Big data driven compliance risk prediction and management method

Abstract

The invention relates to the technical field of risk data processing, and provides a big data driven compliance risk prediction and management method. The method comprises the steps that big data retrieval is carried out, and a risk feature library is mined; traversing the feature library, marking positive and negative sample sequences based on feature correlation, performing polarization processing, and determining risk sample data; supervising and training the risk prediction model by taking the prediction time phase as a constraint; reading target data, performing fuzzy screening and vectorization, and determining a standardized data system; transmitting the data to a prediction model, matching risk features and generating a risk single column; and performing compliance risk control management based on the single-row determination plan. The technical problems of inaccurate risk prediction and low management efficiency caused by huge business data and complex and changeable compliance risks are solved, the risk plan is automatically generated based on the prediction result, the accuracy and timeliness of compliance risk prediction are improved, the pertinence and effectiveness of risk management are enhanced, and the risk management efficiency is improved. And the risk influence degree is reduced.

Description

Technical Field

[0001] The present application relates to the field of computer science and technology, specifically to the field of risk data processing technology, and in particular to a big data-driven compliance risk prediction and management method. Background Art

[0002] With the continuous development of global financial and commercial activities, the complexity and importance of compliance risk management are also increasing. In the current market environment, enterprises are faced with the need to process massive amounts of business data. These data are not only large in size but also diverse in structure, covering transaction records, customer information, financial data, and various external regulatory requirements. In addition, with the diversification and internationalization of business scenarios, the speed of change of risk factors has also greatly accelerated, and traditional risk management methods are difficult to cope with this complex situation. Traditional compliance risk prediction and management methods usually rely on static rules or empirical judgments, which may have been able to meet certain needs in the past, but in the face of modern high-frequency and real-time business environments. In addition, with the increasingly stringent and changeable regulatory requirements, the pressure on corporate compliance risk management is also increasing. Enterprises need to flexibly respond to market changes while complying with regulations to avoid compliance risks caused by lagging management. These challenges make traditional risk management methods increasingly insufficient in the modern business environment. Traditional risk management methods are unable to handle large and complex business data, and are difficult to cope with rapidly changing risk factors, resulting in inaccurate risk prediction and inefficient management. Summary of the invention

[0003] This application provides a big data-driven compliance risk prediction and management method, aiming to solve the technical problems of inaccurate risk prediction and inefficient management due to the huge amount of business data and the complexity and variability of compliance risks.

[0004] In view of the above problems, this application provides a big data-driven compliance risk prediction and management method.

[0005] The present application provides a big data-driven compliance risk prediction and management method, which includes: interacting with business features, performing big data retrieval and mining a risk feature library, wherein the risk feature library is based on a risk inducement set-a risk type set-a risk impact effect set; traversing the risk feature library, marking positive sample sequences and negative sample sequences based on feature correlation, and performing sample polarization processing to determine risk sample data, wherein the positive sample sequence is a correlation feature pair, and the negative sample sequence is a non-correlation feature pair, wherein feature correlation includes inter-set correlation and intra-set correlation; based on the risk sample data, supervised training of a risk prediction model with a prediction phase as a constraint; reading target business data and performing data fuzzy screening, performing data vectorization in combination with a data encoder, and determining a standardized data system, wherein the data encoder includes multiple data format blocks; transmitting the standardized data system to the risk prediction model, performing risk feature matching and risk prediction, and generating a risk list; determining a risk plan based on the risk list, and performing compliance risk risk control management.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: The above-mentioned big data-driven compliance risk prediction and management method first builds a comprehensive risk feature library, which analyzes the characteristics of interactive business and subdivides risks into different causes, types and impact effects. Subsequently, by using this risk feature library, by comparing the correlation between features, it is possible to distinguish which feature combinations are truly risk-related and which are irrelevant, determine the risk sample data, and train a risk prediction model based on these labeled sample data. This model can predict compliance risks that may occur in the future based on historical data and feature relationships. When there is new business data to be checked, the data is first pre-processed, including fuzzy screening and vectorization, so that the model can analyze the data more easily. After that, the processed data is input into the risk prediction model, and the model will match and predict the risk features to generate a risk list. Then, based on this risk list, a corresponding risk response plan is formulated to achieve effective management and control of compliance risks and ensure the sound operation of the enterprise.

[0007] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0009] Figure 1 A flowchart of a big data-driven compliance risk prediction and management method in one embodiment; Figure 2 A flowchart of sample polarization processing of a big data-driven compliance risk prediction and management method in one embodiment. DETAILED DESCRIPTION

[0010] The embodiments of the present application provide a big data-driven compliance risk prediction and management method to solve the technical problems of inaccurate risk prediction and inefficient management due to the huge amount of business data and the complexity and variability of compliance risks.

[0011] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0012] It should be noted that the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0013] Examples, such as Figure 1 As shown, the present application provides a big data-driven compliance risk prediction and management method, the method comprising: Interactive business features, perform big data retrieval and mine a risk feature library, wherein the risk feature library is based on a risk cause set, a risk type set, and a risk impact effect set.

[0014] In the embodiment of the present application, when processing complex businesses, the system terminal will first interact with the business terminal to understand the characteristics of these businesses, and then use big data technology to conduct extensive information retrieval and deep mining, and build a risk feature library through the relevant data retrieved and mined. This risk feature library organizes risk information in the form of feature pairs, including the risk inducement (i.e., the factors that cause the risk to occur) - the risk type (i.e., the nature and classification of the risk) - the risk impact effect (i.e., the possible consequences of the risk once it occurs) feature pairs. In this way, it is possible to more comprehensively understand and grasp the various risks that may be encountered in the business, providing a solid foundation for subsequent risk prediction and management.

[0015] The risk feature library is traversed, positive sample sequences and negative sample sequences are marked based on feature correlation, and sample polarization processing is performed to determine risk sample data, wherein the positive sample sequence is a correlation feature pair, and the negative sample sequence is a non-correlation feature pair, wherein feature correlation includes inter-set correlation and intra-set correlation.

[0016] In one embodiment, in the process of traversing the risk feature library, the system terminal uses the Pearson correlation coefficient to analyze the features in the risk feature library to determine the correlation between them, calculate the inter-set correlation and the intra-set correlation, and quantify the inter-set correlation and the intra-set correlation. The former refers to how the features between different risk inducement sets, risk type sets, or risk impact effect sets affect each other, while the latter refers to the correlation between the features within the same set. Subsequently, based on the correlation of these features, the data in the risk feature library is divided into positive sample sequences and negative sample sequences. The positive sample sequence consists of feature pairs that show strong correlation. These feature pairs are often closely related during the occurrence or evolution of risks, and are of great significance for risk identification and management. In contrast, the negative sample sequence contains non-correlated feature pairs, which are usually not directly meaningful or have little impact in risk analysis. Among them, strong correlation and non-correlation are determined based on positive correlation thresholds and negative correlation thresholds. The system terminal marks the feature pairs whose inter-set correlation and intra-set correlation are both greater than or equal to the positive correlation threshold as positive sample sequences, indicating that they may be closely related during the occurrence or evolution of risks. The feature pairs whose inter-set correlation and intra-set correlation are both less than or equal to the negative correlation threshold are regarded as negative sample sequences, indicating that they may not have direct significance or have little impact in risk analysis. Afterwards, the system terminal maximizes the obtained positive sample sequence and minimizes the negative sample sequence according to the pre-built loss incentive function, and integrates the processed positive sample sequence and negative sample sequence to obtain risk sample data. This step aims to amplify the difference between positive samples and negative samples, so that the subsequent risk prediction model can more easily learn the essential laws of risk characteristics during the training process, thereby improving the accuracy of the risk prediction model.

[0017] Further, if Figure 2 As shown, the present application provides the sample polarization processing, including: A loss excitation function is constructed with correlation as the first excitation feature and recognition probability as the second excitation feature, wherein the loss excitation function is used to determine the polarization direction and the polarization coefficient; and the positive sample sequence and the negative sample sequence are polarized in combination with the loss excitation function.

[0018] Preferably, in order to make the positive sample sequence and the negative sample sequence samples more representative and effective, the system terminal uses the correlation of the feature pair as the first excitation feature, the recognition probability as the second excitation feature, and combines the two excitation features with the corresponding threshold to construct a loss excitation function. The purpose of designing this function is to determine the polarization direction and polarization coefficient of the sample through the output of the function during the polarization process. Among them, the correlation of the feature pair is obtained by calculating the mean of the inter-set correlation and intra-set correlation of the feature pair. The recognition probability is set by the feature pair based on historical experience. The more difficult the feature is to recognize, the lower the recognition probability. The polarization direction determines how the sample moves or adjusts in the feature space to better distinguish between positive and negative samples. The polarization coefficient is used to quantify the amplitude of this adjustment. Subsequently, the system terminal analyzes and calculates the positive sample sequence and the negative sample sequence according to this loss excitation function, determines the polarization direction according to the positive and negative of the calculation result, determines the polarization coefficient according to the value of the calculation result, and maximizes and minimizes the positive sample sequence and the negative sample sequence according to the determined polarization direction and polarization coefficient to obtain risk sample data.

[0019] Furthermore, the present application provides an initialization deployment strategy, including: The positive sample sequence is traversed, and a maximization coefficient is determined in combination with the loss excitation function, and a maximization process is performed on the matching positive sample sequence to determine a first polarization sample; the negative sample sequence is traversed, and a minimization coefficient is determined in combination with the loss excitation function, and a minimization process is performed on the matching negative sample sequence to determine a second polarization sample; the first polarization sample and the second polarization sample are integrated to determine the risk sample data.

[0020] Optionally, the system terminal traverses the positive sample sequence, and in each traversal, the correlation and recognition probability of the feature pair of each positive sample are input into the constructed loss excitation function to determine the maximization coefficient and maximization direction of each positive sample. The purpose of this maximization coefficient is to maximize the correlation between the positive sample and the risk feature when performing the maximization process, while minimizing the non-correlation between them and the non-risk feature. The maximization process involves operations such as adjusting the position of the sample in the feature space, amplifying the feature dimension of the sample, and changing the weight of the sample. Through this maximization process, the system terminal obtains the first polarization sample, and the representation of this first polarization sample in the feature space will become more significant and easy to identify, thereby improving the recognition accuracy of the risk prediction model. Similarly, the system terminal traverses the negative sample sequence and determines a minimization coefficient and minimization direction for each negative sample in combination with the loss excitation function. The goal of this minimization coefficient is to reduce the interference of the negative sample on the model recognition when performing the minimization process, so that it can produce a greater difference in features with the positive sample, so as to facilitate the model to distinguish. Through minimization processing, the system terminal obtains a second polarization sample, which will no longer be easily identified as a positive sample, further improving the recognition accuracy and reliability of the risk prediction model.

[0021] Further, the present application provides polarization coefficients, including: The polarization coefficient of the positive sample sequence is positively correlated with the first excitation feature and negatively correlated with the second excitation feature; the polarization coefficient of the negative sample sequence is positively correlated with the second excitation feature.

[0022] Optionally, the polarization coefficient of the positive sample sequence is positively correlated with the first excitation feature and negatively correlated with the second excitation feature. When the correlation of the positive sample sequence is greater, its polarization coefficient is also greater, which means that the sample will be amplified during the training process so that the model can more accurately identify such samples. However, if the probability of its current recognition by the model is low, it indicates that the performance of the model on this sample still has room for improvement. Therefore, the smaller the recognition probability of the positive sample, the larger its polarization coefficient will be adjusted to encourage the model to pay more attention to these samples during the training process, thereby increasing its recognition probability. Unlike the positive sample sequence, the polarization coefficient of the negative sample sequence is mainly positively correlated with its recognition probability. This is because the model should be able to clearly distinguish between positive and negative samples as much as possible. When the recognition probability of a negative sample is large (that is, the probability that the model mistakenly identifies it as a positive class is high), its polarization coefficient will also increase accordingly, speeding up the minimization process to weaken or suppress the model's misrecognition of the sample.

[0023] Based on the risk sample data, the risk prediction model is supervised and trained with the prediction phase as a constraint.

[0024] In one embodiment, when training a risk prediction model, the system terminal uses the prediction phase, that is, the time range to be predicted, as a constraint of the model, and uses this prediction phase to divide the risk sample data into multiple time windows of fixed length. Each window contains a series of time steps as the input of the LSTM model. Subsequently, the neural network model is constructed using the LSTM unit. The LSTM model can process sequence data and pass state information between time steps to capture long-term dependencies in the sequence. After that, the LSTM model is trained using the training set data. During the training process, the data of each time window is used as input and the corresponding risk label is used as output. The model parameters are optimized by the back propagation algorithm to reduce the prediction error. The trained model is then tested using the risk sample data that is not used for training, its performance is evaluated, and the model parameters are adjusted, such as the learning rate, the number of network layers, the time window size, etc., to ensure the objectivity and accuracy of the evaluation results. When the test passes, the system terminal outputs the LSTM model as a risk prediction model to provide technical support for risk management.

[0025] The target business data is read and fuzzy screened, and the data is vectorized in combination with a data encoder to determine a standardized data system, wherein the data encoder includes multiple data format blocks.

[0026] In one embodiment, in the process of data processing, the system terminal first reads relevant data from the target business system. These data come from a variety of data sources, including but not limited to databases, log files, API interfaces, etc., which store key information of business operations in different formats and structures. In order to extract valuable information from these massive and possibly chaotic data, the system terminal uses a fuzzy screening method to preliminarily screen and filter the original data to remove invalid and redundant data. This process is intended to reduce the complexity of data processing and improve the efficiency and accuracy of subsequent steps. Fuzzy screening allows data items that do not meet the requirements to be excluded according to certain rules or standards in the case of incomplete and accurate matching. This rule or standard is set according to actual needs, for example, irrelevant records are eliminated based on keyword matching, and data with missing key fields are eliminated based on data integrity checks. After the data is screened, in order to improve the efficiency and accuracy of data processing, the system terminal uses a data encoder to vectorize the data. Data vectorization is a key step in data preprocessing, which is to extract valid entity information and the relationship between them from the data and convert this information into a numerical feature vector. This process is the basis for subsequent data analysis and model training. The data encoder is a powerful tool that contains multiple data format blocks to cope with different types of data sources. For example, for image data, a data encoder is used to extract image features and convert them into vectors; for text data, a data encoder is used to convert text content into word vectors or sentence vectors. These different data format blocks ensure that the data encoder can flexibly handle various types of data and uniformly convert them into a format suitable for processing by machine learning algorithms. In order to ensure the consistency and comparability of data, the system terminal determines a standardized data system. This system specifies key attributes such as data format, unit, and precision to ensure that all vectorized data follows unified specifications. The establishment of a standardized data system helps to improve the accuracy and reliability of data analysis, and also promotes the sharing and exchange of data between different systems and platforms.

[0027] Further, the present application provides the method of combining the data encoder to perform data vectorization, including: Acquire and filter business data, divide it based on data format, and determine multiple groups of business data; traverse the multiple groups of business data, match and vectorize the multiple data format blocks, and determine vector data; based on data entity relationships, integrate the vector data and determine the standardized data system.

[0028] Preferably, during the data processing, the system terminal first transfers the business data obtained and screened from various sources to the data encoder. Due to the diverse sources, these data will contain a variety of different formats and structures, such as text, images, etc. In order to process these data more efficiently, the system terminal will divide them based on their formats in the data encoder, group data of similar formats into one group, and thus determine multiple groups of business data. Subsequently, multiple groups of business data are traversed and processed. In this step, according to the multiple data format blocks (such as text paragraphs, image areas, etc.) contained in the data, the corresponding algorithm in the data encoder is called for matching and vectorization processing. For example, for text paragraphs, text encoding algorithms such as TF-IDF, Word2Vec, BERT, etc. are used for vectorization, and image encoding algorithms such as CNN, ResNet, ViT, etc. are used for image areas for feature extraction and vectorization. The purpose of this process is to convert the original data into a vector form that can be understood and processed by the machine for subsequent data analysis. After completing the vectorization processing, the system terminal integrates these vector data according to the relationship between the data entities. Data entity relationships refer to the association or dependency between data items, such as the relationship between users and orders, the relationship between products and services, etc. By integrating these vector data, a more complete and accurate data view is constructed, thereby determining a standardized data system. The standardized data system is one of the important results of data processing. It defines the structure, format, unit and other standards of data, ensuring the consistency and comparability of data. This system provides a solid foundation for subsequent data analysis, enabling data to more effectively support business decisions.

[0029] The standardized data system is transmitted to the risk prediction model to perform risk feature matching and risk prediction to generate a risk list.

[0030] In one embodiment, at the final stage of the data processing flow, the standardized data system will be transmitted to the risk prediction model. When the standardized data system enters the model, the model will automatically match the risk characteristics of the data, that is, check whether there are elements or patterns in the data that match the known risk patterns. Through the matching of risk characteristics, the model can evaluate the potential risk level of the data, and make risk predictions based on this, obtain multiple risk sequences, and fuse them through the output layer to generate a risk list. This risk list lists the possible risk events in detail. In this way, the system terminal can formulate a response strategy based on the risk list to reduce the impact of potential risks on business operations and ensure the stable operation of the business.

[0031] Furthermore, the present application provides the risk generation list, including: Perform risk feature matching to determine risk-related feature vectors; traverse the risk-related feature vectors, perform feature coupling analysis, and determine a first risk sequence; traverse the first risk sequence and determine a second risk sequence based on temporal evolution and spatial extensibility, wherein the risk sequence is used to measure risk events; and integrate the first risk sequence and the second risk sequence in time series to generate the risk single column.

[0032] Preferably, in the process of risk prediction, the system terminal first uses the risk prediction model to match the risk characteristics. This step is to extract risk-related features from the standardized data system and obtain risk-related feature vectors for subsequent analysis. These risk-related feature vectors are the basis for risk assessment, and they represent potential risk signals in the data. Subsequently, in the risk prediction model, these risk-related feature vectors are traversed to perform feature coupling analysis. Feature coupling analysis refers to analyzing the interaction and correlation between different risk features to determine the risk pattern they form together. In the feature coupling analysis process, the system terminal defines a series of risk coupling rules based on business knowledge and expert experience. These rules describe how different feature vectors interact to generate risks. For example, a high inventory level plus a low sales growth rate may indicate the risk of inventory backlog; a high debt ratio plus a cash flow shortage may indicate a financial crisis, etc. Then traverse each feature vector and check whether it has a potential risk association with other feature vectors according to the defined coupling rules. Combine multiple feature vectors together to analyze the risks that may arise from their joint action. This is achieved by combining coupling rules using logical judgments, conditional statements, etc. According to the analysis results of the coupling rules, potential risk points are identified and summarized to obtain the first risk sequence. The first risk sequence is a preliminary classification of risk events, revealing the internal logic and association of risk factors. After that, the system terminal starts from the first risk point in the first risk sequence, analyzes each risk point one by one, and considers its possible temporal evolution and spatial expansibility. For each risk point, analyze its current status, development trend and possible change trajectory. Among them, temporal evolution refers to the development and change process of risk over time, including the origin, development, climax and decline of risk. Analyzing temporal evolution helps predict the future change trend and possible impact of risk. Spatial expansibility refers to the spread and impact of risk within a spatial range, including the geographical distribution of risk, scope of influence, propagation path, etc. Analyzing spatial expansibility helps understand the potential impact of risk on different regions or fields. For each risk point, the system terminal combines historical data, industry trends, policy environment and other factors to predict its future development trend, determine the state changes of risk points at different time nodes, and the impact of these changes on the overall risk level. Assess the potential for risk spread and the scope of impact in different spatial scopes (such as regions, sectors, supply chains, etc.). Analyze how risk points spread through these spatial paths and the obstacles or acceleration factors that may be encountered during the spread. Then, based on the analysis results of temporal evolution and spatial scalability, reorder and expand the risk points in the first risk sequence. Classify risk points with similar temporal evolution characteristics or spatial expansion paths to form a more refined second risk sequence.The second risk sequence is a deepening and expansion of the first risk sequence, which more comprehensively reflects the complexity and dynamics of risk events. Finally, the system terminal integrates the first and second risk sequences and organizes them into a coherent risk list through time series analysis. The risk list is a comprehensive description and assessment of risk events, which includes information on the source, type, intensity, development trend, impact range, etc. Through the risk list, you can clearly understand the current risk situation and formulate an effective risk plan accordingly.

[0033] Based on the risk list, a risk plan is determined to conduct risk control management of compliance risks.

[0034] In one embodiment, the system terminal first obtains the various risk points listed in the risk list, and has a clear understanding of the nature, potential impact and possible development path of each risk. This step is the basis for ensuring that the risk plan can accurately correspond to the risk point. Subsequently, a detailed risk plan is formulated for each risk point. The risk plan should include risk prevention measures, monitoring mechanisms, emergency response plans and subsequent recovery strategies. Preventive measures are intended to reduce the possibility of risk occurrence, monitoring mechanisms are used to promptly detect risk signs, emergency response plans are intended to take prompt action when risks occur, and recovery strategies are intended to minimize the impact of risks on the business. These risk plans are set based on historical experience and expert advice. In the process of formulating risk plans, industry standards must be strictly followed to ensure the compliance of the plans. This includes reviewing all operating procedures, control measures and compliance requirements involved in the plan to ensure that they meet regulatory requirements and avoid possible compliance risks. After the risk plan is formulated, the system terminal incorporates it into the enterprise's risk control management system to ensure that the plan can be effectively implemented. This includes establishing a risk monitoring and reporting mechanism to ensure that risk conditions can be understood and decisions can be made in a timely manner.

[0035] Furthermore, the present application provides a positive serialization sorting for the collision service, including: Acquire parallel business data and perform business collision analysis; if there is a collision business and it meets the first collision level, temporarily divide the processing domain, wherein the processing domain includes block segmentation and block fusion of the multiple data format blocks, and prediction space division of the risk prediction model; if there is a collision business and it meets the second collision level, perform positive serialization sorting of the collision business based on business priority.

[0036] Preferably, the system terminal continuously acquires parallel business data, which may come from different business modules or systems. After obtaining these data, the system terminal performs collision analysis on these data, that is, checks whether there is a potential conflict between business processing. If a collision business is found, the magnitude of the collision will be evaluated. If the collision meets the first collision magnitude (that is, the conflict is more serious or the impact range is wider), a strategy of temporarily dividing the processing domain is adopted. The processing domain here not only includes block segmentation and block fusion at the data level, that is, dividing the data into different processing blocks according to the collision situation and merging them when necessary, but also involves the prediction space division of the risk prediction model, that is, adjusting the calculation scope and focus of the risk prediction model according to the risk assessment of the business collision. If the collision meets the second collision magnitude (that is, the conflict is lighter or can be resolved by adjusting the priority), the collision business will be positively serialized based on the business priority. This means that the system terminal will rearrange the processing order of the collision business according to the preset priority rules to ensure that key businesses can be processed first.

[0037] Furthermore, this application provides for compliance risk management, including Track the risk control effect and conduct risk control deviation analysis to determine the effect deviation data, where the risk control deviation includes the risk control data deviation and the risk control trend deviation; identify the effect deviation data, and if the risk control data deviation meets the first deviation threshold and the risk control trend deviation meets the second deviation threshold, perform deviation tracing and risk control feedback adjustment.

[0038] Preferably, in risk control, it is crucial to continuously track the effect of risk control, which helps to understand the gap between the actual effect of current risk control measures and the expected goal. In order to further optimize the risk control strategy, the system terminal conducts risk control deviation analysis, which aims to find out the deficiencies in the implementation of risk control. Specifically, the risk control deviation analysis focuses on the analysis of risk control data deviation and risk control trend bias. The risk control data deviation refers to the difference between the actual risk control data and the expected or standard data, which reflects the accuracy of the risk control measures at the specific execution level; while the risk control trend bias refers to the directional deviation of the risk control effect over time, which reveals the stability and adaptability of the risk control strategy in the long-term operation. When the risk control data deviation meets the first deviation threshold and the risk control trend also meets the second deviation threshold, deviation tracing will be performed. This step is to find out the specific reasons for the risk control deviation, which may be caused by data quality problems, defects in the risk control model, operational errors in the execution process, or changes in the external environment. On the basis of deviation tracing, risk control feedback adjustment is performed. This includes adjusting and optimizing existing risk control strategies, models or execution processes based on the results of deviation analysis to ensure that risk control measures can more accurately and effectively respond to potential risks. At the same time, a sound feedback mechanism should be established to timely discover and deal with new risk control deviation issues in the future, and continuously improve the level and effectiveness of risk control management.

[0039] In summary, the embodiments of the present application have at least the following technical effects: The embodiment of the present application uses interactive business features to retrieve and mine a risk feature library based on a risk inducement set, a risk type set, and a risk impact effect set in big data. Subsequently, the risk feature library is traversed, positive sample sequences and negative sample sequences are marked based on feature correlation, and sample polarization processing is performed to determine risk sample data. Using these risk sample data, the risk prediction model is supervised and trained under the constraints of the prediction phase. In the prediction stage, the target business data is read, and after data fuzzy screening, the data is vectorized using a data encoder containing multiple data format blocks to form a standardized data system. Afterwards, the system is input into the risk prediction model for risk feature matching and prediction, and a risk single column is generated. Based on the risk single column, the risk plan is determined and the risk control management of compliance risk is performed. In addition, business collision processing, risk control effect tracking and deviation analysis are also involved. In parallel business data processing, if a collision business is found and meets different collision magnitudes, temporary division of the processing domain or business priority sorting strategies will be adopted respectively. After the execution of risk control management, the system will track the risk control effect, analyze risk control deviations, including data deviations and trend deviations, and conduct deviation tracing and risk control feedback adjustments when necessary to continuously optimize the risk control management effect. These technical effects jointly solve the technical problems of inaccurate risk prediction and inefficient management due to the huge amount of business data and the complexity and variability of compliance risks, and realize the automatic generation of risk plans based on prediction results, improve the accuracy and timeliness of compliance risk prediction, enhance the pertinence and effectiveness of risk management, and reduce the degree of risk impact.

[0040] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0041] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0042] This specification and drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A big data-driven compliance risk prediction and management method, characterized by: The method comprises: Interactive business features, perform big data retrieval and mine risk feature library, wherein the risk feature library is based on risk cause set - risk type set - risk impact effect set; Traversing the risk feature library, marking positive sample sequences and negative sample sequences based on feature correlation, and performing sample polarization processing to determine risk sample data, wherein the positive sample sequence is a correlation feature pair, and the negative sample sequence is a non-correlation feature pair, wherein the feature correlation includes inter-set correlation and intra-set correlation; Based on the risk sample data, supervise the training of the risk prediction model with the prediction phase as a constraint; Read the target business data and perform data fuzzy screening, combine the data encoder to perform data vectorization, and determine the standardized data system, wherein the data encoder includes multiple data format blocks; Transmitting the standardized data system to the risk prediction model, performing risk feature matching and risk prediction, and generating a risk list; Based on the risk list, a risk plan is determined to conduct risk control management of compliance risks.

2. The big data driven compliance risk prediction and management method according to claim 1, characterized in that: The sample polarization process comprises: A loss excitation function is constructed using correlation as a first excitation feature and recognition probability as a second excitation feature, wherein the loss excitation function is used to determine the polarization direction and the polarization coefficient; In combination with the loss excitation function, polarization processing is performed on the positive sample sequence and the negative sample sequence.

3. The big data driven compliance risk prediction and management method according to claim 2, characterized in that: The sample polarization process comprises: Traversing the positive sample sequence, determining a maximization coefficient in combination with the loss excitation function, performing maximization processing on the matched positive sample sequence, and determining a first polarization sample; Traversing the negative sample sequence, determining a minimization coefficient in combination with the loss excitation function, performing minimization processing on the matched negative sample sequence, and determining a second polarization sample; The first polarization sample and the second polarization sample are integrated to determine the risk sample data.

4. The big data driven compliance risk prediction and management method according to claim 2, characterized in that: The polarization coefficient of the positive sample sequence is positively correlated with the first excitation feature and negatively correlated with the second excitation feature; the polarization coefficient of the negative sample sequence is positively correlated with the second excitation feature.

5. The big data driven compliance risk prediction and management method according to claim 1, characterized in that: The step of combining a data encoder to perform data vectorization includes: Obtain and filter business data, divide it based on data format, and determine multiple groups of business data; Traversing the multiple groups of business data, matching and vectorizing the multiple data format blocks, and determining vector data; Based on the data entity relationship, the vector data is integrated to determine the standardized data system.

6. The big data driven compliance risk prediction and management method according to claim 1, characterized in that: The generating of risk list includes: Perform risk feature matching to determine risk-related feature vectors; Traversing the risk-related feature vectors, performing feature coupling analysis, and determining a first risk sequence; Traversing the first risk sequence, and determining a second risk sequence based on temporal evolution and spatial expansibility, wherein the risk sequence is used to measure risk events; The first risk sequence and the second risk sequence are integrated in time series to generate the risk single sequence.

7. The big data driven compliance risk prediction and management method according to claim 1, characterized in that: The method further comprises: Obtain parallel business data and conduct business collision analysis; If there is a collision business and the first collision level is met, temporarily dividing the processing domain, wherein the processing domain includes block segmentation and block fusion of the multiple data format blocks, and prediction space division of the risk prediction model; If there is a collision service and the second collision level is satisfied, the collision service is sorted in a positive serialization manner based on the service priority.

8. The big data driven compliance risk prediction and management method according to claim 1, characterized in that: After conducting compliance risk control management, it includes: Track the risk control effect and conduct risk control deviation analysis to determine the effect deviation data, where risk control deviation includes risk control data deviation and risk control trend deviation; The effect deviation data is identified. If the risk control data deviation satisfies a first deviation threshold and the risk control trend deviation satisfies a second deviation threshold, deviation tracing and risk control feedback adjustment are performed.

Images