A deep learning-based enterprise payment risk management system
Through the enterprise payment risk management system based on deep learning, a transaction cycle chain and risk node cluster is built, and personalized evaluation and management of payment risks is achieved, which solves the problem that traditional systems are difficult to identify and prevent payment risks, and improves payment security and risk management effects.
Patent Information
- Application Number
- CN202411000764.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-25
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-07-25
AI Technical Summary
Traditional enterprise payment risk management systems rely on manual experience and rule formulation, making it difficult to effectively identify and prevent payment risks, resulting in high payment risks and difficult to cope with complex payment methods under the digital economy.
Adopt a deep learning-based enterprise payment risk management system, obtain optimized historical payment data through the extreme value filtering module, build a transaction cycle chain module, the risk singularity module identifies frequent transaction nodes, the risk situation evolution module analyzes the risk node cluster, and risk assessment The module conducts personalized risk quantitative assessment, and the payment management module conducts differentiated risk control decisions, and builds a systematic risk management framework.
It has achieved a comprehensive understanding of payment behavior and risk management, reduced payment risks, improved payment security, more accurately identified and warned of payment risks, and provided personalized risk management strategies and control measures.
Smart Images

Figure CN118537018B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of payment risk management, and in particular to an enterprise payment risk management system based on deep learning. Background Art
[0002] With the rapid development of the digital economy, corporate payment activities are becoming more frequent and complex. In addition to conventional accounts payable and accounts receivable management, companies also need to deal with various emerging payment methods, such as electronic payment, mobile payment, etc. At the same time, corporate payment activities are also faced with various risks such as fraud, overdue, and broken capital chain. Once a payment risk occurs, it will cause serious financial losses and reputation damage to the company. The traditional corporate payment risk management system mainly relies on manual experience and rule-making, and it is difficult to identify and prevent various payment risks in a timely and effective manner. There is often a problem of high payment risk in enterprises. With the continuous development of big data and artificial intelligence technology, there is an urgent need to explore an intelligent corporate payment risk management system to achieve intelligent analysis and accurate early warning of payment behavior. Summary of the invention
[0003] In order to solve the above technical problems, the present invention proposes an enterprise payment risk management system based on deep learning to solve at least one of the above technical problems.
[0004] To achieve the above object, the present invention provides an enterprise payment risk management system based on deep learning, comprising the following steps:
[0005] The extreme value filtering module is used to: obtain the enterprise's historical payment data and online payment information; perform extreme value sample filtering on the enterprise's historical payment data to obtain optimized historical payment data; obtain a historical payment transaction feature sequence based on the optimized historical payment data;
[0006] The cycle chain module is used to: analyze the transaction flow direction of the historical payment transaction feature sequence to obtain transaction flow direction data; reconstruct the global transaction cycle based on the transaction flow direction data to build the enterprise transaction cycle chain;
[0007] The risk singularity module is used to: calculate the transaction frequency between enterprises on the enterprise transaction cycle chain to obtain frequent transaction nodes; perform transaction risk analysis on the frequent transaction nodes to obtain the risk singularity of the transaction cycle chain;
[0008] The risk situation evolution module is used to: identify the adjacent risk diffusion of the risk singularity of the transaction cycle chain to obtain the risk node cluster; perform risk situation evolution on the risk node cluster to construct a dynamic risk profile of the enterprise;
[0009] The risk assessment module is used to: use the enterprise dynamic risk profile to conduct personalized risk quantitative assessment on online payment information, so as to obtain the enterprise payment risk level;
[0010] The payment control module is used to make differentiated risk control decisions on online payment information based on the enterprise payment risk level and build an enterprise risk payment control model.
[0011] The present invention acquires the enterprise's historical payment data and online payment information to establish a comprehensive understanding of payment behavior, and eliminates abnormal data through extreme value sample filtering to obtain more accurate historical payment data. By extracting the characteristic sequence of historical payment transactions, the law and trend of payment behavior are revealed. By analyzing the transaction flow direction, the flow direction and path of payment transactions are understood, and the pattern of enterprise capital flow is revealed. The global transaction cycle reconstruction integrates and optimizes the transaction cycle, and establishes an enterprise transaction cycle chain to provide a basis for subsequent analysis and decision-making. By calculating the transaction frequency between enterprises, the nodes with frequent transactions, that is, the nodes with frequent transactions with other enterprises, are determined. Transaction risk analysis is performed on the nodes with frequent transactions to identify nodes with higher transaction risks, that is, risk singularities in the transaction cycle chain, to provide a basis for subsequent risk analysis and management. By expanding the adjacent risks Scattered identification is used to determine the risk node clusters adjacent to the risk singularity, that is, the node groups with a high correlation with the risk nodes. The risk situation evolution analysis of the risk node clusters is carried out to reveal the development trend and evolution law of risks, and a dynamic risk profile of the enterprise is constructed to provide a reference for risk management and response. By using the dynamic risk profile of the enterprise, a personalized risk quantitative assessment is conducted on online payment information to help enterprises accurately assess the risk level of each payment, and classify and sort it according to the risk level, so as to better understand and manage payment risks. By making differentiated risk management decisions based on the enterprise payment risk level, corresponding risk management strategies and control measures are adopted for payment information of different risk levels, and an enterprise risk payment management model is constructed to help enterprises establish a systematic risk management framework, effectively reduce payment risks and improve payment security.
[0012] Preferably, the extreme value filtering module is specifically used for:
[0013] Obtain the company's historical payment data and online payment information;
[0014] Perform extreme value sample filtering on the enterprise's historical payment data to obtain optimized historical payment data;
[0015] Extracting transaction features from the optimized historical payment data to obtain historical payment transaction feature data;
[0016] Heterogeneous data fusion processing is performed on historical payment transaction feature data to obtain a historical payment transaction feature sequence.
[0017] The present invention establishes a comprehensive understanding of payment behavior by acquiring the company's historical payment data and online payment information. These data include key information such as past payment records, transaction direction, amount, time, etc., which provide basic data for subsequent analysis and decision-making. By filtering the historical payment data with extreme value samples and excluding abnormal values and outliers, more accurate and reliable historical payment data are obtained, which helps to improve the accuracy of subsequent analysis and modeling and reduce misjudgments and deviations caused by abnormal data. By extracting transaction features from the optimized historical payment data, key feature information such as transaction amount, transaction type, counterparty, business node and payment path are extracted from it. These feature data help understand the pattern, law and correlation of payment behavior and provide a basis for subsequent risk analysis and decision-making. By performing heterogeneous data fusion processing on the historical payment transaction feature data, data from different sources and different types are integrated and unified, which helps to eliminate the differences between data, improve the consistency and comparability of data, and obtain a series of ordered historical payment transaction feature sequences. These sequences are used for subsequent analysis and modeling to reveal the pattern and trend of payment risks.
[0018] Preferably, the periodic chain module is specifically used for:
[0019] Identify missing values in the company's historical payment data to obtain missing payment values;
[0020] Perform proportional statistics on missing values of payments to generate missing value ratios;
[0021] Based on the missing value ratio, the payment missing values are classified by random correlation, the randomness of the missing values is determined, and random missing values and non-random missing values are generated;
[0022] Perform mean filling processing on random missing values to obtain mean-filled payment data;
[0023] Performing outlier evaluation on non-random missing values based on a preset normative data sample, and determining that the non-random missing value is an abnormal extreme value sample when the non-random missing value exceeds the preset normative data sample range;
[0024] Perform extreme value sample filtering on abnormal extreme value samples to obtain extreme value filtered payment data;
[0025] Optimized historical payment data is obtained based on mean filling payment data and extreme value filtering payment data.
[0026] The present invention identifies missing values in historical payment data to understand which data items are missing, and evaluates the integrity of historical payment data by counting the proportion of missing values, which helps to understand the degree of missing payment data. By classifying missing values by random correlation and judging the randomness of missing values, it helps to distinguish random missing values from non-random missing values, and provide guidance for subsequent processing and filling strategies. By performing mean filling processing on random missing values, missing data can be restored to a certain extent, which helps to improve the integrity and availability of historical payment data. By performing outlier evaluation on non-random missing values, it is judged whether they exceed the preset standard data sample range, thereby identifying abnormal extreme value samples. By filtering abnormal extreme value samples, abnormal values and outliers are excluded to obtain more accurate and reliable extreme value filtered payment data. By combining the payment data that has been processed by mean filling and extreme value filtering, optimized historical payment data is obtained. These data have high integrity, accuracy and reliability, and provide more reliable basic data for enterprise payment risk management.
[0027] Preferably, the risk singularity module is specifically used for:
[0028] Identify transaction nodes of historical payment transaction feature sequences and mark node timestamps to obtain transaction feature nodes;
[0029] Analyze the transaction flow direction of transaction feature nodes to obtain transaction flow direction data;
[0030] Based on the transaction flow direction data, the transaction path is traced back to obtain the transaction path;
[0031] Reconstruct the global transaction cycle of the transaction path and build the enterprise transaction cycle chain.
[0032] The present invention divides continuous payment transaction sequences into different transaction nodes by performing node identification and timestamp marking on historical payment transaction feature sequences, which helps to convert complex transaction data into more manageable and analyzable transaction feature nodes. By analyzing the flow direction between transaction feature nodes, the association and flow between different transaction nodes can be understood, which helps to reveal the process and path of payment transactions. By performing path tracing processing on transaction flow direction data, the specific transaction path from the initial node to the final node can be determined, which helps to gain a deep understanding of the trajectory and evolution of payment transactions and provides support for discovering potential risk points and abnormal situations. By performing global transaction cycle reconstruction on the transaction path, the enterprise's transaction cycle chain can be identified, which helps to determine the periodic pattern and cycle length of payment transactions and provide a reference for enterprise payment risk management. The construction of the cycle chain also helps to discover periodic risks and laws, and then take corresponding risk management measures.
[0033] Preferably, the risk situation evolution module is specifically used for:
[0034] Calculate the transaction frequency between enterprises in the enterprise transaction cycle chain and generate transaction frequency data;
[0035] Based on the transaction frequency data, frequent nodes are extracted from the transaction feature nodes to obtain frequent transaction nodes;
[0036] Conduct transaction risk analysis on nodes with frequent transactions and generate transaction risk data;
[0037] Based on the potential transaction risk data, risk vulnerability analysis is conducted on the enterprise transaction cycle chain to obtain the risk singularity of the transaction cycle chain.
[0038] The present invention calculates the transaction frequency of the enterprise transaction cycle chain to understand the transaction frequency between the enterprise and other participants, which helps to determine the transaction activity of the enterprise in a specific time period and find out whether there are abnormal high-frequency or low-frequency transactions. By analyzing the transaction frequency data, frequently occurring transaction feature nodes are extracted. These frequent nodes represent common patterns and behaviors in payment transactions, which help to identify conventional transaction activities. By extracting frequent transaction nodes, the normal transaction behavior of the enterprise is better understood. By performing transaction risk analysis on frequent transaction nodes, the potential risks brought by each frequent node are evaluated, which includes identifying potential fraud, violations or abnormal transaction behaviors, as well as other factors that lead to payment risks. The generated transaction risk data helps the enterprise identify and prioritize transaction nodes with higher risks to reduce payment risks. By analyzing the potential transaction risk data, risk loopholes in the enterprise transaction cycle chain are identified, which includes discovering weaknesses, abnormal nodes or potential risk concentration areas in the transaction cycle chain. By analyzing the risk singularities of the transaction cycle chain, the enterprise identifies and repairs potential risk loopholes and improves the effect of payment risk management.
[0039] Preferably, the risk assessment module is specifically used for:
[0040] Identify the adjacent risk diffusion of the risk singularity of the transaction cycle chain and generate the risk diffusion trajectory;
[0041] Based on the risk diffusion trajectory, the risk probability distribution of the enterprise transaction cycle chain is calculated to obtain the risk node cluster;
[0042] Evolve the risk situation of risk node clusters and generate cycle chain risk situation evolution data;
[0043] Conduct dynamic risk prediction on the evolution data of cycle chain risk trends and build a dynamic risk profile of the enterprise.
[0044] The present invention identifies the neighboring risk diffusion of the risk singularity of the transaction cycle chain, determines the neighboring nodes around the risk singularity, and tracks its risk diffusion path, which helps to understand the process of risk propagation from the singularity to the surrounding nodes, and reveals the affected transaction nodes and paths. By analyzing the risk diffusion trajectory, the risk probability distribution of each node on the enterprise transaction cycle chain is calculated, which helps to determine the relative risk level of the risk node and aggregate it into a risk node cluster. By obtaining the risk node cluster, the enterprise can better understand the distribution of risks, focus on high-risk areas, and take corresponding risk management measures. By performing risk situation evolution analysis on the risk node cluster, the evolution trend and change pattern of the risk nodes in the cycle chain are understood, which helps to discover the dynamic changes of the risk nodes, identify the risk growth trend or attenuation trend, and provide real-time risk situation information for the enterprise payment risk management. By performing dynamic risk prediction on the cycle chain risk situation evolution data, the development trend and risk evolution of the future risk nodes are predicted, which helps the enterprise to timely identify and predict potential risk points, formulate corresponding risk management strategies, and build a dynamic risk profile of the enterprise. The dynamic risk profile provides real-time risk status and trend information to help the enterprise make accurate decisions and countermeasures.
[0045] Preferably, the specific steps of calculating the risk probability distribution of the enterprise transaction cycle chain based on the risk diffusion trajectory to obtain the risk node cluster are:
[0046] Based on the risk diffusion trajectory, the risk probability distribution of each node in the enterprise transaction cycle chain is calculated to obtain the risk probability of the transaction node in the cycle chain;
[0047] Define the duration of risk diffusion simulation;
[0048] The risk probability of the transaction nodes in the cycle chain is assigned to the node probability of the enterprise transaction cycle chain, and an iterative risk diffusion simulation calculation is performed according to the risk diffusion simulation duration to identify potential risk infection nodes;
[0049] The potential risk infection nodes are divided into risk areas to obtain risk node clusters.
[0050] The present invention calculates the risk probability distribution of the enterprise transaction cycle chain node by node to determine the potential risk level of each node, which helps to understand the risk level of each transaction node. By defining the risk diffusion simulation duration and determining the time span of the risk diffusion simulation, it helps to set the time range of the risk diffusion analysis so as to simulate and evaluate the propagation process of potential risks. By assigning the risk probability of the cycle chain transaction nodes to each node of the enterprise transaction cycle chain and performing iterative risk diffusion simulation calculations to simulate the risk propagation process in the cycle chain, it helps to identify the nodes infected by the risk, that is, those that become the starting point or intermediate nodes of risk propagation in the risk diffusion simulation. By dividing the potential risk infected nodes into risk areas, nodes with similar risk characteristics are clustered into risk node clusters, which helps to classify and organize the risk nodes in the cycle chain, so that the enterprise can better understand the spatial distribution of risks, focus on high-risk areas in the cluster, and take corresponding risk management measures.
[0051] Preferably, the risk assessment module is specifically used for:
[0052] Use enterprise dynamic risk profiles to conduct personalized risk analysis on online payment information to obtain current payment potential risk data;
[0053] Classify potential risk data into risk types and mark potential risk types;
[0054] Conduct risk feature mining based on potential risk types to generate enterprise risk type feature data;
[0055] Quantitatively evaluate the enterprise risk type characteristic data to obtain the enterprise payment risk level.
[0056] The present invention uses the enterprise's dynamic risk profile to perform personalized risk analysis on online payment information, evaluates the current payment information according to the enterprise's actual situation, helps to identify potential risks in current payments, and provides early warning information on payment security. By classifying the potential risk data into risk types, different risks are classified and labeled to better understand and analyze different types of risks, which helps to identify various potential risks in payment, such as fraud risk, security vulnerability risk, etc., by mining risk characteristics based on potential risk types and analyzing the characteristics and patterns of different risk types, it helps to identify the common characteristics and laws of different types of risks, so as to better understand the nature and influencing factors of risks. By quantitatively evaluating the enterprise's risk type characteristic data and quantitatively analyzing and evaluating the enterprise's payment risk, it helps to convert risks into measurable indicators, thereby obtaining the enterprise's payment risk level. The payment risk level helps the enterprise understand the overall risk status of its payment activities.
[0057] Preferably, the payment control module is specifically used for:
[0058] Make differentiated risk management decisions for online payment information based on the enterprise payment risk level to generate enterprise risk management strategies;
[0059] Conduct data mining driven modeling of enterprise risk management strategies and build an enterprise risk payment management model.
[0060] The present invention makes differentiated risk management decisions on online payment information based on the enterprise's payment risk level, and takes corresponding management measures according to different payment risk levels. This helps enterprises to formulate targeted risk management strategies according to the severity of payment risks, so as to minimize the impact of potential risks on the enterprise. By performing data mining-driven modeling on the enterprise's risk management strategy, the enterprise's risk management experience and data are combined to establish a risk payment management model, which helps to systematically analyze and predict payment risks, provide decision support and early warning mechanisms, and help enterprises manage and control risks more effectively. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] Figure 1 This is a schematic diagram of an embodiment of an enterprise payment risk management system based on deep learning according to the present invention; DETAILED DESCRIPTION
[0062] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.
[0063] This application example provides an enterprise payment risk management system based on deep learning. The execution subject of the enterprise payment risk management system based on deep learning includes but is not limited to: mechanical equipment, data processing platform, cloud server node, network upload device, etc. equipped with the system can be regarded as the general computing node of this application, and the data processing platform includes but is not limited to: at least one of an audio and image management system, an information management system, and a cloud data management system.
[0064] See also Figure 1 As shown, the present invention provides an enterprise payment risk management system based on deep learning, and the enterprise payment risk management system based on deep learning includes the following steps:
[0065] The extreme value filtering module is used to: obtain the enterprise's historical payment data and online payment information; perform extreme value sample filtering on the enterprise's historical payment data to obtain optimized historical payment data; obtain a historical payment transaction feature sequence based on the optimized historical payment data;
[0066] The cycle chain module is used to: analyze the transaction flow direction of the historical payment transaction feature sequence to obtain transaction flow direction data; reconstruct the global transaction cycle based on the transaction flow direction data to build the enterprise transaction cycle chain;
[0067] The risk singularity module is used to: calculate the transaction frequency between enterprises on the enterprise transaction cycle chain to obtain frequent transaction nodes; perform transaction risk analysis on the frequent transaction nodes to obtain the risk singularity of the transaction cycle chain;
[0068] The risk situation evolution module is used to: identify the adjacent risk diffusion of the risk singularity of the transaction cycle chain to obtain the risk node cluster; perform risk situation evolution on the risk node cluster to construct a dynamic risk profile of the enterprise;
[0069] The risk assessment module is used to: use the enterprise dynamic risk profile to conduct personalized risk quantitative assessment on online payment information, so as to obtain the enterprise payment risk level;
[0070] The payment control module is used to make differentiated risk control decisions on online payment information based on the enterprise payment risk level and build an enterprise risk payment control model.
[0071] The present invention acquires the enterprise's historical payment data and online payment information to establish a comprehensive understanding of payment behavior, and eliminates abnormal data through extreme value sample filtering to obtain more accurate historical payment data. By extracting the characteristic sequence of historical payment transactions, the law and trend of payment behavior are revealed. By analyzing the transaction flow direction, the flow direction and path of payment transactions are understood, and the pattern of enterprise capital flow is revealed. The global transaction cycle reconstruction integrates and optimizes the transaction cycle, and establishes an enterprise transaction cycle chain to provide a basis for subsequent analysis and decision-making. By calculating the transaction frequency between enterprises, the nodes with frequent transactions, that is, the nodes with frequent transactions with other enterprises, are determined. Transaction risk analysis is performed on the nodes with frequent transactions to identify nodes with higher transaction risks, that is, risk singularities in the transaction cycle chain, to provide a basis for subsequent risk analysis and management. By expanding the adjacent risks Scattered identification is used to determine the risk node clusters adjacent to the risk singularity, that is, the node groups with a high correlation with the risk nodes. The risk situation evolution analysis of the risk node clusters is carried out to reveal the development trend and evolution law of risks, and a dynamic risk profile of the enterprise is constructed to provide a reference for risk management and response. By using the dynamic risk profile of the enterprise, a personalized risk quantitative assessment is conducted on online payment information to help enterprises accurately assess the risk level of each payment, and classify and sort it according to the risk level, so as to better understand and manage payment risks. By making differentiated risk management decisions based on the enterprise payment risk level, corresponding risk management strategies and control measures are adopted for payment information of different risk levels, and an enterprise risk payment management model is constructed to help enterprises establish a systematic risk management framework, effectively reduce payment risks and improve payment security.
[0072] In the embodiment of the present invention, refer to Figure 1 , is a schematic diagram of an embodiment of an enterprise payment risk management system based on deep learning of the present invention. In this example, the enterprise payment risk management system based on deep learning specifically includes:
[0073] The extreme value filtering module is used to: obtain the enterprise's historical payment data and online payment information; perform extreme value sample filtering on the enterprise's historical payment data to obtain optimized historical payment data; obtain a historical payment transaction feature sequence based on the optimized historical payment data;
[0074] In this embodiment, after obtaining the authorization of the enterprise, all payment transaction record data of the enterprise in the past 3-5 years are collected from channels such as the enterprise's financial system and bank reconciliation system. At the same time, the enterprise's recent online payment information, including payment time, payment amount, payment object, etc., is obtained. The collected historical payment data is cleaned and standardized to ensure that the data format is unified and complete. Outliers and duplicate data are identified and eliminated to ensure the accuracy and reliability of the data. Maximum and minimum samples in the historical payment data are identified and filtered to eliminate the interference of these abnormal data on subsequent analysis. A method based on the interquartile range is used to identify and eliminate samples that exceed 1.5 times the upper and lower interquartile range as outliers. Based on the optimized historical payment data, a series of characteristic indicators such as payment amount, payment frequency, and payment object type are extracted, and these characteristic indicators are organized into a time series payment transaction characteristic sequence to prepare data for subsequent transaction flow analysis.
[0075] The cycle chain module is used to: analyze the transaction flow direction of the historical payment transaction feature sequence to obtain transaction flow direction data; reconstruct the global transaction cycle based on the transaction flow direction data to build the enterprise transaction cycle chain;
[0076] In this embodiment, the flow direction of each transaction is analyzed, that is, the direction from the payer to the payee. Association rule mining, social network analysis and other methods are used to identify the correlation between the two parties to the transaction and the transaction flow direction. For each transaction, its flow information is recorded to form a transaction flow direction data set. The transaction flow direction data is aggregated and analyzed to identify the transaction chain structure within the enterprise. Through dimensions such as related transaction parties and transaction time, related payment transactions are connected in series to build a global transaction cycle chain for the enterprise. The transaction cycle chain reflects the overall direction of capital flow within the enterprise and helps to identify the core transaction links of the enterprise. The constructed transaction cycle chain is further analyzed to extract characteristic indicators such as cycle length, transaction frequency, and capital scale. These characteristic indicators can be used to describe the temporal dynamic characteristics of corporate transaction activities and provide valuable information for subsequent risk identification and prediction.
[0077] The risk singularity module is used to: calculate the transaction frequency between enterprises on the enterprise transaction cycle chain to obtain frequent transaction nodes; perform transaction risk analysis on the frequent transaction nodes to obtain the risk singularity of the transaction cycle chain;
[0078] In this embodiment, the transaction frequency is calculated for each transaction node (i.e., both parties to the transaction). The transaction frequency is calculated by counting the number of transactions of the node within a certain time window, which reflects the transaction activity of the node. The transaction frequencies of all transaction nodes are sorted to identify the "frequent transaction nodes" with the highest transaction frequency. The frequent transaction nodes are further analyzed, and their potential transaction risks are evaluated in combination with other characteristic indicators of the nodes (such as transaction amount, transaction cycle, etc.). Anomaly detection, machine learning and other methods are used to identify those transaction cycle chain nodes with abnormal performance and higher risks, namely "transaction risk singularities". Transaction risk singularities involve malicious transactions, cashing out, abnormal fund flows, etc., which require special attention and control. Based on the analysis of frequent transaction nodes and risk singularities, a series of indicators reflecting transaction risks are extracted, such as transaction frequency index, abnormal transaction ratio, etc. These indicators can be used to quantitatively evaluate the overall transaction risk level of the enterprise and provide support for subsequent risk warning and decision-making.
[0079] The risk situation evolution module is used to: identify the adjacent risk diffusion of the risk singularity of the transaction cycle chain to obtain the risk node cluster; perform risk situation evolution on the risk node cluster to construct a dynamic risk profile of the enterprise;
[0080] In this embodiment, starting from the transaction risk singularity, the correlation between it and the surrounding nodes is analyzed, the adjacent diffusion mechanism of risk is explored, and graph analysis, association rule mining and other methods are used to identify other transaction nodes directly or indirectly related to the risk singularity to form a "risk node cluster". The risk node cluster reflects the propagation trend of risk in the transaction cycle chain, which helps to comprehensively identify the risk surface faced by the enterprise. The risk node cluster is dynamically tracked and analyzed to observe its changing trend in the time series. Combined with the transaction characteristic indicators of the node, such as transaction frequency, transaction amount, transaction cycle, etc., the evolution trend of the risk node cluster is analyzed, the changing law of the risk node cluster is identified, the overall risk situation faced by the enterprise is constructed, and the dynamic risk profile of the enterprise is constructed, including risk node clusters, risk propagation paths, risk evolution trends and other dimensions. The risk profile reflects the overall risk status of the enterprise's internal transaction activities, provides support for subsequent risk warning and control, regularly updates the risk profile, tracks the dynamic changes of the enterprise's risk status, and provides data support for decision-making.
[0081] The risk assessment module is used to: use the enterprise dynamic risk profile to conduct personalized risk quantitative assessment on online payment information, so as to obtain the enterprise payment risk level;
[0082] In this embodiment, a payment risk quantification model is designed based on key indicators in the enterprise risk profile, such as risk node clusters, risk propagation paths, risk evolution trends, etc., and combined with the specific characteristics of each payment transaction, such as the payment object, payment amount, payment time, etc., it is correlated with the overall risk status of the enterprise and analyzed to output a risk score or level for each payment transaction, reflecting the relative position of the payment behavior in the overall risk landscape of the enterprise.
[0083] The payment control module is used to make differentiated risk control decisions on online payment information based on the enterprise payment risk level and build an enterprise risk payment control model.
[0084] In this embodiment, payment transactions are divided into different risk levels, using a multi-level classification, such as low risk, medium risk and high risk. The appropriate classification standard is determined according to the actual situation. The risk level classification should match the risk preference and tolerance of the enterprise to ensure the pertinence and effectiveness of risk management measures. Differentiated risk management strategies and measures are formulated for payment transactions of different risk levels. For low-risk payments, relatively loose management measures such as automated approval and credit limit monitoring are adopted. For medium-risk payments, manual review, secondary confirmation and other management measures are added to strengthen monitoring. For high-risk payments, more stringent approval processes, real-time warnings, transaction limits and other management measures are adopted. The above-mentioned differentiated management strategies and measures are integrated into a systematic payment risk management model. The model should include multi-dimensional management rules and decision-making logic, such as risk level judgment, approval process, monitoring indicators, etc. Through dynamic adjustment and optimization of the model, the accuracy and effectiveness of enterprise payment risk management are continuously improved.
[0085] In this embodiment, the extreme value filtering module is specifically used for:
[0086] Obtain the company's historical payment data and online payment information;
[0087] Perform extreme value sample filtering on the enterprise's historical payment data to obtain optimized historical payment data;
[0088] Extracting transaction features from the optimized historical payment data to obtain historical payment transaction feature data;
[0089] Heterogeneous data fusion processing is performed on historical payment transaction feature data to obtain a historical payment transaction feature sequence.
[0090] In this embodiment, the enterprise's historical payment data is collected, including previous payment records, transaction information and related financial data. These data come from the enterprise's internal database, financial system or other related systems. Online payment information is obtained, such as the enterprise's online payment records and transaction information on the e-commerce platform or third-party payment platform. This information is obtained through API interfaces or data export. Outlier detection and extreme value sample filtering are performed on the enterprise's historical payment data. Statistical methods, data mining techniques or machine learning algorithms are used to identify and filter abnormal payment records. Outliers and extreme samples are eliminated to obtain an optimized historical payment data set. In this way, noise and abnormalities in the data are removed, and the accuracy and reliability of subsequent analysis are improved. Transaction features are extracted from the optimized historical payment data. Common transaction features include transaction amount, transaction type (such as purchase, refund, transfer, etc.), transaction counterparty (such as the payer and payee of the transaction), transaction business node (such as order creation, payment completion, etc.), payment path (such as payment channel, payment method, etc.). Using data processing and feature extraction methods, the historical payment data is converted into a structured historical payment transaction feature data set. Each payment transaction record will be represented as a data point containing the above features. Perform heterogeneous data fusion on historical payment transaction feature data. This involves integrating data from different data sources or different formats to form a unified data sequence. According to the chronological order, the historical payment transaction feature data is sorted and organized in the order in which the transactions occurred to form a historical payment transaction feature sequence. Optionally, further data preprocessing is performed on the historical payment transaction feature sequence, such as data smoothing, missing value processing or standardization, etc., to facilitate subsequent analysis and modeling.
[0091] In this embodiment, the specific steps of performing extreme value sample filtering on the enterprise historical payment data to obtain optimized historical payment data are:
[0092] Identify missing values in the company's historical payment data to obtain missing payment values;
[0093] Perform proportional statistics on missing values of payments to generate missing value ratios;
[0094] Based on the missing value ratio, the payment missing values are classified by random correlation, the randomness of the missing values is determined, and random missing values and non-random missing values are generated;
[0095] Perform mean filling processing on random missing values to obtain mean-filled payment data;
[0096] Performing outlier evaluation on non-random missing values based on a preset normative data sample, and determining that the non-random missing value is an abnormal extreme value sample when the non-random missing value exceeds the preset normative data sample range;
[0097] Perform extreme value sample filtering on abnormal extreme value samples to obtain extreme value filtered payment data;
[0098] Optimized historical payment data is obtained based on mean filling payment data and extreme value filtering payment data.
[0099] In this embodiment, the historical payment data of the enterprise is traversed, all records with missing values are identified, the total number of missing values is counted, and the proportion of missing values in the entire data set is calculated to obtain the missing value ratio. For example, a payment data set of an enterprise has 1000 records, of which 200 records have missing values, and the missing value ratio is 20%. According to the missing value ratio, a random mechanism is used to classify the missing values to determine whether they are randomly missing. When the missing value ratio is low (such as less than 10%), it is determined to be completely missing at random (MCAR). When the missing value ratio is high (such as more than 20%), it is necessary to further analyze the distribution characteristics of the missing values to determine whether they are non-random missing (NMAR). When the missing values of some payment records are concentrated in a specific transaction type or transaction amount range, It is determined to be non-random missing. For data determined to be randomly missing, the mean filling method is used to fill the missing values. Specifically, the missing values are replaced by the average value of the attribute. For example, if the average value of a payment attribute is 1,000 yuan, all missing values are replaced with 1,000 yuan. For data determined to be non-random missing, further outlier analysis and filtering are required. A standardized data sample needs to be set as a reference for judging outliers. The non-random missing values are compared with the standardized sample. Any data that exceeds the range of the standardized sample is identified as an abnormal extreme value sample. Finally, these abnormal extreme value samples are filtered out from the data set to obtain payment data after outlier processing. After the preprocessing steps above, an optimized payment data set after missing value filling and outlier filtering is finally obtained.
[0100] In this embodiment, the periodic chain module is specifically used for:
[0101] Identify transaction nodes of historical payment transaction feature sequences and mark node timestamps to obtain transaction feature nodes;
[0102] Analyze the transaction flow direction of the transaction feature nodes to obtain transaction flow direction data;
[0103] Based on the transaction flow direction data, the transaction path is traced back to obtain the transaction path;
[0104] Reconstruct the global transaction cycle of the transaction path and build the enterprise transaction cycle chain.
[0105] In this embodiment, transaction node identification is performed on the historical payment transaction feature sequence. According to predefined rules or algorithms, the historical payment transaction feature sequence is analyzed and identified to determine the location of the transaction node, mark the timestamp of the transaction feature node, assign a timestamp to each transaction feature node, record the time when the transaction node occurs, analyze the flow relationship between the transaction feature nodes, and infer the flow direction between the transaction feature nodes by comparing the timestamps and other relevant information of adjacent transaction feature nodes. According to the analysis results, the transaction feature nodes are connected to form transaction flow direction data. The flow relationship between the transaction nodes is represented by a graph theory method, such as a directed graph or a flow chart. According to the transaction flow direction, Directed data, starting from the initial node to trace the path, in the directed graph or flow chart, by traversing the connection relationship between nodes and edges, find all transaction paths, record the order and node information of each transaction path, use lists, trees or other data structures to store the transaction path information, comprehensively consider the time sequence and node information of the transaction path, reconstruct the global transaction cycle, and choose different methods to define the transaction cycle according to different business needs and analysis purposes, such as by time range, by business process, etc., to build an enterprise transaction cycle chain, connect the transaction paths in the order of the transaction cycle to form an enterprise transaction cycle chain, each transaction cycle chain represents a complete transaction cycle, including multiple transaction paths.
[0106] In this embodiment, the risk singularity module is specifically used for:
[0107] Calculate the transaction frequency between enterprises in the enterprise transaction cycle chain and generate transaction frequency data;
[0108] Based on the transaction frequency data, frequent nodes are extracted from the transaction feature nodes to obtain frequent transaction nodes;
[0109] Conduct transaction risk analysis on nodes with frequent transactions and generate transaction risk data;
[0110] Based on the potential transaction risk data, risk vulnerability analysis is conducted on the enterprise transaction cycle chain to obtain the risk singularity of the transaction cycle chain.
[0111] In this embodiment, based on the enterprise transaction cycle chain constructed previously, the transaction frequency between each enterprise is counted. For example, the transaction frequency between enterprise A and supplier B is twice a month, while the transaction frequency with customer C is once a quarter. In this way, complete transaction frequency data between enterprises is obtained to reflect the transaction activity between each enterprise. According to the transaction frequency data calculated in the previous step, nodes with higher transaction frequency, i.e., frequent transaction nodes, are identified. For example, supplier B is a frequent transaction node, and its transaction frequency with enterprise A is higher. These frequent nodes usually represent the key transaction links and supply chain nodes of the enterprise. For the above-mentioned frequent transaction nodes, further The first step is to analyze its potential transaction risks, consider factors such as overdue payments, supplier credit risk, transaction disputes, etc., and evaluate the risk level of each node. For example, supplier B has a certain credit risk, and its transaction with enterprise A has a certain default risk. Combining the transaction cycle chain, transaction frequency and transaction risk data obtained in the previous steps, a comprehensive risk vulnerability analysis of the entire transaction chain is conducted to find the risk "singularity" on the transaction cycle chain, that is, the key nodes and links that cause failures or interruptions in the entire transaction chain. For example, supplier B is a node with frequent transactions and higher risks. If there is a problem with it, it will seriously affect the overall transaction cycle of enterprise A.
[0112] In this embodiment, the risk situation evolution module is specifically used to:
[0113] Identify the adjacent risk diffusion of the risk singularity of the transaction cycle chain and generate the risk diffusion trajectory;
[0114] Based on the risk diffusion trajectory, the risk probability distribution of the enterprise transaction cycle chain is calculated to obtain the risk node cluster;
[0115] Evolve the risk situation of risk node clusters and generate cycle chain risk situation evolution data;
[0116] Conduct dynamic risk prediction on the evolution data of cycle chain risk trends and build a dynamic risk profile of the enterprise.
[0117] In this embodiment, for the risk singularity in the transaction cycle chain, that is, the potential abnormal or high-risk node, adjacent risk diffusion identification is performed by analyzing the risk transfer of adjacent transaction paths. According to the risk transfer rule or algorithm, the probability or degree of risk diffusion from the risk singularity to the adjacent transaction path is determined, and a risk diffusion trajectory is generated. The transmission path and diffusion degree of the risk from the risk singularity to the adjacent path are recorded. Based on the risk diffusion trajectory, the risk probability distribution of each node in the enterprise transaction cycle chain is calculated, and the risk probability of the node is calculated using a statistical method, a probability model or a machine learning algorithm. According to the risk probability of the node, the nodes with similar risk levels are clustered into risk node clusters, and a clustering algorithm or a threshold is used to calculate the risk probability of the node. A division method is used to determine the formation of node clusters, and risk situation evolution analysis is performed on risk node clusters, which involves monitoring and analyzing the emergence and evolution of risk node clusters at different time points, and recording the cycle chain risk situation evolution data, including the changing trend of risk node clusters, quantity changes, and risk level changes. Based on the cycle chain risk situation evolution data, dynamic risk forecasting is performed, and time series analysis, machine learning models or other forecasting methods are used to predict the future development trend of risk node clusters, and a dynamic risk profile of the enterprise is constructed. The forecast results are integrated with other key information of the enterprise, including the risk level, development trend, scope of impact, and recommended risk response measures of risk node clusters.
[0118] In this embodiment, the specific steps of calculating the risk probability distribution of the enterprise transaction cycle chain based on the risk diffusion trajectory to obtain the risk node cluster are:
[0119] Based on the risk diffusion trajectory, the risk probability distribution of each node in the enterprise transaction cycle chain is calculated to obtain the risk probability of the transaction node in the cycle chain;
[0120] Define the duration of risk diffusion simulation;
[0121] The risk probability of the transaction nodes in the cycle chain is assigned to the node probability of the enterprise transaction cycle chain, and an iterative risk diffusion simulation calculation is performed according to the risk diffusion simulation duration to identify potential risk infection nodes;
[0122] The potential risk infection nodes are divided into risk areas to obtain risk node clusters.
[0123] In this embodiment, the probability distribution of risk occurrence is calculated for each transaction node. For example, the risk probability of supplier B is 0.2, which means that there is a 20% probability of a risk event. These node risk probabilities can reflect the potential risk level of each link in the transaction chain. An appropriate risk diffusion simulation duration is defined, that is, the length of time for simulating the risk to propagate in the transaction chain. This duration can be set according to actual conditions. For example, the risk diffusion process in the next 6 months is simulated. The setting of the simulation duration needs to consider the transaction cycle characteristics of the enterprise and the speed of risk propagation. The above transaction node risk probabilities are assigned to the entire transaction cycle chain. According to the set risk diffusion simulation duration, The risks on the transaction chain are iteratively calculated and diffused. This process simulates the risk propagation process and final impact between various transaction nodes. For example, the risk of supplier B will gradually spread to other transaction links of enterprise A through its transaction relationship with enterprise A. Through the aforementioned risk diffusion simulation, potential nodes that are "infected" by risks are identified, and these high-risk nodes are clustered and regionalized to obtain interrelated risk node clusters. These clusters represent risk "hot spots" that appear on the transaction chain and require special attention and control. For example, certain key transaction links of enterprise A will form a risk cluster, and targeted risk management measures need to be taken.
[0124] In this embodiment, the risk assessment module is specifically used to:
[0125] Use enterprise dynamic risk profiles to conduct personalized risk analysis on online payment information to obtain current payment potential risk data;
[0126] Classify potential risk data into risk types and mark potential risk types;
[0127] Conduct risk feature mining based on potential risk types to generate enterprise risk type feature data;
[0128] Quantitatively evaluate the enterprise risk type characteristic data to obtain the enterprise payment risk level.
[0129] In this embodiment, personalized risk analysis is performed in combination with the current payment information. This is achieved by comparing the current payment information with the risk characteristics and patterns in the enterprise's dynamic risk profile. Potential risk data in the current payment information, including abnormal amounts, abnormal times, abnormal payment methods, etc., is identified and recorded. Risk types are classified according to the characteristics and patterns of the potential risk data. Classification is performed based on a predefined risk type list or using a machine learning algorithm. Each potential risk data is labeled to determine the risk type to which it belongs, which is helpful for subsequent risk feature mining and quantitative evaluation. According to the potential risk type, risk feature mining is performed for each risk type, including analyzing the payment amount, Payment time, payment frequency and other characteristics related to the risk type, generate enterprise risk type characteristic data, record the risk characteristics of each risk type, use data tables, feature vectors or other data structures to represent the risk type and its corresponding characteristic data, design risk assessment indicators and assessment methods, conduct quantitative assessments based on the enterprise risk type characteristic data, including establishing a scoring model, formulating risk weights, etc., evaluate the enterprise risk type characteristic data, calculate the risk score or risk level of each risk type, and comprehensively evaluate the assessment results of each risk type to obtain the enterprise's payment risk level, which is divided according to pre-set rules or standards, such as low risk, medium risk, and high risk levels.
[0130] In this embodiment, the payment control module is specifically used to:
[0131] Make differentiated risk management decisions for online payment information based on the enterprise payment risk level to generate enterprise risk management strategies;
[0132] Conduct data mining driven modeling of enterprise risk management strategies and build an enterprise risk payment management model.
[0133] In this embodiment, differentiated risk management decisions are made for online payment information of different levels according to the payment risk level of the enterprise. This means that different risk management measures are taken for payment behaviors of different risk levels, and risk management strategies corresponding to different risk levels are designed and formulated, including risk warning mechanisms, risk monitoring measures, payment limit settings, identity verification requirements, etc. According to the payment risk level and the corresponding risk management strategy, a risk management strategy document or guide for the enterprise is generated to guide the actual risk management work. Appropriate data mining techniques and algorithms are used to explore potential patterns, laws and correlations in the data. Based on the results of data mining, an enterprise risk payment management model is established. It is a prediction model, classification model or other suitable model used to assist enterprises in making risk management decisions in actual operations. The established risk payment management model is verified and evaluated to ensure its accuracy and reliability. The constructed enterprise risk payment management model is applied to actual risk management decisions, and risk management and control measures are optimized according to the output results of the model.
[0134] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.
[0135] The above is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.
Claims
1. An enterprise payment risk management system based on deep learning, characterized in that: The enterprise payment risk management system based on deep learning includes an extreme value filtering module, a cycle chain module, a risk singularity module, a risk situation evolution module, a risk assessment module and a payment control module; The extreme value filtering module is used to: obtain the enterprise's historical payment data and online payment information; perform extreme value sample filtering on the enterprise's historical payment data to obtain optimized historical payment data; obtain a historical payment transaction feature sequence based on the optimized historical payment data; The cycle chain module is used to: analyze the transaction flow direction of the historical payment transaction feature sequence to obtain transaction flow direction data; reconstruct the global transaction cycle based on the transaction flow direction data to build the enterprise transaction cycle chain; The risk singularity module is used to: Calculate the transaction frequency between enterprises in the enterprise transaction cycle chain and generate transaction frequency data; Based on the transaction frequency data, frequent nodes are extracted from the transaction feature nodes to obtain frequent transaction nodes; Conduct transaction risk analysis on nodes with frequent transactions and generate transaction risk data; Based on the potential transaction risk data, risk vulnerability analysis is performed on the enterprise transaction cycle chain to obtain the risk singularity of the transaction cycle chain; The risk situation evolution module is used to: Identify the adjacent risk diffusion of the risk singularity of the transaction cycle chain and generate the risk diffusion trajectory; Based on the risk diffusion trajectory, the risk probability distribution of the enterprise transaction cycle chain is calculated to obtain the risk node cluster; Evolve the risk situation of risk node clusters and generate cycle chain risk situation evolution data; Conduct dynamic risk forecasting on the risk trend evolution data of the cycle chain and build a dynamic risk profile of the enterprise; The specific steps of calculating the risk probability distribution of the enterprise transaction cycle chain based on the risk diffusion trajectory to obtain the risk node cluster are as follows: Based on the risk diffusion trajectory, the risk probability distribution of each node in the enterprise transaction cycle chain is calculated to obtain the risk probability of the transaction node in the cycle chain; Define the duration of risk diffusion simulation; Assign node probabilities to the enterprise transaction cycle chain using the cycle chain transaction node risk probabilities, and perform iterative risk diffusion simulation calculations based on the risk diffusion simulation duration to identify potential risk infection nodes; Divide the potential risk infection nodes into risk areas to obtain risk node clusters; The risk assessment module is used to: Use enterprise dynamic risk profiles to conduct personalized risk analysis on online payment information to obtain current payment potential risk data; Classify potential risk data into risk types and mark potential risk types; Conduct risk feature mining based on potential risk types to generate enterprise risk type feature data; Quantitatively evaluate the enterprise risk type characteristic data to obtain the enterprise payment risk level; The payment control module is used to make differentiated risk control decisions on online payment information based on the enterprise payment risk level and build an enterprise risk payment control model.
2. The enterprise payment risk management system based on deep learning according to claim 1 is characterized in that: The extreme value filtering module is used to: obtain the enterprise's historical payment data and online payment information; perform extreme value sample filtering on the enterprise's historical payment data to obtain optimized historical payment data; obtain a historical payment transaction feature sequence based on the optimized historical payment data, specifically including: Obtain the company's historical payment data and online payment information; Perform extreme value sample filtering on the enterprise's historical payment data to obtain optimized historical payment data; Extracting transaction features from the optimized historical payment data to obtain historical payment transaction feature data; Heterogeneous data fusion processing is performed on historical payment transaction feature data to obtain a historical payment transaction feature sequence.
3. The enterprise payment risk management system based on deep learning according to claim 2 is characterized in that: The specific steps of filtering the extreme value samples of the enterprise's historical payment data to obtain optimized historical payment data are as follows: Identify missing values in the company's historical payment data to obtain missing payment values; Perform proportional statistics on missing values of payments to generate missing value ratios; Based on the missing value ratio, the payment missing values are classified by random correlation, the randomness of the missing values is determined, and random missing values and non-random missing values are generated; Perform mean filling processing on random missing values to obtain mean-filled payment data; Performing outlier evaluation on non-random missing values based on a preset normative data sample, and determining that the non-random missing value is an abnormal extreme value sample when the non-random missing value exceeds the preset normative data sample range; Perform extreme value sample filtering on abnormal extreme value samples to obtain extreme value filtered payment data; Optimized historical payment data is obtained based on mean filling payment data and extreme value filtering payment data.
4. The enterprise payment risk management system based on deep learning according to claim 1 is characterized in that: The cycle chain module is used to: analyze the transaction flow direction of the historical payment transaction feature sequence to obtain transaction flow direction data; reconstruct the global transaction cycle based on the transaction flow direction data to build the enterprise transaction cycle chain, specifically including: Identify transaction nodes of historical payment transaction feature sequences and mark node timestamps to obtain transaction feature nodes; Analyze the transaction flow direction of transaction feature nodes to obtain transaction flow direction data; Based on the transaction flow direction data, the transaction path is traced back to obtain the transaction path; Reconstruct the global transaction cycle of the transaction path and build the enterprise transaction cycle chain.
5. The enterprise payment risk management system based on deep learning according to claim 1 is characterized in that: The payment control module is used to make differentiated risk control decisions on online payment information based on the enterprise payment risk level and build an enterprise risk payment control model, which specifically includes: Make differentiated risk management decisions for online payment information based on the enterprise payment risk level to generate enterprise risk management strategies; Conduct data mining driven modeling of enterprise risk management strategies and build an enterprise risk payment management model.
