Feature recognition management method and system for electronic payment

Abstract

This application provides a method and system for feature recognition management in electronic payments, relating to the field of feature recognition technology. The method includes: parsing the feature recognition relationships of electronic payment features, including risk relationships and security lock-in relationships; performing feature decomposition based on security lock-in relationships, evaluating the loss and confidence levels of the decomposed features, and configuring a recognition dispersion index for risk dispersion features; performing risk feature recognition compensation based on risk relationships, and configuring a recognition compensation index for risk compensation relationship features; and matching the processing execution parameters of the recognition features according to the recognition dispersion index and recognition compensation index to manage the recognition features. This application solves the technical problem in existing technologies where the traditional model uses indiscriminate full-data collection and static processing strategies, resulting in excessive exposure of sensitive data and low feature recognition security. By dynamically assessing feature risks and dynamically selecting feature processing strategies, the security of feature recognition is improved.

Description

Technical Field

[0001] This application relates to the field of feature recognition technology, and in particular to a feature recognition management method and system for electronic payments. Background Technology

[0002] Currently, the feature recognition in the electronic payment field generally adopts the traditional model of indiscriminately collecting and centrally verifying all user features. While this can ensure the security of the payment process to a certain extent, its static and crude nature leads to an excessive exposure of sensitive data, resulting in significant data leakage and privacy risks. Moreover, traditional security verification methods usually rely on fixed feature recognition processes and lack the ability to dynamically assess and adjust for real-time risks in the transaction environment. In particular, verification measures are insufficiently protective in high-risk scenarios, while excessive intervention occurs in low-risk scenarios, causing an imbalance between security and user experience.

[0003] In summary, existing technologies suffer from technical problems due to the traditional approach of indiscriminate full data collection and static processing, which leads to excessive exposure of sensitive data, low security of feature recognition, and failure to effectively address dynamic risks in the trading environment, further affecting the efficiency of feature recognition and processing. Summary of the Invention

[0004] The purpose of this application is to provide a feature recognition management method and system for electronic payments, in order to solve the technical problems in the prior art, which are that the traditional mode adopts indiscriminate full data collection and static processing strategies, resulting in an excessive exposure of sensitive data, low feature recognition security, and failure to effectively deal with dynamic risks in the transaction environment, which further affects the efficiency of feature recognition processing.

[0005] In view of the above problems, this application provides a feature recognition management method and system for electronic payments.

[0006] Firstly, this application provides a feature recognition management method for electronic payments. This method is implemented through a feature recognition management system for electronic payments. The method includes: parsing the feature recognition relationships of electronic payment features, including risk relationships and security lock-in relationships; performing feature decomposition based on the security lock-in relationships, evaluating the loss amount and confidence level of the decomposed features, and configuring a risk dispersion index for risk dispersion features; performing risk feature recognition compensation based on the risk relationships, configuring a risk compensation relationship feature recognition compensation index according to the compensated risk safety coefficient; and matching the processing execution parameters of the recognition features based on the recognition dispersion index and the recognition compensation index to perform recognition management of the recognition features.

[0007] Optionally, multiple features collected throughout the entire electronic payment cycle are acquired, including one or more combinations of biometrics, behavioral features, device features, and environmental features; security analysis is performed on the inherent attributes of each feature to obtain a static security level, wherein the inherent attributes include the uniqueness, immutability, difficulty of copying, and privacy sensitivity of the feature; based on the static security level, risk analysis is performed in conjunction with the dynamic context of each feature in the electronic payment process to obtain a dynamic risk coefficient, wherein the dynamic context includes one or more of transaction time, transaction location, transaction amount, transaction object, and network environment; a feature risk profile is established based on the static security level and the dynamic risk coefficient, and feature identification relationships are determined based on the feature risk profile; wherein, feature identification relationships where the static security level meets the preset security requirements and the dynamic risk coefficient is less than the risk threshold are determined as the security locking relationship; feature identification relationships where the dynamic risk coefficient is greater than or equal to the risk threshold are determined as the risk relationship.

[0008] Optionally, a feature security evaluation matrix is ​​established, including uniqueness weight, immutability weight, copying difficulty weight, and privacy sensitivity weight; each feature is scored on four dimensions—uniqueness, immutability, copying difficulty, and privacy sensitivity—based on its attribute characteristics to obtain dimension score values; based on the dimension score values ​​and their corresponding weights, a comprehensive security evaluation value for each feature is calculated through weighted average; and static security levels are classified according to the range of the comprehensive security evaluation values.

[0009] Optionally, multiple predefined risk factors are extracted from the dynamic context. These risk factors include transaction time anomaly, geographical location anomaly, transaction amount anomaly, counterparty unfamiliarity, and network environment risk. Historical normal transaction data and fraudulent transaction data are collected to form a training sample set, including a positive example sample set and a negative example sample set. Using the risk factors as input features and transaction risk labels as outputs, the model is trained and converged using machine learning methods based on the positive and negative example sample sets to obtain a dynamic risk assessment model. The dynamic context of each feature in the electronic payment process is injected into the dynamic risk assessment model, and the dynamic risk coefficient is output.

[0010] Optionally, feature data of historical risk samples and known attack patterns are obtained as an adversarial sample set; the adversarial sample set is input into an initial risk identification model based on feature risk profile for adversarial training, and key adversarial risk features are extracted through model iteration optimization; the static security level and dynamic risk weight are corrected based on the adversarial risk features, and the feature risk profile is reconstructed.

[0011] Optionally, a generative adversarial network architecture is used, wherein the generator is used to generate forged feature data simulating attacks, and the discriminator is the initial risk identification model; through the game-like adversarial between the generator and the discriminator, the discriminator's ability to identify fraudulent features is continuously strengthened, and the key adversarial risk features that most affect the discrimination result are identified in the adversarial process.

[0012] Optionally, the original feature data of the security locking relationship is decomposed into multiple feature segments; dispersed features are extracted based on the multiple feature segments to construct multiple dispersed feature combinations; the recognition rate is compared with the original feature data using the dispersed feature combinations to obtain the recognition loss and recognition confidence difference; the dispersed feature combination recognition evaluation is performed based on the recognition loss and recognition confidence difference to obtain the recognition dispersion index.

[0013] Optionally, for the features of the risk relationship, one or more of the following three compensation strategies are selected: multi-feature assisted identification, dynamic verification enhancement, and behavioral biometric fusion. After the risk relationship features are processed by the compensation strategy, dynamic context is used for risk analysis to evaluate the risk safety coefficient of the selected compensation strategy. The risk safety coefficient represents the degree of improvement in feature recognition security after the compensation strategy is adopted.

[0014] Optionally, based on the identification dispersion index and the identification compensation index, combinations of dispersed features or compensation strategies that meet the identification security requirements are selected, and feature processing execution parameters, including dispersed identification features and compensated identification features, are obtained; the identification management adjustment of the corresponding features is performed using the feature processing execution parameters; wherein the identification security requirements are adaptively set according to the security requirement level of the electronic payment scenario.

[0015] Secondly, this application also provides a feature recognition management system for electronic payments, used to execute the feature recognition management method for electronic payments as described in the first aspect, wherein the feature recognition management system for electronic payments includes: a relationship parsing module, used to parse the feature recognition relationships of electronic payment features, including risk relationships and security lock relationships; a feature decomposition module, used to perform feature decomposition based on the security lock relationships, evaluate the loss amount and confidence level of the decomposed features, and configure the recognition dispersion index of risk dispersion features; a recognition compensation module, used to perform risk feature recognition compensation according to the risk relationships, and configure the recognition compensation index of risk compensation relationship features according to the compensated risk security coefficient; and a parameter matching module, used to match the processing execution parameters of the recognition features according to the recognition dispersion index and the recognition compensation index, and perform recognition management of the recognition features.

[0016] One or more technical solutions provided in this application have at least the following beneficial effects: By analyzing the feature recognition relationships of electronic payment features, including risk relationships and security locking relationships; performing feature decomposition based on the security locking relationships, evaluating the loss amount and confidence level of the decomposed features, and configuring a risk dispersion index for risk dispersion features; performing risk feature recognition compensation based on the risk relationships, and configuring a risk compensation relationship feature recognition compensation index according to the compensated risk security coefficient; and matching the processing execution parameters of the recognition features based on the recognition dispersion index and recognition compensation index to manage the recognition features. In other words, by dynamically selecting feature processing strategies for feature recognition based on the feature's own security risk level, confidence level, and contextual risk in the current payment transaction, the security of feature recognition is improved, while the processing efficiency of payment features is also increased.

[0017] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the feature recognition management method for electronic payments used in this application.

[0020] Figure 2 This is a schematic diagram of the feature recognition management system for electronic payments used in this application.

[0021] Figure labeling: Relationship parsing module 11, Feature decomposition module 12, Recognition compensation module 13, Parameter matching module 14. Detailed Implementation

[0022] This application provides a feature recognition management method and system for electronic payments, solving the technical problems in existing technologies where the traditional approach of indiscriminate full data collection and static processing leads to excessive exposure of sensitive data, low feature recognition security, and an inability to effectively address dynamic risks in the transaction environment, further impacting feature recognition processing efficiency. By dynamically selecting feature processing strategies based on the feature's inherent security risk level, confidence level, and contextual risk in the current payment transaction, the application improves both feature recognition security and payment feature processing efficiency.

[0023] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0024] Example 1, please refer to the appendix. Figure 1 This application provides a feature recognition management method for electronic payments, wherein the feature recognition management method for electronic payments is executed through a feature recognition management system for electronic payments, and the feature recognition management method for electronic payments specifically includes the following steps:

[0025] Analyze the feature recognition relationships of electronic payment features, including risk relationships and security lock-in relationships.

[0026] Furthermore, this application also includes the following steps: acquiring multiple features collected during the entire electronic payment cycle, including one or more combinations of biometric features, behavioral features, device features, and environmental features; performing security analysis on the inherent attributes of each feature to obtain a static security level, wherein the inherent attributes include the uniqueness, immutability, copying difficulty, and privacy sensitivity of the feature; based on the static security level, performing risk analysis in conjunction with the dynamic context of each feature in the electronic payment process to obtain a dynamic risk coefficient, wherein the dynamic context includes one or more of transaction time, transaction location, transaction amount, transaction object, and network environment; establishing a feature risk profile based on the static security level and dynamic risk coefficient, and determining feature identification relationships based on the feature risk profile; wherein feature identification relationships where the static security level meets the preset security requirements and the dynamic risk coefficient is less than the risk threshold are determined as the security locking relationship; feature identification relationships where the dynamic risk coefficient is greater than or equal to the risk threshold are determined as the risk relationship.

[0027] Furthermore, this application also includes the following steps: establishing a feature security evaluation matrix, including uniqueness weight, immutability weight, copying difficulty weight, and privacy sensitivity weight; scoring each feature according to its attribute characteristics on four dimensions: uniqueness, immutability, copying difficulty, and privacy sensitivity, to obtain dimension score values; calculating the comprehensive security evaluation value of each feature by weighting based on the dimension score values ​​and their corresponding weights; and classifying static security levels according to the range of the comprehensive security evaluation values.

[0028] Furthermore, this application also includes the following steps: extracting multiple predefined risk factors from the dynamic context, the risk factors including transaction time anomaly, geographical location anomaly, transaction amount anomaly, counterparty unfamiliarity, and network environment risk; collecting historical normal transaction data and fraudulent transaction data to form a training sample set, including a positive example sample set and a negative example sample set; using the risk factors as input features and transaction risk labels as outputs, training and converging the model based on the positive example sample set and the negative example sample set using machine learning methods to obtain a dynamic risk assessment model; injecting the dynamic context of each feature in the electronic payment process into the dynamic risk assessment model, and outputting the dynamic risk coefficient.

[0029] Specifically, during the electronic payment process, the payment platform acquires various data from the user's biometrics, behavioral characteristics, device characteristics, and environmental characteristics. These characteristics are continuously collected and updated in real time throughout the entire electronic payment cycle. The entire electronic payment cycle is the entire process from when the user initiates a payment to when the transaction is completed. The payment system needs to perform multi-dimensional identification and verification of the user's identity, device, and transaction environment to ensure the security and legality of the transaction.

[0030] Biometrics are identity verification information related to a user's physiological characteristics, such as fingerprints, facial recognition, iris scanning, and voiceprint recognition. They are highly unique and difficult to forge, and are therefore widely used in identity verification. Behavioral characteristics are based on the user's behavioral patterns during the payment process, such as keystroke speed, mouse trajectory, and touchscreen swiping habits. They are dynamic and can reflect the user's identity and be used to monitor abnormal transaction behavior. Device characteristics are characteristics related to the user's device, such as device ID, device model, operating system, device fingerprint, device hardware identifier, and browser information. They are used to identify the user's device and increase transaction security. Environmental characteristics are characteristics related to the user's environment, including the user's geographical location and network environment. They are used to determine the user's actual location and identify whether there is any risky behavior, such as making payments in abnormal areas.

[0031] Generally, electronic payment systems do not rely solely on a single type of feature, but rather enhance identity verification during the payment process through a combination of multiple features. For example, combining a user's biometrics, device characteristics, and behavioral characteristics can improve the security of identity verification. Different combinations of features will generate different risk assessments in different payment scenarios. For instance, a combination of biometrics and device characteristics may be considered sufficiently secure for low-risk transactions, but for high-risk transactions, more features, such as behavioral and environmental characteristics, need to be combined for a more detailed risk analysis.

[0032] The inherent attributes of features are the essential characteristics of various features used in the identification process, determining the security and usability of each feature. These include the uniqueness, immutability, difficulty of replication, and privacy sensitivity of the feature. Uniqueness indicates whether a feature can uniquely identify an individual; immutability indicates whether a feature is easily tampered with or forged; difficulty of replication indicates whether a feature is easily copied or forged; and privacy sensitivity indicates the degree of sensitivity of the feature in terms of privacy protection.

[0033] Security analysis is performed on the inherent attributes of each feature to identify its performance across four dimensions, resulting in a dimensional score. For example, fingerprints and facial recognition have high uniqueness and may score 5 points; while passwords or simple PIN codes can be stolen and tampered with, so they may score 2 points; biometric features such as fingerprints and irises are difficult to replicate, so they score higher, such as 5 points.

[0034] Based on the dimensional scores and their corresponding weights, a weighted average is used to calculate the overall security evaluation value for each feature. First, a weight is assigned to each inherent attribute to reflect the importance of each dimension to the overall security assessment. For example, uniqueness has a weight of 0.3, immutability 0.4, duplication difficulty 0.2, and privacy sensitivity 0.1. The performance of each feature across each dimension is evaluated, and a score is given. For example, for a fingerprint feature with a uniqueness score of 5, immutability score of 5, duplication difficulty score of 5, and privacy sensitivity score of 4, the weighted average gives the feature an overall security evaluation value of 5*0.3 + 5*0.4 + 5*0.2 + 4*0.1 = 4.9.

[0035] Static security levels are defined based on the range of comprehensive security evaluation values. In the electronic payment process, this means that each feature is classified into a security level based on its inherent attributes during the security evaluation. Depending on the comprehensive security evaluation value, features may be categorized into different security levels, such as high security, medium security, and low security. For example, a comprehensive security evaluation value > 4.5 indicates high security, 3.5 < 4.5 indicates medium security, and a comprehensive security evaluation value ≤ 3.5 indicates low security.

[0036] In electronic payments, the security of user characteristics is affected by a variety of factors, including not only the user's identity but also the real-time environment of the transaction. Dynamic context refers to various real-time, environment-related information during the transaction process, including one or more of the following: transaction time, transaction location, transaction amount, transaction counterparty, and network environment. Transaction time is the specific time the transaction occurred, usually represented by a timestamp; transaction location is the geographical location where the transaction took place, typically obtained through the device's GPS or IP address. For example, a user making a payment far from their usual residence may be considered unusual; transaction amount is the size of the transaction; larger transaction amounts are generally considered to have higher risk; transaction counterparty is the other party in the transaction; and network environment refers to the network environment used for the transaction, particularly whether it is on public Wi-Fi or whether there is a risk of cyberattacks.

[0037] By employing various technical means during the payment process, dynamic contextual information related to the transaction is obtained. Multiple predefined risk factors are extracted from this dynamic context, including transaction time anomaly, geographical location anomaly, transaction amount anomaly, unfamiliarity of the counterparty, and network environment risk. These risk factors are potential risk indicators extracted from the dynamic context, used to quantify the degree of transaction anomaly. For example, if a user conducts a transaction at 11 PM on a holiday, the time is recorded; if the transaction location is determined via GPS or IP address, and the user transacts in a location significantly different from their usual residence, this information is recorded; if a user makes a large transaction, the amount is used as dynamic context data; if a user transacts with a merchant for the first time, and this merchant may be new or unfamiliar, the counterparty's information is recorded; and whether the network used for the transaction is an insecure network, such as public Wi-Fi, is also considered.

[0038] Historical normal and fraudulent transaction data are collected to construct a training sample set, including positive and negative sample sets. Historical normal transaction data refers to user transaction records under normal trading conditions over a past period, while fraudulent transaction data refers to transaction data where fraudulent activities occurred. Using risk factors as input features and transaction risk labels as outputs, the model is trained using machine learning methods based on the positive and negative sample sets. Transaction risk labels include transaction type and transaction risk coefficient, corresponding one-to-one with historical normal and fraudulent transaction data. Using Support Vector Machines (SVMs) as an example, a classic classification model, binary classification is performed by finding a hyperplane that maximizes the margin. Since different features have different dimensions and value ranges, the data in the training sample set first needs to be standardized or normalized so that the values ​​of each feature are within the same range, such as 0 to 1.

[0039] The training sample set is divided into an 80% training set and a 20% test set. The training set is used to train the model, and the test set is used to evaluate the model's performance. Using the training set data, the model is trained using the SVM algorithm. SVM maximizes the margin between sample points by finding the decision boundary, aiming to distinguish between legitimate and fraudulent transactions. During training, cross-validation is used to tune hyperparameters. After training, the model is validated using the test set, and metrics such as accuracy, precision, recall, and F1 score are calculated. For example, assuming the model has an accuracy of 98%, a recall of 95%, a precision of 92%, and an F1 score of 94%. Based on the test results, if the accuracy is low, further adjustments are made to the model parameters or optimizations such as feature selection and dimensionality reduction are performed. Once the model training is complete, the trained dynamic risk assessment model is used to evaluate each new transaction.

[0040] In each new electronic payment transaction, the dynamic context of the current transaction is automatically extracted, and multiple risk factors are extracted from the dynamic context and input into a dynamic risk assessment model for risk analysis to obtain a dynamic risk coefficient. For example, if the transaction occurs during the user's normal time, the transaction amount is small, and the transaction partner is a familiar merchant, the model may output a low risk coefficient, such as 0.1. If the transaction occurs at an unusual time, such as 3 a.m., the transaction amount is unusually high, and an abnormal network is used, the model may output a higher risk coefficient, such as 0.8.

[0041] The dynamic risk coefficient is a key indicator used in risk assessment, representing the level of risk in a current transaction. It is typically compared to a certain risk threshold to determine whether the transaction carries risk. In electronic payments, each feature has its inherent static security level and a dynamic risk coefficient in a specific payment scenario. By combining the static security level and the dynamic risk coefficient, a feature risk profile can be created for each feature. A feature risk profile is a comprehensive description considering both the static security level and the dynamic risk coefficient. For example, for biometric features such as fingerprints, with a high static security level of 0.9 and a low dynamic risk coefficient of 0.2, its feature risk profile would be 0.9, 0.2.

[0042] Based on the feature-based risk profile, the identification relationship of each feature is further determined. In this process, it is determined whether each feature belongs to a secure locking relationship or a risky relationship. If a feature's static security level has reached the preset security requirements (e.g., a static security level greater than or equal to 0.8) and its dynamic risk coefficient is less than the preset risk threshold (e.g., a dynamic risk coefficient < 0.5), then the feature's identification relationship is determined to be a secure locking relationship, and the feature's security is guaranteed. Excessive dynamic verification is not required, and it can be directly used for identification. If a feature's dynamic risk coefficient is greater than or equal to the risk threshold (e.g., a dynamic risk coefficient ≥ 0.5), meaning the feature exhibits high risk in the current trading environment, then the feature's identification relationship is determined to be a risky relationship. More stringent verification or other preventative measures are required for this feature. Features with secure locking relationships and risky relationships are categorized and processed accordingly: for secure locking relationships, the feature can be directly trusted without complex verification; for risky relationships, stronger verification methods are needed, such as multi-factor verification and dynamic risk assessment. For example, if the feature type is a biometric feature, such as a fingerprint, its static security level is 0.9, its dynamic risk coefficient is 0.2, its feature risk profile is [0.9, 0.2], and its identification relationship is a secure lock relationship; if the feature type is a behavioral feature, such as a sliding trajectory, its static security level is 0.6, its dynamic risk coefficient is 0.7, its feature risk profile is [0.6, 0.7], and its identification relationship is a risk relationship; if the feature type is a device feature, such as a device ID, its static security level is 0.5, its dynamic risk coefficient is 0.8, its feature risk profile is [0.5, 0.8], and its identification relationship is a risk relationship.

[0043] By combining static security levels and dynamic risk coefficients, high-risk transactions are accurately identified, and verification measures are strengthened as needed, reducing the risk of data leakage and identity theft. For low-risk transactions, verification steps are reduced to improve the efficiency of the payment process; while for high-risk transactions, additional verification is triggered promptly to ensure transaction security.

[0044] Furthermore, this application also includes the following steps: obtaining feature data of historical risk samples and known attack patterns as an adversarial sample set; inputting the adversarial sample set into an initial risk identification model based on feature risk profile for adversarial training, and extracting key adversarial risk features through model iterative optimization; and reconstructing the feature risk profile by correcting the static security level and dynamic risk weight based on the adversarial risk features.

[0045] Furthermore, this application also includes the following steps: using a generative adversarial network architecture, wherein the generator is used to generate forged feature data simulating attacks, and the discriminator is the initial risk identification model; through the game-like adversarial between the generator and the discriminator, the ability of the discriminator to identify fraudulent features is continuously strengthened, and the key adversarial risk features that most affect the discrimination result are identified in the adversarial process.

[0046] Specifically, historical risk samples are sample data from past electronic payment transactions that have been identified as normal or fraudulent. These include genuine transaction records, categorized by labels indicating whether the transaction is normal or fraudulent; positive examples represent normal transactions, while negative examples represent known fraudulent transactions. Feature data representing known attack patterns refers to feature data collected using known attack patterns to simulate potential risks and vulnerabilities. By extracting features from historical data that indicate fraudulent transactions and combining them with normal transaction data, an adversarial sample set is formed. This adversarial sample set typically includes genuine feature samples from normal transactions and forged feature samples from malicious attacks.

[0047] Generative Adversarial Networks (GANs) are a machine learning architecture consisting of two main components: a generator and a discriminator. These two components compete against each other in a game-like process, continuously improving the model's performance. The generator receives random noise and generates fake feature data simulating attacks, such as forged biometrics, device features, and behavioral characteristics, with the aim of deceiving the discriminator. The discriminator, based on an initial risk identification model, determines whether the input feature data comes from real transaction samples or is attack data forged by the generator. The discriminator's goal is to identify fraudulent transaction data and distinguish it from legitimate transactions. During training, the generator and discriminator are trained through a game-like competition. The generator strives to generate more realistic fake data, while the discriminator strives to distinguish between fake and real data. As training progresses, the generator gradually learns to generate more deceptive fake features, while the discriminator becomes increasingly adept at identifying such fake data. Specifically, the generator receives random noise input and generates fake feature data. The discriminator receives real and fake samples as input and outputs its judgment of whether a sample is real or fake. The discriminator calculates the loss based on the output and the actual results, and performs backpropagation. The generator optimizes its generation strategy based on feedback from the discriminator, gradually improving the quality of the forged data. This process is repeated multiple times until the forged data generated by the generator becomes very close to the real samples, while the discriminator becomes increasingly sensitive, able to identify subtle fraudulent features. Through repeated game training, both the generator and the discriminator continuously improve their performance. Ultimately, the discriminator will be able to identify which features are most important for distinguishing between normal and forged samples—the key adversarial risk features.

[0048] During iterative training, the generator gradually produces more deceptive feature data, while the discriminator continuously optimizes its recognition capabilities. Through iterative optimization of the model, key adversarial risk features become more apparent. Once the adversarial network is trained and key adversarial risk features are extracted from the training, the original feature risk profile can be optimized and corrected based on these features. Some features may be found to be more vulnerable to attack during training, so their static security level may need to be lowered. For example, if facial recognition features are frequently forged and successfully deceive the discriminator, then the static security level of that feature should be reduced. Through adversarial training, it is discovered that some features are more sensitive in high-risk scenarios, thus requiring an increase in their dynamic risk coefficient. For instance, in situations with high network risk, the dynamic risk coefficient of certain features may need to be increased to enhance their protection. Through these corrections, the feature risk profile can be reconstructed, enabling a more accurate assessment of the security and risk of each feature in different transaction scenarios, and thus allowing for appropriate verification and processing.

[0049] By employing a generative adversarial network (GAN) training method, the model's recognition capabilities were significantly improved, particularly in handling unknown and variant fraudulent behaviors with greater flexibility and accuracy. Through multi-round game-based training, the generator and discriminator were continuously optimized, making the model's identification of fraudulent transactions more precise. The false positive and false negative rates for fraud detection were greatly reduced, improving the user experience.

[0050] Based on the security locking relationship, feature decomposition is performed, and the loss amount and confidence of the decomposed features are evaluated. The risk dispersion index is then configured to identify the dispersion features.

[0051] Furthermore, this application also includes the following steps: decomposing the original feature data of the security locking relationship into multiple feature segments; extracting dispersed features based on the multiple feature segments to construct multiple dispersed feature combinations; comparing the recognition rate of the dispersed feature combinations with the original feature data to obtain the recognition loss and recognition confidence difference; and evaluating the recognition of the dispersed feature combinations based on the recognition loss and recognition confidence difference to obtain the recognition dispersion index.

[0052] Specifically, secure locking relationships are feature recognition relationships related to low-risk transactions obtained through security analysis during the electronic payment feature identification process. The original feature data of secure locking relationships is decomposed, with each original feature data being split into multiple smaller feature fragments, each containing a portion of the original feature information. By decomposing the feature data, the risk of data leakage can be reduced, and the model's resistance to attacks can be improved.

[0053] Dispersed feature extraction is performed based on multiple feature segments, considering various dimensions such as density, quantity, proportion, step size, and iterative search, resulting in multiple combinations of dispersed features. Dispersed feature combination involves combining feature information extracted from different feature segments to construct multiple different feature combination methods based on different metrics, such as feature density, quantity, proportion, and step size, forming various different combination methods.

[0054] The recognition rate is compared between multiple combinations of dispersed features and the original feature data. The recognition rate refers to the proportion of correctly identified features. The recognition loss represents the degree of decrease in recognition accuracy when using combinations of dispersed features; a lower loss indicates a smaller impact of the dispersed feature combinations on recognition accuracy. The recognition confidence difference represents the change in confidence level caused by the combination of dispersed features during feature recognition, indicating the change in the confidence level of the recognition result. Confidence level refers to the model's degree of certainty about the result; a smaller difference indicates higher stability of the recognition result.

[0055] The identification dispersion index is obtained by evaluating the dispersion of features based on the identification loss and identification confidence difference. A higher identification dispersion index indicates that the dispersion effect of the combination is good, with low loss and high stability. The identification dispersion index combines the identification loss and identification confidence difference to evaluate the security and accuracy of feature dispersion. By decomposing features and performing dispersion extraction, the risk of single feature leakage is reduced; through multiple combinations and comparisons, the optimal dispersion feature combination is selected, which can enhance the ability to identify fraudulent behavior while maintaining high accuracy. By optimizing the identification loss and confidence difference, the identification results are ensured to have high stability, reducing false positives and false negatives.

[0056] Based on the aforementioned risk relationships, risk characteristics are identified and compensated. According to the risk safety coefficient of the compensation, the identification and compensation index of the risk compensation relationship characteristics is configured.

[0057] Furthermore, this application also includes the following steps: for the features of the risk relationship, select one or more combinations from three compensation strategies: multi-feature assisted identification, dynamic verification enhancement, and behavioral biometric fusion; perform feature compensation processing on the risk relationship features using the compensation strategy, and then perform risk analysis in dynamic context to evaluate the risk safety coefficient of the selected compensation strategy, wherein the risk safety coefficient represents the degree of improvement in feature recognition security after adopting the compensation strategy.

[0058] Specifically, when an electronic payment system identifies certain characteristics as high-risk, compensation strategies are needed to enhance its identification capabilities. One or more of the following three compensation strategies can be selected and combined: multi-feature-assisted identification, dynamic verification enhancement, and behavioral biometric fusion. Multi-feature-assisted identification improves the accuracy of risk feature identification by introducing additional auxiliary features. For example, if a transaction's device characteristics already indicate a high risk, supplementary biometric features can be used to assist identification. Combining data from multiple features improves identification accuracy. Dynamic verification enhancement combines real-time transaction behavior data for dynamic verification, thereby improving the ability to identify risk features. For example, the time and location of a transaction may differ from a user's usual patterns, indicating potential fraud. Dynamic verification strategies collect and analyze transaction behavior features in real time, such as IP address, transaction time, and location, and compare them with historical data to enhance risk identification. Behavioral biometric fusion combines multiple biometric features with behavioral features to enhance user identity verification. For example, a user's fingerprint information can be combined with typing speed and mouse click patterns. By combining biometrics and behavioral patterns, the accuracy of fraud detection can be effectively improved, especially in high-risk scenarios such as large transactions or logins from unfamiliar devices.

[0059] According to the compensation strategy, feature compensation processing is applied to the characteristics of risk relationships, compensating for certain high-risk features. Taking an online payment scenario as an example, suppose a user's transaction amount differs significantly from historical transaction amounts, the system will identify the transaction as potentially fraudulent. To reduce the risk of false positives, a dynamic verification enhancement strategy is activated, checking the user's login location and device in real time, or requiring secondary verification. Dynamic contextual risk analysis is performed on the compensated features, calculating a risk safety coefficient. This coefficient represents the degree of security improvement in feature recognition after adopting the compensation strategy. If the recognition accuracy improves after compensation, the compensation strategy is effective; if the compensation fails to achieve the expected results, the compensation strategy may need to be readjusted. For example, suppose that after applying multi-feature assisted recognition compensation to a transaction, the recognition accuracy increases from 90% to 95%, while the false positive rate decreases from 5% to 2%. In this case, the risk safety coefficient is high, indicating that the compensation strategy successfully improved security. The risk safety coefficient measures the degree of security improvement in feature recognition after applying the compensation strategy; the higher the risk safety coefficient, the stronger the ability to identify risky features and the higher the security after adopting the compensation strategy.

[0060] After evaluating the effectiveness of the compensation strategy, the identification compensation index of the risk compensation relationship features is configured based on changes in the risk security coefficient. The identification compensation index is a quantitative indicator reflecting the effectiveness of the compensation strategy in improving identification security. For example, if the risk security coefficient is improved by 20% through the implementation of a dynamic verification enhancement strategy, then the identification compensation index of this strategy is 0.2. A higher identification compensation index indicates that the compensation strategy is effective and can continue to be used in high-risk transactions. Adjustments are made based on changes in the identification compensation index. If the identification compensation index of the compensation strategy is high, its identification mechanism is further optimized to automatically adapt to different payment scenarios. For example, in scenarios involving large transactions and logins from unfamiliar devices, a behavioral biometric fusion strategy is preferred, while for small payments and transactions from known devices, a multi-feature assisted identification strategy is chosen.

[0061] By introducing compensation strategies, the most suitable identification strategy can be selected based on different risk characteristics, thereby improving the overall identification accuracy and security. For example, combining multi-feature-assisted identification and dynamic verification enhancement can effectively reduce the false alarm rate and false negative rate.

[0062] Based on the identification dispersion index and identification compensation index, the processing execution parameters of the identification features are matched to perform identification management on the identification features.

[0063] Furthermore, this application also includes the following steps: based on the identification dispersion index and the identification compensation index, screening for dispersion feature combinations or compensation strategies that meet the identification security requirements, obtaining feature processing execution parameters, including dispersion identification features and compensation identification features; using the feature processing execution parameters to perform identification management adjustments for the corresponding features; wherein the identification security requirements are adaptively set according to the security requirement level of the electronic payment scenario.

[0064] Specifically, security requirements are determined based on factors such as payment amount, user risk assessment, and device environment. Based on the identification dispersion index and identification compensation index, dispersed feature combinations or compensation strategies that meet the security requirements are selected from the evaluated feature combinations. Appropriate feature combinations are chosen based on the identification dispersion index; the purpose of feature dispersion is to reduce the potential risks arising from feature concentration. Feature combinations with high identification dispersion indices are given priority for further security assessment. Combining the identification compensation index, feature identification strategies with good compensation effects are selected. For example, if certain features perform poorly in high-risk transaction scenarios, compensation strategies such as dynamic verification strategies and behavioral biometric fusion are introduced to improve security.

[0065] From the selected combinations of dispersed features and compensation strategies, a set of feature processing execution parameters is generated, specifying the feature combinations and compensation strategies to be applied. In other words, once a feature combination or compensation strategy that meets the security requirements is determined, corresponding feature processing execution parameters are set for these features, including the specific configuration of the dispersed feature combination or compensation strategy, i.e., the selected dispersed identification features and compensated identification features. Feature processing execution parameters are used for actual feature recognition management, determined based on the identification dispersion index and identification compensation index, including the required feature combination, compensation strategy, and related algorithm settings. For example, for a cross-border payment of 5000 yuan using a new device, the transaction is assessed as a high-risk scenario with a high security requirement level. A compensation strategy based on device features, behavioral features, and dynamic verification codes is selected. The compensation strategy chosen is face recognition and device fingerprint recognition. The feature processing execution parameters are device ID, behavioral features, and dynamic verification codes. Payment security is significantly improved, with an identification accuracy of 99.98%. The user verification process is relatively smooth, but due to the adoption of strong verification measures, the identification latency is slightly increased, and the average user waiting time increases by approximately 3 seconds.

[0066] Configure relevant parameters for feature recognition based on different security requirements. For example, in low-risk payment scenarios, only basic device feature recognition and behavioral feature verification are needed. In high-risk payment scenarios, multiple verifications may be required, such as dynamic CAPTCHAs, biometrics, and multi-dimensional analysis of behavioral features. Once the feature processing execution parameters are configured, adjust the actual feature recognition management based on these parameters. Evaluate the weight of each feature; for example, in low-risk scenarios, device ID recognition has a higher weight, while in high-risk scenarios, the weight of dynamic CAPTCHAs and biometrics may increase. Select a recognition strategy, such as static verification based on device information, or a dynamic method combining behavioral analysis and real-time verification. Adjust the sensitivity of the recognition algorithm, such as increasing the threshold of biometric technology, performing secondary verification in certain situations, or adjusting the model's judgment criteria through machine learning to improve the accuracy of a specific feature.

[0067] The system monitors the effectiveness of feature recognition in real time during the payment process. If a recognition error occurs at any stage, it automatically adjusts based on pre-defined execution parameters or activates compensation strategies to enhance recognition capabilities. If an anomaly is detected in the user's device during recognition, such as an unregistered device, compensation strategies are triggered, such as requiring SMS verification or dynamic password verification. Based on feedback from the feature recognition results, the feature recognition strategy is automatically adjusted. For features with poor recognition performance, the weight of that feature is reduced or other compensation strategies are introduced. Through feature recognition management adjustments, the feature recognition strategy can be flexibly adjusted according to the specific circumstances of the transaction, enhancing security in high-risk payment scenarios and effectively reducing the occurrence of fraud.

[0068] In summary, the feature recognition management method for electronic payments provided in this application has the following beneficial effects: It analyzes the feature recognition relationships of electronic payment features, including risk relationships and security lock-in relationships; it decomposes features based on the security lock-in relationships, evaluates the loss amount and confidence level of the decomposed features, and configures a risk dispersion index for risk dispersion features; it performs risk feature recognition compensation based on the risk relationships, and configures a risk compensation index for risk compensation relationship features according to the compensated risk security coefficient; and it matches the processing execution parameters of the recognition features based on the recognition dispersion index and the recognition compensation index to manage the recognition features. In other words, by dynamically selecting feature processing strategies for feature recognition based on the feature's own security risk level, confidence level, and contextual risk in the current payment transaction, the security of feature recognition is improved, while the processing efficiency of payment features is also increased.

[0069] Example 2: Based on the same inventive concept as the feature recognition management method for electronic payments in Example 1, this application also provides a feature recognition management system for electronic payments. Please refer to the appendix. Figure 2 The feature recognition management system for electronic payments includes: a relationship parsing module 11, used to parse the feature recognition relationships of electronic payment features, including risk relationships and security lock relationships; a feature decomposition module 12, used to decompose features based on the security lock relationships, evaluate the loss amount and confidence level of the decomposed features, and configure the recognition dispersion index of risk dispersion features; a recognition compensation module 13, used to perform risk feature recognition compensation according to the risk relationships, and configure the recognition compensation index of risk compensation relationship features according to the compensated risk security coefficient; and a parameter matching module 14, used to match the processing execution parameters of the recognition features according to the recognition dispersion index and the recognition compensation index, and to perform recognition management of the recognition features.

[0070] Furthermore, the relationship parsing module 11 in the feature recognition management system for electronic payment is also used to: acquire multiple features collected and applied throughout the entire electronic payment cycle, including one or more combinations of biometric features, behavioral features, device features, and environmental features; perform security analysis on the inherent attributes of each feature to obtain a static security level, wherein the inherent attributes include the uniqueness, immutability, copying difficulty, and privacy sensitivity of the feature; based on the static security level, perform risk analysis in conjunction with the dynamic context of each feature in the electronic payment process to obtain a dynamic risk coefficient, wherein the dynamic context includes one or more of transaction time, transaction location, transaction amount, transaction object, and network environment; establish a feature risk profile based on the static security level and dynamic risk coefficient, and determine feature recognition relationships based on the feature risk profile; wherein feature recognition relationships with a static security level that meets the preset security requirements and a dynamic risk coefficient that is less than the risk threshold are determined as the security locking relationship; feature recognition relationships with a dynamic risk coefficient that is greater than or equal to the risk threshold are determined as the risk relationship.

[0071] Furthermore, the relationship parsing module 11 in the feature recognition management system for electronic payments is also used to: establish a feature security evaluation matrix, including uniqueness weight, immutability weight, copying difficulty weight, and privacy sensitivity weight; score each feature according to its attribute features in four dimensions: uniqueness, immutability, copying difficulty, and privacy sensitivity, to obtain dimension score values; calculate the comprehensive security evaluation value of each feature by weighting based on the dimension score values ​​and their corresponding weights; and classify static security levels according to the range of the comprehensive security evaluation values.

[0072] Furthermore, the relationship parsing module 11 in the feature recognition management system for electronic payments is also used to: extract multiple predefined risk factors from the dynamic context, including transaction time anomaly, geographical location anomaly, transaction amount anomaly, counterparty unfamiliarity, and network environment risk; collect historical normal transaction data and fraudulent transaction data to form a training sample set, including a positive example sample set and a negative example sample set; use the risk factors as input features and transaction risk labels as outputs, and perform model training convergence based on the positive example sample set and negative example sample set using machine learning methods to obtain a dynamic risk assessment model; inject the dynamic context of each feature in the electronic payment process into the dynamic risk assessment model, and output the dynamic risk coefficient.

[0073] Furthermore, the relationship parsing module 11 in the feature recognition management system for electronic payment is also used to: acquire feature data of historical risk samples and known attack patterns as an adversarial sample set; input the adversarial sample set into the initial risk recognition model based on the feature risk profile for adversarial training, and extract key adversarial risk features through model iteration optimization; and correct the static security level and dynamic risk weight based on the adversarial risk features to reconstruct the feature risk profile.

[0074] Furthermore, the relationship parsing module 11 in the feature recognition management system for electronic payments is also used to: use a generative adversarial network architecture, wherein the generator is used to generate forged feature data simulating attacks, and the discriminator is the initial risk recognition model; through the game-like adversarial between the generator and the discriminator, the ability of the discriminator to identify fraud features is continuously strengthened, and the key adversarial risk features that can most affect the judgment result are identified in the adversarial process.

[0075] Furthermore, the feature decomposition module 12 in the feature recognition management system for electronic payment is also used to: decompose the original feature data of the security lock relationship into multiple feature segments; extract dispersed features based on the multiple feature segments to construct multiple dispersed feature combinations; compare the recognition rate of the dispersed feature combinations with the original feature data to obtain the recognition loss and recognition confidence difference; and evaluate the recognition of the dispersed feature combinations based on the recognition loss and recognition confidence difference to obtain the recognition dispersion index.

[0076] Furthermore, the identification compensation module 13 in the feature recognition management system for electronic payment is also used to: select one or more combinations of three compensation strategies—multi-feature assisted identification, dynamic verification enhancement, and behavioral biometric fusion—for the features of the risk relationship; perform risk analysis in dynamic context after using the compensation strategy to perform feature compensation processing on the risk relationship features; and evaluate the risk security coefficient of the selected compensation strategy, wherein the risk security coefficient represents the degree of improvement in feature recognition security after adopting the compensation strategy.

[0077] Furthermore, the parameter matching module 14 in the feature recognition management system for electronic payment is also used to: filter dispersed feature combinations or compensation strategies that meet the recognition security requirements according to the recognition dispersion index and the recognition compensation index, obtain feature processing execution parameters, including dispersed recognition features and compensated recognition features; and use the feature processing execution parameters to adjust the recognition management of the corresponding features; wherein the recognition security requirements are adaptively set according to the security requirement level of the electronic payment scenario.

[0078] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Figure 1The feature recognition management method and specific examples for electronic payment in Embodiment 1 are also applicable to the feature recognition management system for electronic payment in this embodiment. Through the foregoing detailed description of the feature recognition management method for electronic payment, those skilled in the art can clearly understand the feature recognition management system for electronic payment in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0079] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0080] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A feature recognition management method for electronic payments, characterized in that, include: Analyze the feature recognition relationships of electronic payment characteristics, including risk relationships and security lock-in relationships; Based on the security locking relationship, feature decomposition is performed, and the loss amount and confidence of the decomposed features are evaluated. The risk dispersion index is then configured to identify the dispersion features. Based on the aforementioned risk relationships, risk characteristics are identified and compensated, and an identification and compensation index for risk compensation relationship characteristics is configured according to the compensated risk safety coefficient. Based on the identification dispersion index and identification compensation index, the processing execution parameters of the identification features are matched to perform identification management on the identification features; Analyzing the feature recognition relationships of electronic payment characteristics, including risk relationships and security lock-in relationships, including: Acquire multiple features collected throughout the entire electronic payment lifecycle, including one or more combinations of biometrics, behavioral features, device features, and environmental features; A security analysis is performed on the inherent attributes of each feature to obtain a static security level. The inherent attributes include the uniqueness, immutability, difficulty of copying, and privacy sensitivity of the feature. Based on the static security level, risk analysis is performed by combining the dynamic context of each feature in the electronic payment process to obtain a dynamic risk coefficient. The dynamic context includes one or more of the following: transaction time, transaction location, transaction amount, transaction object, and network environment. A feature risk profile is established based on the static security level and dynamic risk coefficient, and feature recognition relationships are determined based on the feature risk profile. Specifically, the feature identification relationship where the static security level meets the preset security requirements and the dynamic risk coefficient is less than the risk threshold is determined as the security locking relationship; the feature identification relationship where the dynamic risk coefficient is greater than or equal to the risk threshold is determined as the risk relationship. Based on the aforementioned security locking relationship, feature decomposition is performed, and the decomposed features are evaluated for loss and confidence. An identification dispersion index for risk dispersion features is configured, including: The original feature data of the security locking relationship is decomposed into multiple feature fragments; Based on the multiple feature segments, dispersed feature extraction is performed to construct multiple dispersed feature combinations; By comparing the recognition rate of the dispersed feature combination with the original feature data, the recognition loss and recognition confidence difference are obtained. Based on the identification loss and identification confidence difference, a dispersion feature combination identification evaluation is performed to obtain the identification dispersion index; Based on the aforementioned risk relationships, risk characteristics are identified and compensated. According to the compensated risk safety coefficient, an identification and compensation index for risk compensation relationship characteristics is configured, including: For the characteristics of the risk relationship, one or more combinations of three compensation strategies are selected from multi-feature assisted identification, dynamic verification enhancement, and behavioral biometric fusion. After performing feature compensation processing on the risk relationship features using a compensation strategy, risk analysis is performed in dynamic context to evaluate the risk safety coefficient of the selected compensation strategy. The risk safety coefficient represents the degree of improvement in feature recognition security after adopting the compensation strategy. Based on the identification dispersion index and identification compensation index, the processing execution parameters of the identification features are matched to perform identification management, including: Based on the identification dispersion index and identification compensation index, select the combination of dispersion features or compensation strategies that meet the identification security requirements, and obtain the feature processing execution parameters, including dispersion identification features and compensation identification features. The corresponding feature recognition, management, and adjustment are performed using the aforementioned feature processing execution parameters; The security requirements are adaptively set according to the security needs of the electronic payment scenario.

2. The feature recognition management method for electronic payment according to claim 1, characterized in that, Security analysis is performed on the inherent attributes of each feature to obtain the static security level, including: Establish a feature security evaluation matrix, including uniqueness weight, immutability weight, copying difficulty weight, and privacy sensitivity weight; Each feature is scored based on its attribute characteristics across four dimensions: uniqueness, immutability, difficulty of replication, and privacy sensitivity, resulting in a dimension score value. Based on the dimensional score and its corresponding weight, a comprehensive security evaluation value for each feature is calculated by weighting. Static safety levels are classified based on the range of comprehensive safety evaluation values.

3. The feature recognition management method for electronic payment according to claim 1, characterized in that, Based on the static security level, risk analysis is performed by combining the dynamic context of each feature in the electronic payment process to obtain dynamic risk coefficients, including: Multiple predefined risk factors are extracted from the dynamic context, including transaction time anomaly, geographical location anomaly, transaction amount anomaly, unfamiliarity with the counterparty, and network environment risk. Collect historical normal transaction data and fraudulent transaction data to form a training sample set, including a positive sample set and a negative sample set; Using the aforementioned risk factors as input features and transaction risk labels as outputs, a dynamic risk assessment model is obtained by training and converging the model using machine learning methods based on the positive and negative sample sets. The dynamic risk assessment model is injected with the dynamic context of each feature in the electronic payment process, and the dynamic risk coefficient is output.

4. The feature recognition management method for electronic payment according to claim 1, characterized in that, Based on the static security level and dynamic risk coefficient, a characteristic risk profile is established, which then includes: Obtain feature data of historical risk samples and known attack patterns as an adversarial sample set; The adversarial sample set is input into the initial risk identification model based on feature risk profile for adversarial training, and key adversarial risk features are extracted through model iterative optimization. Based on the adversarial risk characteristics, the static security level and dynamic risk weight are modified to reconstruct the characteristic risk profile.

5. The feature recognition management method for electronic payment according to claim 4, characterized in that, The adversarial example set is input into the initial risk identification model based on feature risk profiles for adversarial training, including: A generative adversarial network architecture is used, in which a generator is used to generate forged feature data simulating attacks, and a discriminator is the initial risk identification model. Through the game-like confrontation between the generator and the discriminator, the discriminator's ability to identify fraud features is continuously strengthened, and the key adversarial risk features that can most affect the discrimination result are identified during the confrontation process.

6. A feature recognition management system for electronic payments, characterized in that, The step of implementing the feature recognition management method for electronic payment according to any one of claims 1 to 5, wherein the feature recognition management system for electronic payment comprises: The relationship parsing module is used to parse the feature recognition relationships of electronic payment features, including risk relationships and security lock-in relationships; The feature decomposition module is used to decompose features based on the security locking relationship, evaluate the loss and confidence of the decomposed features, and configure the identification dispersion index of risk dispersion features. The identification and compensation module is used to identify and compensate for risk characteristics based on the risk relationship, and to configure the identification and compensation index of the risk compensation relationship characteristics according to the risk safety coefficient of the compensation. The parameter matching module is used to match the processing execution parameters of the identification features according to the identification dispersion index and the identification compensation index, and to perform identification management on the identification features.

Images