revising the weights of the first and second objectives helps to enhance the standardization of the weights and improve the stability of the non-performing asset valuation model. Furthermore, adjusting the degree of weight deviation aligns with the aforementioned adjustments to the first and second initial weights, enhancing the adaptability of the non-performing asset valuation model to different types of non-performing assets. The weight balancing factor, introduced based on the degree of feature contribution and weight deviation, can objectively quantify the importance of time-series feature evaluation vectors and nonlinear static feature evaluation vectors in the model, thereby achieving a more reasonable weight allocation, avoiding the disruption of the original weight logic by conventional normalization, and ensuring the rationality of the fusion results.

[0088] In an optional embodiment, the method further includes: after the non-performing asset disposal is completed, generating a disposal effect assessment report based on the actual recovered value of the non-performing asset and the assessed value of the non-performing asset.

[0089] A pre-set template for a disposal effectiveness evaluation report is provided. Upon receiving the actual recovered value and assessed value of non-performing assets, the system automatically generates or guides the user to fill out the report based on this template. For example... Figure 4 As shown, the disposal effect assessment report may include one or more of the following: asset identification, disposal path, actual recovered value of non-performing assets, assessed value of non-performing assets, and recovery rate of non-performing assets.

[0090] Example 2 Figure 5 This is a schematic diagram of the module of the intelligent non-performing asset processing device based on value assessment provided in Embodiment 2 of this application.

[0091] In some embodiments, the intelligent non-performing asset processing device based on valuation may include multiple functional modules composed of computer program segments. The computer programs for each program segment of the intelligent non-performing asset processing device based on valuation may be stored in the memory of an electronic device and executed by at least one processor to perform (see details). Figure 1 (Description) This describes a function for intelligent processing of non-performing assets based on value assessment. According to the functions it performs, it can be divided into multiple functional modules. These functional modules may include: a first acquisition module 501, a second acquisition module 502, an asset assessment module 503, a strategy generation module 504, an asset processing module 505, and a report generation module 506. The term "module" in this application refers to a series of computer program segments that can be executed by at least one processor and perform a fixed function, stored in memory. In this embodiment, the functions of each module will be detailed in subsequent embodiments.

[0092] The first acquisition module 501 is used to acquire basic information on non-performing assets and multimodal dynamic data; The second acquisition module 502 is used to acquire related information of non-performing assets based on the debtor information and creditor information of the non-performing assets; The asset valuation module 503 is used to obtain the valuation value of non-performing assets and the disposal risk coefficient based on debtor information, creditor information, multimodal dynamic data, non-performing asset correlation information and non-performing asset amount through a trained non-performing asset valuation model. The strategy generation module 504 is used to obtain a set of non-performing asset disposal strategies based on the non-performing asset valuation and disposal risk coefficient. The asset processing module 505 is used to process non-performing assets based on a set of non-performing asset processing strategies. The report generation module 506 is used to generate a disposal effect evaluation report based on the actual recovered value and the assessed value of the non-performing assets after the disposal of non-performing assets is completed.

[0093] It should be understood that the various variations and specific embodiments of the intelligent non-performing asset processing method based on value assessment provided in Embodiment 1 above are also applicable to the intelligent non-performing asset processing device based on value assessment in this embodiment. Through the detailed description of the intelligent non-performing asset processing method based on value assessment described above, those skilled in the art can clearly understand the implementation process of the intelligent non-performing asset processing device based on value assessment in this embodiment. For the sake of brevity, it will not be described in detail here.

[0094] Example 3 This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the intelligent processing method for non-performing assets based on value assessment according to Embodiment 1.

[0095] Example 4 See Figure 6 The diagram shown is a structural schematic of an electronic device according to an embodiment of this application.

[0096] Electronic device 6 includes a memory 61, at least one processor 62, and at least one communication bus 63. Those skilled in the art should understand that... Figure 6 The structure of the electronic device shown does not constitute a limitation of the embodiments of this application. It can be a bus structure or a star structure. The electronic device may also include more or fewer other hardware or software than shown, or different component arrangements.

[0097] In some embodiments, the electronic device is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), programmable gate arrays (FPGAs), digital processors, and embedded devices. The electronic device may also include user equipment, which includes, but is not limited to, any electronic product capable of human-computer interaction with a user via a keyboard, mouse, remote control, touchpad, or voice control device, such as a personal computer, tablet computer, smartphone, or digital camera.

[0098] In the embodiments provided in this application, it should be understood that the disclosed methods, apparatuses, computer-readable storage media, and electronic devices can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple components or modules may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices, components, or modules may be electrical, mechanical, or other forms.

[0099] The components described as separate parts may or may not be physically separate. The components shown as components may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the components can be selected to achieve the purpose of this embodiment according to actual needs.

[0100] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each component can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0101] If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the intelligent data backup method for electronic devices described in the various embodiments of the present invention. The aforementioned storage medium includes: USB flash drive, portable hard drive, read-only memory (ROM). Various media that can store program code, such as only memory, random access memory (RAM), magnetic disks, or optical disks.

[0102] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0103] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0104] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for intelligent processing of non-performing assets based on value assessment, characterized in that, The method includes: Obtain basic information and multimodal dynamic data on non-performing assets; Based on the debtor and creditor information of non-performing assets, obtain related information about non-performing assets; The non-performing asset valuation model, after training, yields the valuation value and disposal risk coefficient of non-performing assets based on the debtor information, the creditor information, the multimodal dynamic data, the non-performing asset association information, and the amount of non-performing assets. Based on the assessed value of the non-performing assets and the disposal risk coefficient, a set of non-performing asset disposal strategies is obtained; The non-performing assets are processed based on the aforementioned set of non-performing asset disposal strategies.

2. The intelligent processing method for non-performing assets based on value assessment according to claim 1, characterized in that, The processing of non-performing assets based on the set of non-performing asset processing strategies includes: Based on expert experience, the set of non-performing asset disposal strategies is screened to obtain candidate non-performing asset disposal strategies; The candidate non-performing asset disposal strategies are decomposed into tasks and structured disposal solutions are generated. The non-performing assets are processed based on the aforementioned structured disposal scheme.

3. The intelligent processing method for non-performing assets based on value assessment according to claim 1, characterized in that, The acquisition of related information on non-performing assets based on debtor and creditor information includes: Based on the debtor information, a correlation information graph is constructed, and the business status information and potential correlation risk information are obtained by searching the correlation information graph links. Based on the aforementioned debt information, target transaction case information and industry disposal benchmark parameters are matched from the market transaction case database; Non-performing asset association information is obtained based on the aforementioned operating status information, potential associated risk information, target transaction case information, and industry disposal benchmark parameters.

4. The intelligent processing method for non-performing assets based on value assessment according to claim 1, characterized in that, The non-performing asset valuation model includes a first sub-model and a second sub-model. The trained non-performing asset valuation model, based on the debtor information, the creditor information, the multimodal dynamic data, the non-performing asset correlation information, and the non-performing asset amount, yields the non-performing asset valuation value and disposal risk coefficient, including: Time-series data and static data are obtained from the multimodal dynamic data and the non-performing asset association information; The time series data is input into the first sub-model, and the time series feature evaluation vector is output through the first sub-model. The debtor information, the creditor information, the static data, and the amount of non-performing assets are input into the second sub-model, and the non-linear static feature evaluation vector is output through the second sub-model. Obtain the first target weight of the first sub-model, and obtain the second target weight of the second sub-model; A fusion evaluation vector is obtained based on the first target weight and the temporal feature evaluation vector, the second target weight and the nonlinear static feature evaluation vector; The fusion evaluation vector is input into the fully connected layer, and the fully connected layer outputs the non-performing asset value assessment value and disposal risk coefficient.

5. The intelligent processing method for non-performing assets based on value assessment according to claim 4, characterized in that, The step of obtaining the first target weight of the first sub-model includes: The first initial weight is determined based on the asset type of the non-performing assets; A timeliness score is calculated based on the acquisition time and update frequency of the multimodal dynamic data. The first correction factor is determined based on the timeliness score; The first initial weight is corrected using the first correction factor to obtain the first target weight.

6. The intelligent processing method for non-performing assets based on value assessment according to claim 4, characterized in that, The process of obtaining the second objective weight of the second sub-model includes: The second initial weight is determined based on the asset type of the non-performing assets; A completeness score is calculated based on the number of missing items and key items in the debtor information and the claim information. A second correction factor is determined based on the integrity score; The second initial weight is corrected using the second correction factor to obtain the second target weight.

7. The intelligent processing method for non-performing assets based on value assessment according to any one of claims 4 to 6, characterized in that, The step of obtaining the fusion evaluation vector based on the first target weight and the time-series feature evaluation vector, the second target weight and the nonlinear static feature evaluation vector includes: Determine whether the sum of the first target weight and the second target weight is 1; When the sum is not 1, the weight deviation is calculated based on the first target weight and the second target weight; The feature contribution ratio is determined based on the time-series feature evaluation vector and the nonlinear static feature evaluation vector. The weight balance factor is obtained based on the degree of weight deviation and the proportion of feature contribution. The first target weight and the second target weight are corrected based on the weight balancing factor to obtain a first corrected weight and a second corrected weight, and the sum of the first corrected weight and the second corrected weight is 1. The fusion evaluation vector is obtained based on the first modified weight and the time-series feature evaluation vector, the second modified weight and the nonlinear static feature evaluation vector.

8. The intelligent processing method for non-performing assets based on value assessment according to claim 1, characterized in that, The method further includes: After the disposal of non-performing assets is completed, a disposal effectiveness evaluation report is generated based on the actual recovered value of the non-performing assets and the assessed value of the non-performing assets.

9. An electronic device, characterized in that, The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the intelligent processing method for non-performing assets based on value assessment as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent processing method for non-performing assets based on value assessment as described in any one of claims 1 to 8.

Images