Transaction risk early warning control method and device based on risk measurement model

Abstract

The application provides a transaction risk early warning control method and device based on a risk measurement model, and relates to the technical field of computer data processing.The method comprises the following steps: acquiring associated party data and associated transaction data corresponding to the associated party; inputting the associated party data and the associated transaction data corresponding to the associated party into an associated party risk measurement model and an associated transaction risk measurement model respectively, and obtaining an associated party risk control threshold and an associated transaction risk control threshold through analysis; multiplying the associated party risk control threshold and the associated transaction risk control threshold corresponding to the associated party to determine a transaction risk control threshold; setting an early warning condition according to the transaction risk control threshold; and when the associated transaction data corresponding to the associated party meets the early warning condition, generating an early warning prompt and controlling the associated transaction according to the early warning prompt.

Description

Technical Field

[0001] The present invention relates to the technical field of computer data processing, and in particular to a transaction risk early warning control method and device based on a risk measurement model. Background Art

[0002] This section is intended to provide a background or context for embodiments of the present invention. No description herein is admitted to be prior art by virtue of its inclusion in this section.

[0003] In their daily operations and management, enterprises proactively verify and confirm related parties and incorporate them into their IT systems for management. Based on the condition that the counterparty is a registered related party, enterprises will extract related transactions from the transaction data retained in all business systems and also incorporate them into their IT systems for management. Extracted related transaction information includes: counterparty name, ID information, business type, transaction date, contract amount, transaction amount, transaction balance, and transaction price.

[0004] In existing technologies, management that requires monitoring related-party transactions primarily involves manual statistical analysis, following steps such as deriving related-party transaction details, screening transactions of target business types, and performing calculations and analysis to determine whether the conduct of various types of transactions complies with regulatory requirements. Existing manual statistical analysis methods have the following shortcomings: Processing is untimely and inaccurate. In actual application scenarios, manual processing may not immediately fulfill regulatory requirements for review and disclosure. It is also inefficient and prone to errors. All regulations of various regulatory agencies regarding related-party transaction management require separate statistical analysis for each item, which is a huge workload and involves multiple steps in the statistical process. Any errors will lead to deviations in the calculation results, affecting judgments and making it difficult to ensure the compliance of related-party transactions.

[0005] Since related-party transaction regulators involve multiple institutions, the scope of related parties and related-party transactions defined by each regulatory agency, as well as the management requirements for related-party transactions, are all different and complex. Manual statistical analysis requires differentiating between regulatory agencies, designing separate calculation rules for each regulation, and performing calculations separately. This is a huge workload and inefficient. The statistical process has many steps, and any errors will lead to deviations in the calculation results, affecting judgment. In addition, manual statistical analysis cannot be carried out daily or in real time. It can only be performed for the current node, which lacks timeliness and accuracy. When some special circumstances arise between two manual statistical analyses, such as a sharp increase in transaction size in a short period of time, which directly triggers the regulatory agency's regulations on fulfilling review and disclosure requirements, there will be a lack of timely early warning prompts, resulting in related-party transactions not complying with regulations and failing to implement regulatory requirements.

[0006] In summary, there is an urgent need for a technical solution that can overcome the above-mentioned defects, effectively analyze the risks existing in transactions, improve the accuracy of statistical analysis, and improve processing efficiency. Summary of the Invention

[0007] In order to solve the problems existing in the prior art, the present invention proposes a transaction risk early warning control method and device based on a risk measurement model.

[0008] In a first aspect of an embodiment of the present invention, a transaction risk early warning control method based on a risk measurement model is proposed, comprising:

[0009] Obtain related party data and related transaction data corresponding to the related parties;

[0010] The related-party data and the related-party transaction data corresponding to the related-party are respectively input into the related-party risk measurement model and the related-party transaction risk measurement model, and the related-party risk control threshold and the related-party transaction risk control threshold are obtained through analysis; wherein the related-party risk measurement model and the related-party transaction risk measurement model are pre-trained in the following manner:

[0011] For the related-party risk measurement model, the related-party characteristic data corresponding to the related-party is collected and the risk level corresponding to the related-party is marked. The marked sample set is divided into a training set and a test set. The machine learning model is trained using the training set. The machine learning model is mapped from the related-party to the risk level in combination with the risk level corresponding to the related-party. The model is tested using the test set to obtain the related-party risk measurement model. The related-party risk control threshold is determined based on the risk level corresponding to the related-party output by the model.

[0012] For the related-party transaction risk measurement model, characteristic data of related-party transactions are collected, and the corresponding risk control thresholds of the related-party transactions are marked. The marked sample set is divided into a training set and a test set. The machine learning model is trained using the training set. The machine learning model is mapped from related-party transactions to risk control thresholds in combination with the risk control thresholds corresponding to the related-party transactions. The model is then tested using the test set to obtain the related-party transaction risk measurement model.

[0013] Multiplying the related-party risk control threshold and the risk control threshold corresponding to the related-party transaction to determine the transaction risk control threshold;

[0014] An early warning condition is set according to the transaction risk control threshold. When the related-party transaction data corresponding to the related-party meets the early warning condition, an early warning prompt is generated and the related-party transaction is controlled according to the early warning prompt.

[0015] In a second aspect of an embodiment of the present invention, a transaction risk early warning control device based on a risk measurement model is proposed, comprising:

[0016] The data acquisition module is used to obtain the related party data and the related transaction data corresponding to the related parties;

[0017] The risk measurement analysis module is configured to input the related-party data and the related-party transaction data corresponding to the related-party into a related-party risk measurement model and a related-party transaction risk measurement model, respectively, and obtain the related-party risk control threshold and the related-party transaction risk control threshold through analysis; wherein the related-party risk measurement model and the related-party transaction risk measurement model are pre-trained in the following manner:

[0018] For the related-party risk measurement model, the related-party characteristic data corresponding to the related-party is collected and the risk level corresponding to the related-party is marked. The marked sample set is divided into a training set and a test set. The machine learning model is trained using the training set. The machine learning model is mapped from the related-party to the risk level in combination with the risk level corresponding to the related-party. The model is tested using the test set to obtain the related-party risk measurement model. The related-party risk control threshold is determined based on the risk level corresponding to the related-party output by the model.

[0019] For the related-party transaction risk measurement model, characteristic data of related-party transactions are collected, and the corresponding risk control thresholds of the related-party transactions are marked. The marked sample set is divided into a training set and a test set. The machine learning model is trained using the training set. The machine learning model is mapped from related-party transactions to risk control thresholds in combination with the risk control thresholds corresponding to the related-party transactions. The model is then tested using the test set to obtain the related-party transaction risk measurement model.

[0020] a risk control threshold determination module, configured to multiply the related party risk control threshold by the risk control threshold corresponding to the related transaction to determine a transaction risk control threshold;

[0021] The transaction warning module is used to set warning conditions according to the transaction risk control threshold, generate warning prompts when the related transaction data corresponding to the related parties meet the warning conditions, and control the related transactions according to the warning prompts.

[0022] In a third aspect of an embodiment of the present invention, a computer device is proposed, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a transaction risk early warning control method based on a risk measurement model is implemented.

[0023] In a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is proposed, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, a transaction risk early warning control method based on a risk measurement model is implemented.

[0024] In a fifth aspect of an embodiment of the present invention, a computer program product is proposed. The computer program product includes a computer program. When the computer program is executed by a processor, the computer program implements a transaction risk early warning control method based on a risk measurement model.

[0025] The transaction risk early warning control method and device based on the risk measurement model proposed in the present invention obtain related party data and related transaction data corresponding to the related parties; input the related party data and the related transaction data corresponding to the related parties into the related party risk measurement model and the related transaction risk measurement model respectively, and obtain the related party risk control threshold and the related transaction risk control threshold after analysis; wherein, the related party risk measurement model and the related transaction risk measurement model are pre-trained in the following manner: for the related party risk measurement model, the related party feature data corresponding to the related parties are collected, and the risk level corresponding to the related parties are marked, and the marked sample set is divided to obtain a training set and a test set; the machine learning model is trained using the training set, and the machine learning model is mapped from the related parties to the risk level in combination with the risk level corresponding to the related parties, and the test set is used for testing to obtain the related party risk measurement model, and the related party risk control threshold is determined according to the risk level corresponding to the related parties output by the model; for the related transaction risk measurement model, the related transaction feature data are collected, and the risk control threshold corresponding to the related transactions are marked, and the marked sample set is divided to obtain a training set and a test set; the machine learning model is trained using the training set, In combination with the risk control thresholds corresponding to the related-party transactions, the machine learning model is trained to map related-party transactions to risk control thresholds, and tested using the test set to obtain a related-party transaction risk measurement model. The related-party risk control threshold is multiplied by the risk control threshold corresponding to the related-party transaction to determine the transaction risk control threshold. Warning conditions are set based on the transaction risk control thresholds. When the related-party transaction data corresponding to the related-party meets the warning conditions, a warning prompt is generated, and the related-party transaction is controlled based on the warning prompt. The overall solution automatically completes warning analysis for related-party transactions that require monitoring, significantly reducing the workload of personnel in related-party transaction management positions, reducing statistical errors, avoiding statistical caliber errors caused by personnel handovers or replacements, improving control accuracy, and enhancing work efficiency. The present invention utilizes a related-party risk measurement model and a related-party transaction risk measurement model to distinguish different risk levels and set corresponding control thresholds based on quantitative indicators. By combining the two control thresholds, the actual control threshold for each related-party in each type of related-party transaction that requires monitoring can be individually determined based on actual risk conditions. Preemptive control of related-party transactions that require monitoring is achieved through automatic calculation and timely warning, ensuring early warning and timely response, reducing the risk of violations, and providing strong technical support for risk warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0027] Figure 1 The present invention is a flowchart of a transaction risk early warning control method based on a risk measurement model according to an embodiment of the present invention.

[0028] Figure 2 It is a schematic diagram of a specific process of information collection according to an embodiment of the present invention.

[0029] Figure 3 It is a schematic diagram of a specific process of preprocessing related party data according to an embodiment of the present invention.

[0030] Figure 4 The figure is a schematic diagram of the training process of the related party risk measurement model according to an embodiment of the present invention.

[0031] Figure 5 It is a schematic diagram of a specific process of preprocessing related party data according to an embodiment of the present invention.

[0032] Figure 6 The figure is a schematic diagram of the training process of the related-party transaction risk measurement model according to an embodiment of the present invention.

[0033] Figure 7 It is a specific flowchart of transaction risk warning and control according to an embodiment of the present invention.

[0034] Figure 8 It is a flowchart of an embodiment of the present invention.

[0035] Figure 9 2 is a schematic diagram of the architecture of a transaction risk early warning control device based on a risk measurement model according to an embodiment of the present invention.

[0036] Figure 10 2 is a schematic diagram of the architecture of a transaction risk early warning control device based on a risk measurement model according to another embodiment of the present invention.

[0037] Figure 11 It is a schematic diagram of the structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided solely to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. Rather, these embodiments are provided to make this disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.

[0039] Those skilled in the art will appreciate that the embodiments of the present invention may be implemented as a system, apparatus, device, method, or computer program product. Therefore, the present disclosure may be implemented in the following forms: entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or in a combination of hardware and software.

[0040] According to an embodiment of the present invention, a transaction risk early warning control method and device based on a risk measurement model are proposed, which relate to the field of computer data processing technology.

[0041] In the embodiments of the present invention, the following terms need to be explained:

[0042] Regulatory agency: The agency that promulgates rules and regulations related to the management of related-party transactions;

[0043] Regulatory rules: Specific provisions on the management of related-party transactions in the rules and regulations related to related-party transactions promulgated by regulatory agencies;

[0044] Related parties: natural persons, legal persons or unincorporated organizations that have an associated relationship with the enterprise as defined in the regulatory rules that the enterprise needs to comply with;

[0045] Related-party transactions refer to the transfer of resources, services, obligations or other benefits between an enterprise and its subsidiaries and related parties.

[0046] The principles and spirit of the present invention are explained in detail below with reference to several representative embodiments of the present invention.

[0047] Figure 1 FIG1 is a flow chart of a transaction risk early warning control method based on a risk measurement model according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0048] S101, obtaining related party data and related transaction data corresponding to the related parties;

[0049] S102, inputting the related party data and the related transaction data corresponding to the related parties into a related party risk measurement model and a related transaction risk measurement model respectively, and obtaining the related party risk control threshold and the related transaction risk control threshold through analysis;

[0050] The related-party risk measurement model and the related-party transaction risk measurement model are pre-trained in the following manner:

[0051] For the related-party risk measurement model, the related-party characteristic data corresponding to the related-party is collected and the risk level corresponding to the related-party is marked. The marked sample set is divided into a training set and a test set. The machine learning model is trained using the training set. The machine learning model is mapped from the related-party to the risk level in combination with the risk level corresponding to the related-party. The model is tested using the test set to obtain the related-party risk measurement model. The related-party risk control threshold is determined based on the risk level corresponding to the related-party output by the model.

[0052] For the related-party transaction risk measurement model, characteristic data of related-party transactions are collected, and the corresponding risk control thresholds of the related-party transactions are marked. The marked sample set is divided into a training set and a test set. The machine learning model is trained using the training set. The machine learning model is mapped from related-party transactions to risk control thresholds in combination with the risk control thresholds corresponding to the related-party transactions. The model is then tested using the test set to obtain the related-party transaction risk measurement model.

[0053] S103, multiplying the related-party risk control threshold and the risk control threshold corresponding to the related-party transaction to determine a transaction risk control threshold;

[0054] S104, setting a warning condition according to the transaction risk control threshold, generating a warning prompt when the related-party transaction data corresponding to the related-party meets the warning condition, and controlling the related-party transaction according to the warning prompt.

[0055] The present invention can automatically complete the early warning analysis of related transactions that need to be monitored, significantly reduce the workload of personnel in related transaction management positions, reduce statistical errors, avoid statistical caliber errors caused by personnel handover or replacement, improve control accuracy, and enhance work efficiency. The present invention adopts a related party risk measurement model and a related transaction risk measurement model, which can distinguish different risk levels and set corresponding control thresholds with quantitative indicators. Through the combination of the two control thresholds, the actual control threshold of each related party in various types of related transactions that need to be monitored can be determined in a personalized manner according to the actual risk situation. Through automatic calculation and timely early warning, advance control of related transactions that need to be monitored is achieved, ensuring early warning, timely response, reducing violation risks, and providing strong technical support for risk warning.

[0056] In order to explain the above-mentioned transaction risk early warning and control method based on the risk measurement model more clearly, each step is explained in detail below.

[0057] In one embodiment, reference Figure 2, which is a schematic diagram of a specific process of information collection according to an embodiment of the present invention, such as Figure 2 As shown, the method includes:

[0058] S201, collect the list of related party customers;

[0059] S202: Collecting related party information and related transaction information corresponding to the related parties based on the related party customer list.

[0060] Enterprises build IT systems to collect lists of related-party clients and proactively identify related-party transactions based on the condition that the counterparty is a registered related party. These transactions are also managed within the IT system. Based on this information, according to the rules governing related-party transactions, combined with the enterprise's registered related parties and related-party transaction information, machine learning methods can be used to establish related-party transaction risk measurement models and related-party risk measurement models, respectively. This allows for effective analysis, early warning, and control of transaction risks.

[0061] Specifically, enterprises establish information technology systems capable of storing and managing detailed information on individuals and organizations with whom they have affiliations, processing and storing various related-party data, such as transaction data and business relationships. Through these IT systems, enterprises collect information on all related parties and create a comprehensive customer list. This list includes all natural persons, legal entities, and unincorporated organizations with direct or indirect relationships with the enterprise. Transaction counterparties are identified as related parties according to regulatory rules. Any transactions with these related parties are considered related-party transactions. Once related-party transactions are identified, their transaction information must also be managed within the IT system. This management includes recording transaction details, such as counterparty name, identification information, business type, transaction date, and contract amount. Regular updates to related-party and related-party transaction data ensure accuracy and timeliness. Data maintenance includes cleansing, deduplication, updating, and archiving. The IT system monitors related-party transactions to ensure compliance with regulatory requirements. It also automatically detects potential compliance issues and prompts management intervention.

[0062] In one embodiment, reference Figure 3 , which is a schematic diagram of a specific process of preprocessing related party data according to an embodiment of the present invention, such as Figure 3 As shown, the method includes:

[0063] S301: Based on the regulatory agency's transaction supervision requirements, collect historical transaction information of related parties under different calibers, including at least transaction scale and frequency, and establish a related party data mart;

[0064] S302, filtering and cleaning related parties, wherein related party data with related transactions is extracted from historical transaction information, invalid and redundant data is eliminated, and the cleaned related party data is unified and standardized;

[0065] S303: Determine a related party risk measurement score based on the related party data mart using multiple dimensions, including the number of related party enterprises, the frequency of transactions, the number of related party transactions with a maximum single amount exceeding a set limit within a set time period, the probability of related party transactions of a preset business type occurring, and the fair value of related party transactions within the set time period.

[0066] S304: A plurality of risk levels corresponding to the related parties are obtained according to the related party risk measurement scores, and a corresponding related party risk control threshold is set for each risk level.

[0067] Specifically, for the related party risk measurement model, the related party characteristic data corresponding to the related parties are collected, and the risk levels corresponding to the related parties are marked, and the marked sample set is divided to obtain a training set and a test set; the training set is used to train the machine learning model, and combined with the risk levels corresponding to the related parties, the machine learning model is mapped from related parties to risk levels, and the test set is used to test to obtain the related party risk measurement model, and the related party risk control threshold is determined according to the risk levels corresponding to the related parties output by the model.

[0068] refer to Figure 4 The specific process includes:

[0069] S401: For the related-party risk measurement model, collect related-party feature data corresponding to the related-party. Feature extraction and analysis are performed based on business dimension feature values, including related-party role, business type, transaction scale, and transaction frequency, to determine the related-party feature data as a sample set for model training.

[0070] S402, using the Bayesian algorithm to perform model training, constructing a classification prediction of the risk level corresponding to the related party and the related party characteristic data, and obtaining a related party risk measurement model;

[0071] S403, testing the related party risk measurement model, automatically correcting the characteristic dimensions of the related parties according to the test results, optimizing the related party characteristic data, and manually adjusting the model parameters.

[0072] Specifically, according to the related-party transaction supervision rules of each regulatory agency, for related parties under the corresponding scope, basic data of related parties shall be established based on the specific circumstances of related-party transactions that need to be monitored in previous years, including transaction scale, frequency, etc.

[0073] Cleanse the data of related parties (extract the data of related parties with related transactions within the historical sample period and eliminate invalid and redundant data). Then unify and standardize the cleaned related party data.

[0074] Relying on the data mart, the risk measurement score K of the related party is automatically calculated from five dimensions: the number of companies under the group where the related party is located, the frequency of past transactions, the number of related transactions with the highest single amount exceeding the set amount (for example, 200 million yuan) in the past 12 months, the possibility of major related transactions in the group where the related party is located, and the fairness of related transactions in the past 12 months. This score can be manually queried and reviewed, and can be manually adjusted according to actual conditions. According to the size of the score, the related parties are managed with graded labels. For example, they are divided into 3 levels, namely: risk level 1-low risk, risk level 2-medium risk, and risk level 3-high risk. Further, according to different risk levels, the control threshold Y2 of the related parties is determined respectively.

[0075] Establish business dimension feature values ​​such as related party roles (such as counterparties, issuers, underwriters, etc.), business types (credit, services, interest transfer, etc.), transaction scale (amount), transaction frequency (number of transactions within 1 year), etc., conduct feature extraction and analysis, and build a correlation matrix of dimension features to prepare for subsequent related party risk score model training.

[0076] Use the Bayesian algorithm to train the model, build a classification prediction of the related party risk level and the related party's dimensional feature classification in the previous section, obtain the risk level classification result, and determine the related party control threshold Y2. At the same time, according to the manual settings and verification results, automatically correct the characteristic dimensions of the related party (for example, divide the transaction frequency into monthly periods) to optimize the related party risk measurement parameter characteristics. Strengthen human-machine training, and intervene in manual adjustments based on machine learning. When the existing hierarchical management system cannot meet the requirements of refined management, the risk level can be further subdivided into n levels. Refer to Table 1, which exemplifies the correspondence between the interval of the related party risk measurement score, the related party risk level, and the related party control threshold.

[0077] Table 1

[0078] Related party risk measurement score range Related party risk level Related Party Control Threshold k1= <K<=100 Risk Level 1 Y2-1 k2=<K<k1 Risk Level 2 Y2-2 … … … k9= <K<k8 Risk Level 9 Y2-9 0=<K<k9 Risk Level 10 Y2-10

[0079] The risk levels and thresholds of related parties are recorded separately by regulatory agency, as shown in Table 2, which exemplifies the corresponding relationship.

[0080] Table 2

[0081]

[0082]

[0083] In one embodiment, reference Figure 5 , which is a schematic diagram of a specific process of preprocessing related party data according to an embodiment of the present invention, such as Figure 5 As shown, the method includes:

[0084] S501: Determine the transaction type, amount, business type to be included in the calculation, and calculation mechanism for related-party transactions based on the regulatory agency's transaction supervision requirements;

[0085] S502: Analyze historical transaction data based on the calculation mechanism, taking into account the transaction type, amount, and business type to be included in the calculation, to determine the transaction scale and increase / decrease of related transactions, and establish a related transaction data mart;

[0086] S503, filtering and cleaning the business data of related-party transactions, including eliminating invalid and redundant data, converting, integrating, discretizing, and normalizing the data;

[0087] S504: Determine the risk control threshold corresponding to the related-party transaction based on the related-party transaction data mart and multiple dimensions including the related-party role, business type, transaction size, risk limit, and increase or decrease range.

[0088] Specifically, for the related-party transaction risk measurement model, characteristic data of related-party transactions are collected, and the risk control thresholds corresponding to the related-party transactions are marked. The marked sample set is divided to obtain a training set and a test set. The machine learning model is trained using the training set, and the mapping from related-party transactions to risk control thresholds is learned for the machine learning model in combination with the risk control thresholds corresponding to the related-party transactions. The model is then tested using the test set to obtain the related-party transaction risk measurement model.

[0089] refer to Figure 6 The specific process includes:

[0090] S601: For the related-party transaction risk measurement model, collect related-party transaction feature data corresponding to the related-party transactions. Feature extraction and analysis are performed based on the business dimension feature values, including the related-party role, business type, transaction size, risk limit, and increase / decrease range, to determine the related-party transaction feature data as a sample set for model training.

[0091] S602: Perform model training using a support vector machine algorithm, use a Gaussian kernel as the kernel function, optimize the model accuracy and generalization ability, and evaluate the model using the F1 score.

[0092] Specifically, according to the related-party transaction supervision rules of each regulatory agency, the types of related-party transactions that need to be monitored, the corresponding amounts, the types of businesses that need to be included in the calculation under the types of related-party transactions that need to be monitored, and the calculation rules should be clearly defined.

[0093] Based on the calculation rules corresponding to various regulatory provisions, combined with the specific circumstances of transactions in previous years for all business types under the related-party transaction types that financial institutions need to monitor, including basic data dimensions such as transaction scale and increase or decrease, as well as external financial information such as industrial and commercial bureaus and the China Banking and Insurance Regulatory Commission, a related-party transaction risk data mart is established based on the big data cloud platform.

[0094] Based on the related-party transaction risk data mart, the related-party list and related-party transaction business type parameters established by the enterprise in the IT system, business data is filtered and cleaned (invalid and redundant data are eliminated) to obtain related-party transaction data that meets business needs.

[0095] Transform and integrate data to ensure data quality, accuracy, and reliability; discretize and reduce the integrated data (according to business management requirements, such as discretizing by equal distance through time intervals and discretizing by equal frequency through the number of transactions to obtain streamlined and effective related transaction data), unify and standardize the basic data of related transactions, and improve the quality and availability of related transaction data.

[0096] Extract and select related-party transaction features, establish business dimension feature values ​​such as related-party roles, business types, transaction sizes, risk limits, and increases and decreases, reduce the dataset dimensions, and lower the computational complexity and memory consumption of subsequent model training, thus preparing for the entire feature engineering and model training of subsequent related-party transactions.

[0097] Because support vector machines (SVMs) are particularly advantageous in processing high-dimensional and nonlinear data, and the characteristics of related-party transaction data are often nonlinear, this paper uses an SVM algorithm for model training. Since linear kernel functions are not suitable for processing nonlinear data, and the Sigmoid kernel function is slightly less effective, this paper uses a Gaussian kernel as the kernel function to optimize the model's accuracy and generalization ability. The model is then evaluated using the F1 score as an evaluation metric.

[0098] The overall related-party transaction SVM strategy backtest is as follows:

[0099] 1. Trading backtesting settings;

[0100] 1.1. Types and amounts of related-party transactions that require monitoring, as well as control thresholds for related-party transaction types;

[0101] 1.2. Backtest period: 20220101-20241231;

[0102] 1.3. Backtest frequency: 1 month;

[0103] 2. Select transaction type;

[0104] Select four transaction types from the related-party transaction system: credit transactions, service transactions, interest transfer transactions, and other transactions for testing;

[0105] 3. Trading strategy signals;

[0106] Feature selection: Monthly related-party transactions are summarized as features to calculate the control threshold corresponding to the type of related-party transaction business;

[0107] 4. Transaction model evaluation;

[0108] Model evaluation method: F1 score;

[0109] In actual application scenarios, it is best not to miss any real related-party transaction type. Considering that missing one may lead to failure to meet regulatory requirements, and at the same time, the accuracy is low, resulting in normal business transactions being incorrectly marked. F1 is the harmonic mean of precision and recall, which can effectively balance the relationship between the two. Therefore, the present invention adopts the performance of the F1 rating model. The F1 score is an evaluation indicator used to measure the performance of the model in classification problems; it comprehensively considers the model's precision (Precision) and recall (Recall), and provides a single value to evaluate the accuracy of the model. The F1 score is defined as the harmonic mean of precision and recall, and the calculation formula is:

[0110]

[0111] Precision: The ratio of positive categories to positive categories, calculated as follows: TP stands for True Positives and FP stands for False Positives.

[0112] Recall: The ratio of positive categories to actual positive categories. The calculation formula is: FN stands for False Negatives.

[0113] The F1 score achieves a balanced precision and recall, making it suitable for scenarios with an imbalance of positive and negative samples. In such cases, using precision or recall alone may not be sufficient to fully evaluate model performance. In multi-classification problems, the F1 score can be calculated for each class and then averaged to provide an overall assessment of the model's performance. During model training, the F1 score can be used as an evaluation metric to help select optimal model parameters and architecture. By tuning the model to maximize the F1 score, one can find a model that strikes the best balance between precision and recall. The requirements for precision and recall may vary in different business scenarios. For example, in some scenarios, avoiding misclassification of positive samples as negative (i.e., improving recall) may be more important than accurately identifying positive samples (i.e., improving precision). By adjusting the F1 score calculation, these varying business requirements can be accommodated. The F1 score provides an intuitive numerical value that allows practitioners to directly understand model performance.

[0114] Based on the related-party transaction risk measurement model, the corresponding control threshold Y1 is automatically generated. Calculate and record it separately according to the regulatory agency. An example is shown in Table 3.

[0115] Table 3

[0116]

[0117] In one embodiment, for S103, the related party risk control threshold and the risk control threshold corresponding to the related transaction are multiplied to determine the transaction risk control threshold.

[0118] Based on the results of the related-party transaction risk measurement model and the related-party risk measurement model, the actual control threshold Y3 for monitoring related-party transactions is calculated by multiplying the related-party control threshold Y1 by the related-party transaction control threshold Y2. Table 4 illustrates the corresponding relationship between the actual control thresholds for monitoring related-party transactions.

[0119] Table 4

[0120]

[0121] In one embodiment, for S104, a warning condition is set according to the transaction risk control threshold. When the related-party transaction data corresponding to the related-party satisfies the warning condition, a warning prompt is generated and the related-party transaction is controlled according to the warning prompt.

[0122] Specifically, refer to Figure 7 , is a schematic diagram of a specific process of transaction risk warning and control according to an embodiment of the present invention. Figure 7 As shown, the method includes:

[0123] S701, multiplying the amount of the related-party transaction to be monitored by the transaction risk control threshold to obtain the warning line of the related-party as a warning condition;

[0124] S702: When the transaction scale of a related-party transaction requiring monitoring reaches the related-party warning line, a warning alert is generated and pushed to the related-party transaction processing personnel, who then notify the relevant business departments to perform risk control and transaction processing.

[0125] The warning threshold for a specific related party is calculated by multiplying the amount of related-party transactions subject to monitoring by the actual control threshold Y3. When Company X, a related party under the caliber of Regulatory Authority A, conducts any transaction under any of the business types subject to monitoring, and the overall transaction volume reaches the corresponding warning threshold, the IT system will generate a warning alert and send it to the related-party transaction management personnel. Upon receiving the alert, the related-party transaction management personnel will notify the relevant business departments to pay attention and develop enhanced risk control plans to ensure compliance with related-party transactions.

[0126] In one embodiment, an enterprise builds an IT system to collect a list of related-party customers; establishes a related-party transaction risk measurement model and a related-party risk measurement model based on regulatory rules; conducts model training and risk assessment using machine learning methods (SVM and Bayes algorithm); extracts related-party transaction and related-party characteristics and classifies risk levels; determines the actual control thresholds for monitoring related-party transactions based on model results; the IT system generates early warning alerts, and management personnel take action based on the early warnings.

[0127] refer to Figure 8 , which is a schematic diagram of the architecture relationship of the risk measurement model of an embodiment of the present invention, such as Figure 8 As shown, in this architecture, big data and machine learning are used to manage and control related-party transaction risks. Business channels, including corporate lending, financial markets, interbank asset management, and subsidiaries, are areas where companies conduct transactions.

[0128] The Related-Party Transaction Big Data Cloud Platform is a platform for centrally processing and storing related-party transaction data. Data sources include business data on corporate credit, interbank asset management, subsidiaries, and other businesses, as well as regulatory information such as industrial and commercial information, information from the People's Bank of China, and information from the China Banking and Insurance Regulatory Commission.

[0129] In the Related-Party Transaction Risk Measurement Model Laboratory, machine learning algorithms (such as Support Vector Machines (SVMs)) are used to assess related-party transaction risk and train models. Related-party transaction risk source data serves as the foundation for model training and evaluation. This data is cleaned and formatted for analysis, ensuring data quality and consistency. This cleansing and conversion process yields standardized and unified basic data. Risk dimensional features of related parties and related-party transactions are further extracted for use in risk assessment. Machine learning techniques such as Support Vector Machines (SVMs) are used for model training, evaluation, and optimization.

[0130] Through the related-party transaction application, identified related-party transactions are used for further risk control and management. Finally, the system automatically generates control parameters and thresholds for target business related-party transactions. The system automatically sets control parameters and thresholds for related-party transactions to facilitate supervision and risk control. The overall architecture improves the efficiency and accuracy of related-party transaction management through automation and intelligent means.

[0131] According to the specific regulations of the regulatory authorities, the amount of related-party transactions that need to be monitored is determined. By setting multi-dimensional indicators, a related-party transaction risk measurement model and a related-party risk measurement model are established respectively to distinguish different risk levels. The corresponding control thresholds are set with quantitative indicators. Through automatic calculation and timely early warning by the system, advance control of related-party transactions that need to be monitored is achieved.

[0132] The transaction risk early warning and control method based on the risk measurement model proposed in the present invention can enable enterprises to better manage related transactions, enhance risk management capabilities, and ensure the stable operation of the business. The overall solution can improve the accuracy and efficiency of identification through automated identification of related parties and related transactions, and reduce errors and omissions in manual operations. By utilizing big data and machine learning technology, related transaction risks can be more accurately assessed and managed, providing enterprises with more effective risk control measures, and can monitor related transactions in real time. Once potential risks are discovered, early warnings can be issued in a timely manner to help enterprises respond quickly; automated control parameters and threshold generation help ensure the compliance of corporate related transactions and reduce the occurrence of violations; through quantitative risk assessment, data support is provided to managers to make more reasonable business decisions. Automated and intelligent processing processes reduce reliance on manual labor, thereby reducing labor costs and operating costs.

[0133] The big data platform of this invention can process and analyze large amounts of complex data, improving enterprises' data utilization efficiency. Furthermore, through continuous model training, evaluation, and optimization, it can continuously enhance the model's predictive accuracy and generalization capabilities. It can be adjusted to meet different business needs and regulatory requirements, demonstrating excellent flexibility and scalability. In practical application scenarios, it integrates information from various sources, such as industrial and commercial information, information from the People's Bank of China, and information from the China Banking and Insurance Regulatory Commission, providing comprehensive data support for risk assessment.

[0134] It is noted that, although the operations of the method of the present application are described in a particular order in the above embodiments and drawings, this does not require or imply that the operations must be performed in this particular order, or that all of the operations are necessarily performed to achieve desirable results. Additionally or alternatively, certain steps can be omitted, combined, performed in a different order, and / or split into multiple steps.

[0135] Having introduced the method of the exemplary embodiments of the present application, next, reference is made to Figure 9 The transaction risk early warning control device based on the risk measurement model of the exemplary embodiments of the present application is introduced.

[0136] The implementation of the transaction risk early warning control device based on the risk measurement model can refer to the implementation of the above method, and the repeated parts will not be described again. The term "module" or "unit" used below can be a combination of software and / or hardware that achieves a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware, or a combination of software and hardware is also possible and contemplated.

[0137] Based on the same inventive concept, the present application also proposes a transaction risk early warning control device based on a risk measurement model, as shown in Figure 9 The device comprises:

[0138] The data acquisition module 910 is configured to acquire associated party data and associated transaction data corresponding to the associated party.

[0139] The risk measurement analysis module 920 is configured to input the associated party data and the associated transaction data corresponding to the associated party into an associated party risk measurement model and an associated transaction risk measurement model, respectively, to obtain an associated party risk control threshold and an associated transaction risk control threshold through analysis; wherein the associated party risk measurement model and the associated transaction risk measurement model are obtained through pre-training in the following manner:

[0140] For the associated party risk measurement model, the associated party feature data corresponding to the associated party is collected, and the risk level corresponding to the associated party is labeled. The labeled sample set is divided to obtain a training set and a test set. The machine learning model is trained using the training set, and the machine learning model is mapped to the risk level of the associated party through learning in combination with the risk level corresponding to the associated party. The test set is used for testing to obtain the associated party risk measurement model. The risk level corresponding to the associated party output by the model is used to determine the associated party risk control threshold.

[0141] For the related-party transaction risk measurement model, characteristic data of related-party transactions are collected, and the corresponding risk control thresholds of the related-party transactions are marked. The marked sample set is divided into a training set and a test set. The machine learning model is trained using the training set. The machine learning model is mapped from related-party transactions to risk control thresholds in combination with the risk control thresholds corresponding to the related-party transactions. The model is then tested using the test set to obtain the related-party transaction risk measurement model.

[0142] The risk control threshold determination module 930 is configured to multiply the related party risk control threshold and the risk control threshold corresponding to the related transaction to determine the transaction risk control threshold;

[0143] The transaction warning module 940 is used to set warning conditions according to the transaction risk control threshold, generate warning prompts when the related transaction data corresponding to the related parties meet the warning conditions, and control the related transactions according to the warning prompts.

[0144] In one embodiment, the data acquisition module 910 is specifically configured to:

[0145] Collect a list of related party customers;

[0146] Based on the related party customer list, related party information and related transaction information corresponding to the related parties are counted.

[0147] In one embodiment, reference Figure 10 , the apparatus further includes: an associated party data processing module 950;

[0148] The related party data processing module 950 is specifically used to:

[0149] According to the regulatory agency's transaction supervision requirements, collect historical transaction information of related parties under different calibers, including at least transaction scale and frequency, and establish a related party data mart;

[0150] Filter and clean related parties, extracting related party data with related transactions from historical transaction information, eliminating invalid and redundant data, and unifying and standardizing the cleaned related party data;

[0151] Determine the related party risk measurement score based on the related party data mart using multiple dimensions, including the number of companies corresponding to the related parties, the frequency of transactions, the number of related party transactions with a maximum single amount exceeding a set limit within a set time period, the probability of related party transactions of preset business types occurring within the related party, and the fair value of related party transactions within a set time period;

[0152] A plurality of risk levels corresponding to the related parties are obtained according to the related party risk measurement scores, and a corresponding related party risk control threshold is set for each risk level.

[0153] In one embodiment, the risk measurement analysis module 920 is specifically configured to:

[0154] For the related-party risk measurement model, the related-party characteristic data corresponding to the related-party is collected. Feature extraction and analysis are performed based on the business dimension characteristic values ​​including the related-party role, business type, transaction scale, and transaction frequency to determine the related-party characteristic data as the sample set for model training;

[0155] Using the Bayesian algorithm to train the model, we construct a classification prediction of the risk level corresponding to the related parties and the related party characteristic data, and obtain a related party risk measurement model;

[0156] The related party risk measurement model is tested, the characteristic dimensions of the related parties are automatically corrected according to the test results, the characteristic data of the related parties are optimized, and the model parameters are adjusted manually.

[0157] In one embodiment, reference Figure 10 , the device further includes: an associated transaction data processing module 960;

[0158] The associated transaction data processing module 960 is specifically used to:

[0159] Determine the transaction type, amount, business type to be included in the calculation, and calculation mechanism for related-party transactions based on the regulatory authorities' transaction supervision requirements;

[0160] Based on the calculation mechanism, historical transaction data is analyzed to determine the transaction scale, increase or decrease of related-party transactions, taking into account the transaction type, amount, and business type that corresponds to the transaction type and needs to be included in the calculation, and a related-party transaction data mart is established;

[0161] Filter and cleanse the business data of related-party transactions, including eliminating invalid and redundant data, and converting, integrating, discretizing, and normalizing the data;

[0162] Based on the related-party transaction data mart, the risk control threshold corresponding to the related-party transaction is determined through multiple dimensions including the related-party role, business type, transaction scale, risk limit and increase or decrease range.

[0163] In one embodiment, the risk measurement analysis module 920 is specifically configured to:

[0164] For the related-party transaction risk measurement model, we collect related-party transaction feature data corresponding to related-party transactions. We perform feature extraction and analysis based on the business dimension feature values, including the related-party role, business type, transaction size, risk limit, and increase / decrease range, to determine the related-party transaction feature data, which serves as the sample set for model training.

[0165] The model is trained using the support vector machine algorithm, and the Gaussian kernel is used as the kernel function to optimize the model accuracy and generalization ability, and the F1 score is used to evaluate the model.

[0166] In one embodiment, the transaction warning module 940 is specifically configured to:

[0167] Multiply the amount of the related-party transaction to be monitored by the transaction risk control threshold to obtain the early warning line of the related-party, which serves as the early warning condition;

[0168] When the transaction scale of a related-party transaction that requires monitoring reaches the related-party's warning line, an early warning prompt will be generated and pushed to the related-party transaction processing personnel, who will then notify the relevant business departments to conduct risk control and transaction processing.

[0169] It should be noted that while the detailed description above mentions several modules of the risk measurement model-based trading risk early warning control device, this division is merely exemplary and not mandatory. In practice, according to embodiments of the present invention, the features and functions of two or more modules described above may be embodied in a single module. Conversely, the features and functions of a single module described above may be further divided and embodied by multiple modules.

[0170] Based on the above invention concept, Figure 11 As shown, the present invention also proposes a computer device 1100, including a memory 1110, a processor 1120, and a computer program 1130 stored in the memory 1110 and executable on the processor 1120. When the processor 1120 executes the computer program 1130, the aforementioned transaction risk early warning control method based on the risk measurement model is implemented.

[0171] Based on the aforementioned inventive concept, the present invention proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the aforementioned transaction risk early warning control method based on the risk measurement model.

[0172] Based on the aforementioned inventive concept, the present invention proposes a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements a transaction risk early warning control method based on a risk measurement model.

[0173] The transaction risk early warning control method and device based on the risk measurement model proposed in the present invention obtain related party data and related transaction data corresponding to the related parties; input the related party data and the related transaction data corresponding to the related parties into the related party risk measurement model and the related transaction risk measurement model respectively, and obtain the related party risk control threshold and the related transaction risk control threshold after analysis; wherein, the related party risk measurement model and the related transaction risk measurement model are pre-trained in the following manner: for the related party risk measurement model, the related party feature data corresponding to the related parties are collected, and the risk level corresponding to the related parties are marked, and the marked sample set is divided to obtain a training set and a test set; the machine learning model is trained using the training set, and the machine learning model is mapped from the related parties to the risk level in combination with the risk level corresponding to the related parties, and the test set is used for testing to obtain the related party risk measurement model, and the related party risk control threshold is determined according to the risk level corresponding to the related parties output by the model; for the related transaction risk measurement model, the related transaction feature data are collected, and the risk control threshold corresponding to the related transactions are marked, and the marked sample set is divided to obtain a training set and a test set; the machine learning model is trained using the training set, In combination with the risk control thresholds corresponding to the related-party transactions, the machine learning model is trained to map related-party transactions to risk control thresholds, and tested using the test set to obtain a related-party transaction risk measurement model. The related-party risk control threshold is multiplied by the risk control threshold corresponding to the related-party transaction to determine the transaction risk control threshold. Warning conditions are set based on the transaction risk control thresholds. When the related-party transaction data corresponding to the related-party meets the warning conditions, a warning prompt is generated, and the related-party transaction is controlled based on the warning prompt. The overall solution automatically completes warning analysis for related-party transactions that require monitoring, significantly reducing the workload of personnel in related-party transaction management positions, reducing statistical errors, avoiding statistical caliber errors caused by personnel handovers or replacements, improving control accuracy, and enhancing work efficiency. The present invention utilizes a related-party risk measurement model and a related-party transaction risk measurement model to distinguish different risk levels and set corresponding control thresholds based on quantitative indicators. By combining the two control thresholds, the actual control threshold for each related-party in each type of related-party transaction that requires monitoring can be individually determined based on actual risk conditions. Preemptive control of related-party transactions that require monitoring is achieved through automatic calculation and timely warning, ensuring early warning and timely response, reducing the risk of violations, and providing strong technical support for risk warning.

[0174] The acquisition, storage, use, and processing of data in the technical solution of this application comply with relevant laws and regulations.

[0175] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, apparatus, or computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer readable program code.

[0176] The present application is described in reference to the flowchart illustrations and / or block diagrams of methods and computer program products according to embodiments of the application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 means for performing each of the functions specified in the flowchart illustrations and / or block diagrams.

[0177] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 means for performing each of the functions specified in the flowchart illustrations and / or block diagrams.

[0178] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart illustrations and / or block diagrams. Figure 1 one or more functions specified in the flowchart illustrations and / or block diagrams. Figure 1 means for performing each of the functions specified in the flowchart illustrations and / or block diagrams.

[0179] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A transaction risk early warning control method based on a risk measurement model, characterized in that: include: Obtain related party data and related transaction data corresponding to the related parties; The related-party data and the related-party transaction data corresponding to the related-party are respectively input into the related-party risk measurement model and the related-party transaction risk measurement model, and the related-party risk control threshold and the related-party transaction risk control threshold are obtained through analysis; wherein the related-party risk measurement model and the related-party transaction risk measurement model are pre-trained in the following manner: For the related-party risk measurement model, the related-party characteristic data corresponding to the related-party is collected and the risk level corresponding to the related-party is marked. The marked sample set is divided into a training set and a test set. The machine learning model is trained using the training set. The machine learning model is mapped from the related-party to the risk level in combination with the risk level corresponding to the related-party. The model is tested using the test set to obtain the related-party risk measurement model. The related-party risk control threshold is determined based on the risk level corresponding to the related-party output by the model. For the related-party transaction risk measurement model, characteristic data of related-party transactions are collected, and the corresponding risk control thresholds of the related-party transactions are marked. The marked sample set is divided into a training set and a test set. The machine learning model is trained using the training set. The machine learning model is mapped from related-party transactions to risk control thresholds in combination with the risk control thresholds corresponding to the related-party transactions. The model is then tested using the test set to obtain the related-party transaction risk measurement model. Multiplying the related-party risk control threshold and the risk control threshold corresponding to the related-party transaction to determine the transaction risk control threshold; Setting a warning condition based on the transaction risk control threshold, generating a warning prompt when the related-party transaction data corresponding to the related-party meets the warning condition, and controlling the related-party transaction based on the warning prompt; Among them, for the related party risk measurement model, related party characteristic data corresponding to the related parties are collected, and the risk levels corresponding to the related parties are marked, and the marked sample set is divided to obtain a training set and a test set; the training set is used to train the machine learning model, and the machine learning model is mapped from related parties to risk levels in combination with the risk levels corresponding to the related parties, and the test set is used to test to obtain the related party risk measurement model, and the related party risk control threshold is determined according to the risk levels corresponding to the related parties output by the model, including: For the related-party risk measurement model, the related-party characteristic data corresponding to the related-party is collected. Feature extraction and analysis are performed based on the business dimension characteristic values ​​including the related-party role, business type, transaction scale, and transaction frequency to determine the related-party characteristic data as the sample set for model training; Using the Bayesian algorithm to train the model, we construct a classification prediction of the risk level corresponding to the related parties and the related party characteristic data, and obtain a related party risk measurement model; Testing the related-party risk measurement model, automatically correcting the characteristic dimensions of the related-party based on the test results, optimizing the related-party characteristic data, and manually adjusting the model parameters; The related-party transaction risk measurement model includes collecting related-party transaction feature data, marking the risk control thresholds corresponding to the related-party transactions, and dividing the marked sample set to obtain a training set and a test set. The machine learning model is trained using the training set, and the machine learning model is mapped from related-party transactions to risk control thresholds in combination with the risk control thresholds corresponding to the related-party transactions. The model is then tested using the test set to obtain the related-party transaction risk measurement model, including: For the related-party transaction risk measurement model, we collect related-party transaction feature data corresponding to related-party transactions. We perform feature extraction and analysis based on the business dimension feature values, including the related-party role, business type, transaction size, risk limit, and increase / decrease range, to determine the related-party transaction feature data, which serves as the sample set for model training. The model is trained using the support vector machine algorithm, and the Gaussian kernel is used as the kernel function to optimize the model accuracy and generalization ability, and the F1 score is used to evaluate the model.

2. The transaction risk early warning control method based on the risk measurement model according to claim 1 is characterized in that: The method includes: Collect a list of related party customers; Based on the related party customer list, related party information and related transaction information corresponding to the related parties are counted.

3. The transaction risk early warning control method based on the risk measurement model according to claim 1 is characterized in that: The method further includes: According to the regulatory agency's transaction supervision requirements, collect historical transaction information of related parties under different calibers, including at least transaction scale and frequency, and establish a related party data mart; Filter and clean related parties, extracting related party data with related transactions from historical transaction information, eliminating invalid and redundant data, and unifying and standardizing the cleaned related party data; Determine the related party risk measurement score based on the related party data mart using multiple dimensions, including the number of companies corresponding to the related parties, the frequency of transactions, the number of related party transactions with a maximum single amount exceeding a set limit within a set time period, the probability of related party transactions of preset business types occurring within the related party, and the fair value of related party transactions within a set time period; A plurality of risk levels corresponding to the related parties are obtained according to the related party risk measurement scores, and a corresponding related party risk control threshold is set for each risk level.

4. The transaction risk early warning control method based on the risk measurement model according to claim 1 is characterized in that: The method further includes: Determine the transaction type, amount, business type to be included in the calculation, and calculation mechanism for related-party transactions based on the regulatory authorities' transaction supervision requirements; Based on the calculation mechanism, historical transaction data is analyzed to determine the transaction scale, increase or decrease of related-party transactions, taking into account the transaction type, amount, and business type that corresponds to the transaction type and needs to be included in the calculation, and a related-party transaction data mart is established; Filter and cleanse the business data of related-party transactions, including eliminating invalid and redundant data, and converting, integrating, discretizing, and normalizing the data; Based on the related-party transaction data mart, the risk control threshold corresponding to the related-party transaction is determined through multiple dimensions including the related-party role, business type, transaction scale, risk limit and increase or decrease range.

5. The transaction risk early warning control method based on the risk measurement model according to claim 1 is characterized in that: Setting a warning condition according to the transaction risk control threshold, generating a warning prompt when the related-party transaction data corresponding to the related-party meets the warning condition, and controlling the related-party transaction according to the warning prompt, including: Multiply the amount of the related-party transaction to be monitored by the transaction risk control threshold to obtain the early warning line of the related party, which serves as the early warning condition; When the transaction scale of a related-party transaction that requires monitoring reaches the related-party's warning line, an early warning prompt will be generated and pushed to the related-party transaction processing personnel, who will then notify the relevant business departments to conduct risk control and transaction processing.

6. A transaction risk early warning control device based on a risk measurement model, characterized in that: include: A data acquisition module is used to obtain related party data and related transaction data corresponding to the related parties; The risk measurement analysis module is configured to input the related-party data and the related-party transaction data corresponding to the related-party into a related-party risk measurement model and a related-party transaction risk measurement model, respectively, and obtain, through analysis, a related-party risk control threshold and a related-party transaction risk control threshold; wherein the related-party risk measurement model and the related-party transaction risk measurement model are pre-trained in the following manner: For the related-party risk measurement model, the related-party characteristic data corresponding to the related-party is collected and the risk level corresponding to the related-party is marked. The marked sample set is divided into a training set and a test set. The machine learning model is trained using the training set. The machine learning model is mapped from the related-party to the risk level in combination with the risk level corresponding to the related-party. The model is tested using the test set to obtain the related-party risk measurement model. The related-party risk control threshold is determined based on the risk level corresponding to the related-party output by the model. For the related-party transaction risk measurement model, characteristic data of related-party transactions are collected, and the corresponding risk control thresholds of the related-party transactions are marked. The marked sample set is divided into a training set and a test set. The machine learning model is trained using the training set. The machine learning model is mapped from related-party transactions to risk control thresholds in combination with the risk control thresholds corresponding to the related-party transactions. The model is then tested using the test set to obtain the related-party transaction risk measurement model. a risk control threshold determination module, configured to multiply the related party risk control threshold by the risk control threshold corresponding to the related transaction to determine a transaction risk control threshold; A transaction warning module is used to set warning conditions according to the transaction risk control threshold, generate warning prompts when the related-party transaction data corresponding to the related-party meets the warning conditions, and control the related-party transaction according to the warning prompts; The risk measurement and analysis module is specifically used to: For the related-party risk measurement model, the related-party characteristic data corresponding to the related-party is collected. Feature extraction and analysis are performed based on the business dimension characteristic values ​​including the related-party role, business type, transaction scale, and transaction frequency to determine the related-party characteristic data as the sample set for model training; Using the Bayesian algorithm to train the model, we construct a classification prediction of the risk level corresponding to the related parties and the related party characteristic data, and obtain a related party risk measurement model; Testing the related-party risk measurement model, automatically correcting the characteristic dimensions of the related-party based on the test results, optimizing the related-party characteristic data, and manually adjusting the model parameters; The risk measurement and analysis module is specifically used to: For the related-party transaction risk measurement model, we collect related-party transaction feature data corresponding to related-party transactions. We perform feature extraction and analysis based on the business dimension feature values, including the related-party role, business type, transaction size, risk limit, and increase / decrease range, to determine the related-party transaction feature data, which serves as the sample set for model training. The model is trained using the support vector machine algorithm, and the Gaussian kernel is used as the kernel function to optimize the model accuracy and generalization ability, and the F1 score is used to evaluate the model.

7. The transaction risk early warning control device based on the risk measurement model according to claim 6, characterized in that: The data acquisition module is specifically used for: Collect a list of related party customers; Based on the related party customer list, related party information and related transaction information corresponding to the related parties are counted.

8. The transaction risk early warning control device based on the risk measurement model according to claim 6, characterized in that: The device also includes: an associated party data processing module; The related party data processing module is specifically used to: According to the regulatory agency's transaction supervision requirements, collect historical transaction information of related parties under different calibers, including at least transaction scale and frequency, and establish a related party data mart; Filter and clean related parties, extracting related party data with related transactions from historical transaction information, eliminating invalid and redundant data, and unifying and standardizing the cleaned related party data; Determine the related party risk measurement score based on the related party data mart using multiple dimensions, including the number of companies corresponding to the related parties, the frequency of transactions, the number of related party transactions with a maximum single amount exceeding a set limit within a set time period, the probability of related party transactions of preset business types occurring within the related party, and the fair value of related party transactions within a set time period; A plurality of risk levels corresponding to the related parties are obtained according to the related party risk measurement scores, and a corresponding related party risk control threshold is set for each risk level.

9. The transaction risk early warning control device based on the risk measurement model according to claim 6, characterized in that: The device also includes: an associated transaction data processing module; The associated transaction data processing module is specifically used to: Determine the transaction type, amount, business type to be included in the calculation, and calculation mechanism for related-party transactions based on the regulatory authorities' transaction supervision requirements; Based on the calculation mechanism, historical transaction data is analyzed to determine the transaction scale, increase or decrease of related-party transactions, taking into account the transaction type, amount, and business type that corresponds to the transaction type and needs to be included in the calculation, and a related-party transaction data mart is established; Filter and cleanse the business data of related-party transactions, including eliminating invalid and redundant data, and converting, integrating, discretizing, and normalizing the data; Based on the related-party transaction data mart, the risk control threshold corresponding to the related-party transaction is determined through multiple dimensions including the related-party role, business type, transaction scale, risk limit and increase or decrease range.

10. The transaction risk early warning control device based on the risk measurement model according to claim 6, characterized in that: The transaction warning module is specifically used to Multiply the amount of the related-party transaction to be monitored by the transaction risk control threshold to obtain the early warning line of the related party, which serves as the early warning condition; When the transaction scale of a related-party transaction that requires monitoring reaches the related-party's warning line, an early warning prompt will be generated and pushed to the related-party transaction processing personnel, who will then notify the relevant business departments to conduct risk control and transaction processing.

11. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 5 is implemented.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

13. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images