Mobile payment supervision method and system based on aggregation payment transaction routing

Abstract

The invention provides a mobile payment supervision method and system based on aggregation payment transaction routing, and the method comprises the steps: further mining cause features and association rules of different deviation types from massive historical deviation transaction data through employing an association rule mining algorithm after the judgment of the deviation types is completed, and building a deviation cause knowledge base; for each deviation transaction, acquiring corresponding cause characteristics and association rules from a deviation cause knowledge base according to the deviation type of the deviation transaction, and judging the specific cause of the deviation transaction by adopting a pattern matching algorithm; matching suitable treatment measures from a pre-established treatment strategy knowledge base according to the type and cause of the deviation transaction; after the treatment measures are determined, real-time information sharing and linkage with a transaction system, a risk control system and the like are achieved, corresponding treatment processes are triggered, and the execution progress and result of the treatment measures are monitored.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a mobile payment supervision method and system based on aggregated payment transaction routing. Background Art

[0002] In the mobile payment supervision system, the identification and processing of transaction path deviations face technical contradictions. On the one hand, the system needs to monitor massive transaction data in real time and quickly discover the deviation between the transaction path and the expected path; on the other hand, the determination of the transaction path involves complex business rules and algorithms, which puts higher requirements on the system's processing capabilities. In addition, the causes of transaction deviations are diverse, which may be caused by system failures, routing rule errors, fraudulent behavior, etc. The deviation characteristics caused by different causes are different, which increases the difficulty of identification and processing. The supervision system needs to calculate the difference between the actual path and the expected path of each transaction in real time in massive transaction data based on the preset transaction path model and rules. Transactions with a difference exceeding the threshold are judged as deviation transactions, and the type and cause of the deviation need to be further identified. This process involves complex data processing, calculation and comparison, which places strict requirements on the performance and stability of the system. At the same time, the setting of transaction path models and rules directly affects the accuracy of deviation determination, and needs to be continuously optimized and adjusted in actual business scenarios. The supervision system also needs to take corresponding disposal measures according to the type and cause of deviation; this requires the supervision system to be able to achieve real-time information sharing and linkage disposal with other systems such as transaction systems and risk control systems. In short, transaction deviation identification and processing involves multiple links such as real-time monitoring of transaction data, calculation of massive data, difference comparison, determination of deviation type, and implementation of disposal measures, which puts forward all-round requirements on the technical capabilities of the mobile payment regulatory system. Summary of the invention

[0003] The present invention provides a mobile payment supervision method based on aggregate payment transaction routing, which mainly includes: Obtain the actual transaction path and expected transaction path of each transaction, and determine whether there is a transaction deviation by comparing the difference between the actual transaction path and the expected transaction path. If the difference exceeds the preset threshold, the transaction is determined to be a deviation transaction and marked as a deviation transaction to be processed; A classification algorithm is used to determine the deviation type of the identified deviation transactions, and the deviation type of each deviation transaction is determined based on the pre-established deviation type knowledge base; After the deviation type is determined, the association rule mining algorithm is used to mine the cause characteristics and association rules of different deviation types from the historical deviation transaction data to establish a deviation cause knowledge base; According to the deviation type of each deviation transaction, the corresponding cause characteristics and association rules are obtained from the deviation cause knowledge base, and the specific cause of the deviation transaction is determined by using the pattern matching algorithm; According to the type and cause of deviation transactions, match the applicable disposal measures from the pre-established disposal strategy knowledge base; After determining the disposal measures, the corresponding disposal process is triggered through real-time information sharing and linkage with the trading system and risk control system, and the progress and results of the disposal measures are monitored; After completing the handling of a single deviation transaction, continue to track the subsequent status changes of the transaction. If any abnormal changes occur, re-trigger the deviation transaction identification process, and optimize and adjust the transaction path model and rules based on the accumulated deviation transaction data.

[0004] The present invention provides a mobile payment supervision system based on aggregate payment transaction routing, which mainly includes: Deviation transaction identification module, used to obtain the actual transaction path and expected transaction path of each transaction, and determine whether there is a transaction deviation by comparing the difference between the actual transaction path and the expected transaction path. If the difference exceeds the preset threshold, the transaction is determined to be a deviation transaction and marked as a deviation transaction to be processed; The deviation type determination module is used to determine the deviation type of the identified deviation transactions using a classification algorithm. Based on the pre-established deviation type knowledge base, it comprehensively considers multi-dimensional features such as transaction amount, transaction time, and transaction object to determine the deviation type of each deviation transaction. The deviation cause analysis module is used to further use the association rule mining algorithm to mine the cause characteristics and association rules of different deviation types from the massive historical deviation transaction data after completing the deviation type determination, and establish a deviation cause knowledge base; The deviation cause determination module is used to obtain the corresponding cause characteristics and association rules from the deviation cause knowledge base for each deviation transaction according to its deviation type, and use the pattern matching algorithm to determine the specific cause of the deviation transaction; A disposal measure matching module is used to match applicable disposal measures from a pre-established disposal strategy knowledge base according to the type and cause of the deviation transaction; The disposal measure execution module is used to trigger the corresponding disposal process and monitor the execution progress and results of the disposal measures by realizing real-time information sharing and linkage with the trading system, risk control system, etc. after the disposal measures are determined; The continuous tracking and optimization module is used to continuously track the subsequent status changes of a single deviation transaction after completing the disposal of the transaction. If any abnormal changes occur, the deviation transaction identification process is re-triggered, and the transaction path model and rules are optimized and adjusted based on the accumulated deviation transaction data.

[0005] The technical solution provided by the embodiment of the present invention may have the following beneficial effects: The present invention discloses a mobile payment supervision method based on aggregate payment transaction routing. By comparing the difference between the actual and expected transaction paths, the deviation transaction is identified. The classification algorithm is used to determine the type of the deviation transaction, and the association rule mining algorithm is used to mine the cause characteristics from the historical data to establish a deviation cause knowledge base. According to the deviation type and cause characteristics, the pattern matching algorithm is used to determine the specific cause, and the applicable disposal measures are matched from the disposal strategy knowledge base. The present invention realizes real-time linkage with the transaction system and the risk control system, triggers the corresponding disposal process and monitors the execution status. After the disposal is completed, the transaction status changes are continuously tracked, and the identification process is re-triggered when necessary. The transaction path model and rules are optimized and adjusted through the accumulated deviation transaction data. The present invention realizes the intelligent identification, classification, cause analysis and disposal of transaction deviations, improves the processing efficiency and accuracy of abnormal transactions, effectively reduces transaction risks, and provides a comprehensive transaction monitoring solution for financial institutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 The present invention is a flow chart of a mobile payment supervision method based on aggregated payment transaction routing.

[0007] Figure 2 It is a schematic diagram of a mobile payment supervision method and system based on aggregated payment transaction routing of the present invention.

[0008] Figure 3 It is another schematic diagram of a mobile payment supervision method and system based on aggregated payment transaction routing of the present invention.

[0009] Figure 4 It is a structural schematic diagram of a mobile payment supervision method and system based on aggregated payment transaction routing of the present invention. DETAILED DESCRIPTION

[0010] The following will describe the technical solutions in the embodiments of the present invention in detail in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention.

[0011] like Figure 1-4 In this embodiment, a mobile payment supervision method and system based on aggregated payment transaction routing may specifically include: Step S101, obtain the actual transaction path and expected transaction path of each transaction, and determine whether there is a transaction deviation by comparing the difference between the actual transaction path and the expected transaction path. If the difference exceeds a preset threshold, the transaction is determined to be a deviation transaction and marked as a deviation transaction to be processed.

[0012] Obtain the actual transaction path and expected transaction path of each transaction, and use the dynamic time warping algorithm to calculate the difference between the actual transaction path and the expected transaction path; determine whether the difference exceeds the preset threshold. If it exceeds the preset threshold, mark the transaction as a pending deviation transaction; for the deviation transactions marked as pending, obtain the detailed differences between the actual transaction path and the expected transaction path in terms of key elements such as time, amount, and transaction object, and comprehensively determine the cause of the transaction deviation in combination with business rules and risk control strategies; extract transaction time, amount, and object features based on the difference information and deviation causes of the deviation transactions, and use the decision tree algorithm to classify the deviation transactions; use corresponding processing strategies for different types of deviation transactions classified by the decision tree algorithm If the deviation is caused by information entry errors, data correction will be carried out; if the deviation is caused by system failure, the system investigation and repair process will be started; if the deviation is caused by abnormal operation, the compliance and rationality of the transaction will be judged in combination with business rules and authorization system; on the basis of deviation transaction processing, an anti-fraud model based on graph computing is adopted to extract transaction subjects, time, and amount elements from deviation transactions, build a transaction relationship graph, mine fraudulent transaction patterns, and identify and warn high-risk transactions in real time; through continuous evaluation of the processing results of deviation transactions, dynamically optimize the preset thresholds and processing strategies, combined with the recognition rate, processing accuracy, and timeliness indicators of deviation transactions, through threshold adjustment and strategy iteration, the management and control effect of deviation transactions is continuously improved.

[0013] For example, taking bank transfer as an example, the expected path may be that customer A transfers 1,000 yuan to customer B through mobile banking, but the actual path is that customer A transfers 800 yuan to customer C through ATM. The dynamic time warping algorithm can effectively measure the similarity of the two paths. The algorithm can handle the situation of inconsistent time series lengths and find the optimal alignment through dynamic programming. If the difference exceeds the preset threshold, such as 0.3, the transaction is marked as a pending deviation transaction. For deviation transactions, it is necessary to deeply analyze the specific differences in key elements such as time, amount, and transaction object. For example, the transfer time changes from the expected working day to the weekend, the amount changes from 1,000 yuan to 800 yuan, and the recipient changes from customer B to customer C. Combining business rules and risk control strategies, possible deviation causes can be inferred, such as customer operation errors, system failures, or abnormal transactions. Using the decision tree algorithm to classify deviation transactions, a multi-level classification model can be constructed. The root node of the decision tree can be the deviation type, such as information entry errors, system failures, abnormal operations, etc. Each node is divided by different features, such as time deviation, amount deviation, object deviation, etc. In this way, deviation transactions can be classified into different leaf nodes, which is convenient for subsequent targeted processing strategies. For different types of deviation transactions, corresponding processing strategies need to be adopted. If the information is entered incorrectly, such as the customer mistakenly enters the name of the payee, it can be solved by data correction. If the deviation is caused by system failure, such as ATM equipment failure causing withdrawal failure but the account has been deducted, it is necessary to start the system troubleshooting and repair process. If the deviation is caused by abnormal operation, such as frequent large-amount transfers in a short period of time, it is necessary to combine business rules and authorization systems to judge the compliance and rationality of the transaction. On the basis of processing deviation transactions, an anti-fraud model based on graph computing can be used to further improve risk control capabilities. The model uses transaction subjects as nodes and transaction relationships as edges to construct a complex transaction network. By analyzing the topological structure, node attributes and edge attributes of the network, potential fraud patterns can be mined. For example, if an account is found to have large-amount transfers with multiple unfamiliar accounts in a short period of time, it may be engaged in illegal behavior; or if multiple accounts are found to form a closed-loop transfer chain, it may be cash-out behavior. In order to continuously optimize the control effect of deviation transactions, it is necessary to evaluate and feedback the processing results. Multiple key indicators can be set, such as the recognition rate, processing accuracy, and processing timeliness of deviation transactions. Through the changing trends of these indicators, the preset thresholds and optimized processing strategies can be dynamically adjusted. For example, if a large number of normal transactions are found to be misjudged as deviation transactions, the preset thresholds can be appropriately increased; if the processing accuracy of a certain type of deviation transactions is found to be low, the corresponding processing strategies can be improved in a targeted manner. The whole process forms a closed-loop optimization mechanism. Through the continuous accumulation of experience and data, the system's recognition ability and processing efficiency will gradually improve. This method can not only effectively control transaction risks, but also provide valuable insights for business decisions.For example, by analyzing the patterns of deviation transactions, we can discover the changing trends of customer behavior and provide a basis for product innovation and service optimization. At the same time, this method also helps to improve the adaptive ability of the system and better cope with the challenges of the ever-changing financial environment and new fraud methods.

[0014] Step S102: using a classification algorithm to determine the deviation type of the identified deviation transactions, and determining the deviation type to which each deviation transaction belongs based on a pre-established deviation type knowledge base.

[0015] Acquire transaction data, which includes transaction amount, transaction time, and transaction object information; clean the transaction data, remove duplicate values ​​and missing values ​​in the transaction data, and convert the transaction data into a unified format; select transaction amount, transaction time, and transaction object from the cleaned transaction data as model training features; normalize the model training features, and scale the feature values ​​of the model training features to between 0 and 1; divide the normalized model training features into a training set and a test set; use the training set and the random forest algorithm to train the deviation transaction determination model; input the test set into the deviation transaction determination model to obtain the probability value of each transaction output by the deviation transaction determination model belonging to each deviation type; obtain a preset probability threshold, and if the probability value is greater than the probability threshold, determine that the transaction belongs to the corresponding deviation type; obtain a pre-established deviation type rule library, which contains each The characteristic rules and judgment conditions of type deviation transactions are stored in the form of "characteristic-deviation type". The transaction data judged as deviation type is matched with the characteristic rules and judgment conditions in the deviation type rule library. If the match is successful, the deviation type judgment result of the transaction data is updated to the deviation type corresponding to the characteristic rules and judgment conditions. If the transaction is judged as a high-risk deviation type, risk warning information is generated and relevant personnel are notified to handle it. The deviation type judgment result of the transaction data is stored together with the transaction data. The deviation type rule library is updated with the stored transaction data and the corresponding deviation type judgment result. The updated deviation type rule library is used to reselect model training features from the transaction data and perform normalization. The deviation transaction judgment model is retrained using the reselected and normalized model training features and the updated deviation type judgment results.

[0016] For example, the acquisition and cleaning of transaction data is the basis for determining deviation transactions. Taking bank transfer business as an example, the original transaction data may contain information such as transfer amount, time, payer, and payee. During the data cleaning process, you may encounter problems such as duplicate records, zero amount, or missing payee information. By removing duplicates and filling missing values, you can get a high-quality data set. For example, for the missing payee name, you can fill it in by querying historical transaction records based on the account number. Feature selection and normalization are the key to building an effective model. In the transfer business, the transaction amount, time, and object are important indicators for determining whether the transaction is biased. The amount may range from a few yuan to millions, the time span may be 24 hours a day, and the transaction object may be different types such as individuals and enterprises. Through normalization, these different dimensional features can be unified into the range of 0 to 1, so that the model can fairly evaluate the importance of each feature. The random forest algorithm has advantages in determining deviation transactions. It can process both numerical (such as amount) and categorical (such as transaction object) features at the same time, and is insensitive to outliers. By building multiple decision trees and taking majority voting results, random forests can effectively reduce the risk of overfitting. In practice, the model performance can be optimized by adjusting parameters such as the number of trees and the maximum depth of each tree. The setting of probability thresholds requires a trade-off between accuracy and recall. For example, setting the threshold to 0.8 means that only when the model determines that a transaction belongs to a certain deviation type with a probability of more than 80%, it will be marked as a deviation transaction. This approach can reduce misjudgments, but may miss some edge cases. In actual applications, the threshold can be adjusted according to business needs and risk preferences. The establishment and updating of the deviation type rule base is a dynamic process. The initial rules may come from expert experience, such as "large transfers in the early morning hours" may be considered high-risk operations. As new deviation types are discovered and confirmed, the rule base will continue to expand. For example, if it is found that multiple small transfers are aggregated to the same account in a short period of time, it may be a new type of financial crime, and this pattern can be added to the rule base. Timely processing of high-risk deviation transactions is crucial to preventing financial risks. When the system identifies possible high-risk transactions, such as multiple large cross-border transfers in a short period of time, risk warning information will be generated immediately. This information may be notified to risk control personnel through SMS, email, etc., so that timely measures can be taken, such as suspending transactions, contacting customers for verification, etc. The storage of transaction data and deviation type determination results provides a basis for subsequent analysis. By mining historical data, new deviation patterns can be discovered or existing rules can be optimized. For example, by analyzing transactions that are judged to be normal but are actually deviations, the blind spots of the model can be discovered, thereby improving the accuracy of the determination. Regular updating and retraining of the model is key to maintaining its effectiveness. As new data accumulates and new rules are added, the model needs to adapt to these changes.For example, if a large number of cases of financial crimes using virtual currencies have been discovered recently, relevant features need to be incorporated into the model and the model needs to be retrained using a data set containing these new cases. The entire process forms a closed-loop system that improves the ability to identify and handle deviant transactions through continuous learning and optimization. This approach also helps improve the system's adaptive capabilities and better cope with the ever-changing financial environment and new fraud methods.

[0017] Step S103, after the deviation type is determined, an association rule mining algorithm is used to mine the cause characteristics and association rules of different deviation types from the historical deviation transaction data to establish a deviation cause knowledge base.

[0018] The historical deviation transaction data of the determined deviation type is obtained, and the K-means clustering algorithm is used to classify the historical deviation transaction data according to the deviation type to obtain sub-datasets of different deviation types; for each sub-dataset, the principal component analysis technology is used to extract the feature vector related to the deviation occurrence to obtain the feature data set; the Apriori algorithm is used to mine the association rules of the feature data set to obtain the feature combination of different deviation types that frequently occur; based on the feature combination, the correlation strength between the features is calculated, and when the correlation strength exceeds the preset threshold, the strong association rule is determined; based on the strong association rule, a mapping relationship between the deviation type and the cause is established, and a deviation cause knowledge base is constructed; the newly added deviation transaction data is obtained, and the principal component analysis technology is used to extract the features of the newly added deviation transaction data to obtain the newly added deviation transaction features; the newly added deviation transaction features are input into the decision tree model, and the deviation causes of the newly added deviation transaction data are determined according to the strong association rules in the deviation cause knowledge base; the deviation causes of the newly added deviation transaction data are fed back to the deviation cause knowledge base, and the deviation cause knowledge base is updated.

[0019] For example, taking bank transfer business as an example, the K-means clustering algorithm can classify historical deviation transaction data into categories such as "abnormal amount", "abnormal frequency", and "abnormal time". For the "abnormal amount" category, principal component analysis technology may extract key features such as transaction amount, account balance, and historical transaction mean. This dimensionality reduction process not only reduces the complexity of the data, but also highlights the most explanatory feature combination. Association rule mining can reveal the potential connection between features. Assume that in the "abnormal amount" category, the Apriori algorithm finds the frequent item set of "large transaction", "non-working hours" and "remote operation". If the support of this combination is 6%, which is higher than the preset 5%, and the confidence is 80%, which is higher than the preset 70%, it can be identified as a strong association rule. This means that when these three features appear at the same time, it is very likely to be a deviation transaction with abnormal amount. Building a knowledge base of deviation causes is the basis for realizing intelligent analysis. The rules mined are stored in the MySQL database, forming the following structure: deviation type (abnormal amount), feature combination (large transaction, non-working hours, remote operation), correlation strength (support 6%, confidence 80%), possible cause (fund theft or financial crime). This structured storage facilitates subsequent rapid retrieval and update. The analysis process of newly added deviation transactions demonstrates the application value of the knowledge base. For example, the system detects a new suspicious transaction: late at night, a large inter-provincial transfer suddenly occurs in an account with long-term low-value transactions. After extracting features through principal component analysis, the decision tree model may match the "abnormal amount" rule in the knowledge base. The system then gives an early warning: the transaction meets the characteristics of fund theft and recommends manual review. The advantage of this method lies in its adaptability. As new deviation transactions are identified and analyzed, the knowledge base will continue to be enriched and optimized. For example, if more and more legal large-value night transfers are found, such as regular settlements of multinational companies, the system will automatically adjust the weights of relevant rules to reduce the false alarm rate. At the same time, emerging deviation patterns, such as the use of virtual currency for financial crimes, can also be captured and added to the knowledge base in a timely manner. This closed-loop system can not only effectively identify known deviation types, but also detect new abnormal patterns in a timely manner. Through continuous learning and optimization, it provides financial institutions with a comprehensive and flexible transaction risk management solution, greatly improving the accuracy of identifying abnormal transactions and the efficiency of processing them.

[0020] Step S104, according to the deviation type of each deviation transaction, the corresponding cause characteristics and association rules are obtained from the deviation cause knowledge base, and the specific cause of the deviation transaction is determined by using a pattern matching algorithm.

[0021] Obtain the deviation transaction record and determine the deviation type of the deviation transaction record; according to the deviation type, obtain the cause feature set corresponding to the deviation type from the pre-established deviation cause knowledge base table, the cause feature set includes time feature, amount feature and transaction counterparty feature; obtain the transaction time of the deviation transaction record, match the transaction time with the time feature, if the transaction time meets the preset time feature threshold, then classify the transaction time into the time feature matching result set; obtain the transaction amount and transaction counterparty information of the deviation transaction record, combine with historical transaction data, and use the Apriori algorithm to generate an association rule set, the association rule set is obtained by mining frequent item sets on historical transaction data, setting support and confidence thresholds, and generating association rules that meet the conditions; match the association rule set with the rule features in the cause feature set, if the match is successful, then classify the rule features into the rule feature matching result set; input the time feature matching result set and the rule feature matching result set into the pre-trained decision tree model, and judge the deviation cause of the deviation transaction record through the decision tree model.

[0022] For example, deviation transaction analysis is an important part of financial risk control, and its core lies in accurately identifying the type of transaction deviation and tracing its cause. Taking bank transfer business as an example, an abnormally large transfer may be initially determined by the system as a deviation transaction of the "abnormal amount" type. To deeply analyze such deviations, the system will first query the pre-established deviation cause knowledge base. The knowledge base may contain features of multiple dimensions, such as time features (non-working hours), amount features (exceeding historical averages), and counterparty features (first transaction objects). These feature sets provide important references for subsequent analysis. Time feature analysis is one of the key steps in determining transaction anomalies. Assume that the large transfer occurred at 3 a.m., which is obviously beyond the normal banking business hours. The system will compare this time point with the preset time feature threshold (such as 9:00-17:00 on working days) and classify it into the time feature matching result set. This step helps to identify possible illegal operations or account theft. The analysis of amount features and counterparty features requires combining historical transaction data and using the Apriori algorithm to generate association rules. For example, the system may find that the average daily transaction amount of the account in the past six months was 1,000 yuan, while the transfer amount this time was as high as 100,000 yuan. At the same time, the counterparty of the transaction is an account that has never had a transaction record. The Apriori algorithm generates frequent item sets based on this information, such as {large transaction, unusual time, new transaction object}. In the process of generating association rules, the system will set appropriate support and confidence thresholds. Assume that the support threshold is set to 1% and the confidence threshold is set to 80%. This means that only rules with a frequency of more than 1% in historical data and a probability of more than 80% when the preconditions are met will be taken into consideration. Such settings can effectively filter out rules with high contingency and improve the accuracy of analysis. The generated association rule set will be matched with the rule features in the cause feature set. If a match is found, the system will classify it into the rule feature matching result set. The significance of this step is that it can reveal potential risk patterns. For example, the system may find that the combination of {large transaction, unusual time, new transaction object} frequently appears in past fraud cases. Finally, the time feature matching results and rule feature matching results will be input into the pre-trained decision tree model. The advantage of the decision tree model is that it is highly interpretable and can clearly show the decision path. In this case, the decision tree may first consider the transaction amount, then the transaction time, and finally the characteristics of the transaction object. Through this layer-by-layer screening, the model may eventually come to the conclusion that "high-risk transactions, suspected account theft". The training process of the decision tree model is also worthy of attention. It is usually based on a large amount of labeled historical data, including normal transactions and various confirmed deviation transactions. During the training process, the model will learn how to make accurate judgments based on the combination of different features.For example, it may learn that while large transactions themselves do not necessarily mean risk, when they occur at unusual times and with new transaction objects, the risk factor increases significantly. This multi-step, multi-dimensional analysis method has significant advantages. It can not only quickly identify potential high-risk transactions, but also provide detailed risk explanations, which helps bank staff conduct subsequent manual review and intervention. At the same time, this process is dynamic and adaptive. As new transaction data continues to accumulate, association rules and decision tree models are regularly updated to ensure that the system can cope with changing fraud methods and transaction patterns. Through this refined analysis of deviation transactions, financial institutions can minimize interference with normal business while ensuring transaction security, improving customer satisfaction and operational efficiency.

[0023] Step S105, matching applicable disposal measures from a pre-established disposal strategy knowledge base according to the type and cause of the deviation transaction.

[0024] A1. Obtain deviation transaction records, extract the types and causes of deviation transactions, and construct deviation transaction feature vectors; A2. Use K-means clustering algorithm to cluster deviation transaction feature vectors to obtain cluster centers of different disposal strategies; A3. Calculate the Euclidean distance between the current deviation transaction feature vector and each cluster center, and select the disposal strategy corresponding to the cluster center with the smallest distance as the candidate disposal strategy; A4. Extract the association rules of the candidate disposal strategy from the disposal strategy knowledge base, and use the association rule matching algorithm to determine whether the current deviation transaction meets the triggering conditions of the candidate disposal strategy; if the triggering conditions are met, the candidate disposal strategy is added to the set of disposal strategies to be selected; otherwise, the candidate disposal strategy is excluded, and the disposal strategy corresponding to the next cluster center with the smallest distance is selected as the new candidate disposal strategy, and steps A3-A4 are repeated until a candidate disposal strategy that meets the triggering conditions is found. A5. Sort the strategies in the candidate disposal strategy set according to the preset priority rules, and select the disposal strategy with the highest priority as the final disposal measure; A6. According to the final disposal measure, combined with business rules and system logic, generate a detailed disposal plan, including disposal time, disposal object, disposal amount, and disposal reason; A7. Send the generated detailed disposal plan to the relevant business system for actual disposal operation, and feed back the disposal execution result to the disposal strategy knowledge base; A8. If the disposal result does not meet expectations, adjust the association rules in the disposal strategy knowledge base, and optimize the matching accuracy of the disposal strategy by modifying the conditions and conclusions of the association rules, and improve the disposal effect of deviation transactions; A9. The adjusted association rules will be applied to the subsequent deviation transaction disposal process to continuously optimize the performance of the disposal strategy.

[0025] For example, the handling of deviation transactions is a key link in financial risk control, and its core lies in accurately identifying abnormal transactions and taking appropriate handling measures. Taking bank credit card business as an example, the system may detect an abnormal large-amount consumption transaction. In order to effectively handle such deviation transactions, it is necessary to first construct a transaction feature vector. This vector may contain multiple dimensions, such as transaction amount, transaction time, merchant type, geographic location, etc. For example, suppose a hotel consumption of 100,000 yuan is detected, which occurs at 2 a.m. and is located in a city that has never been to. These features will be quantified and formed into a multidimensional vector. Next, the system will use the K-means clustering algorithm to classify historical deviation transactions. Assume that after cluster analysis, the system identifies three main cluster centers, corresponding to the three types of "suspected fraudulent use", "high-risk consumption" and "abnormal location consumption". Each cluster center is associated with a specific handling strategy. For example, "suspected fraudulent use" may correspond to the strategy of "immediately freeze the card and contact the cardholder", "high-risk consumption" may correspond to the strategy of "require additional identity verification", and "abnormal location consumption" may correspond to the strategy of "sending risk reminder SMS". The system calculates the Euclidean distance between the current deviation transaction and these cluster centers. Assume that the calculation result shows that the transaction is closest to the "suspected fraud" cluster center. This means that "immediately freeze the card and contact the cardholder" will become the preferred candidate disposal strategy. However, before taking action, the system needs to further verify the applicability of this strategy. To this end, the system queries the disposal strategy knowledge base and extracts rules related to "suspected fraud". These rules may include "large amount of consumption in the early morning hours" and "first consumption in an unconventional place". Through the association rule matching algorithm, the system determines whether the current transaction meets these trigger conditions. In this case, since the transaction time, amount and location all meet the rule description, the strategy of "immediately freeze the card and contact the cardholder" is confirmed to be valid and added to the set of candidate disposal strategies. If the preferred strategy does not meet the trigger conditions, the system will turn to the next closest cluster center and repeat the above process. This method ensures the flexibility and accuracy of the disposal strategy. Finally, the system selects the most suitable disposal measure from the candidate strategies according to the preset priority rules. The priority may be set based on factors such as risk level and customer value. After determining the final disposal strategy, the system will generate a detailed implementation plan. In this case, the plan may include: immediately suspending the card transaction function, contacting the cardholder through a reserved phone number, setting unfreezing conditions within 24 hours, etc. These specific steps will be sent to relevant business systems for execution, such as risk control systems, customer service systems, etc. The disposal results will be recorded and fed back to the knowledge base. If it is finally confirmed that this is a real case of card fraud, the system will strengthen the weight of the relevant rules. On the contrary, if it is a misjudgment, the system will adjust the rules, such as relaxing the amount threshold for transactions in the early morning or lowering the risk score for consumption in unconventional locations.This dynamic adjustment mechanism ensures that the disposal strategy can be continuously optimized to adapt to changing fraud patterns and customer behaviors. Through this multi-level, adaptive disposal process, financial institutions can minimize interference with normal business while ensuring transaction security. For example, for a high-net-worth customer who travels frequently, the system may gradually learn the pattern of his consumption in unconventional places, thereby reducing the false alarm rate and improving the customer experience. At the same time, this method can also quickly respond to new fraud methods, such as by adjusting the parameters of the clustering algorithm or adding new feature dimensions to identify and deal with newly emerging deviant transaction types.

[0026] Step S106, after determining the disposal measures, real-time information sharing and linkage with the transaction system and risk control system are achieved to trigger the corresponding disposal process and monitor the execution progress and results of the disposal measures.

[0027] Obtain transaction data reported by the transaction system; for transaction data, count the number of transactions of a certain type within a sliding window of a fixed time length as the transaction frequency; if the transaction frequency exceeds the preset frequency threshold, trigger an early warning; for transaction data, use regular expressions to determine whether the transaction type meets the preset risk control rules; if it meets the preset risk control rules, use the decision tree model to determine the transaction risk level; if both the transaction risk level and the transaction frequency exceed the preset threshold, calculate the transaction risk score by weighted average method, where the weights of the transaction frequency and the transaction risk level are 3 and 7 respectively; if the transaction risk score exceeds the preset score threshold, determine the transaction disposal measures according to the range of the transaction risk score, and the transaction disposal measures include freezing the account, restricting transactions and manual review; according to the transaction risk level, determine the transaction risk level, and determine the transaction risk level; if both the transaction risk level and the transaction frequency exceed the preset threshold ... The transaction system and the risk control system perform corresponding operations according to the disposal instructions, among which the transaction system performs account freezing and transaction restriction operations, and the risk control system performs manual review operations; the transaction system and the risk control system use the execution results as execution feedback information, and the execution feedback information includes the execution results, execution time and executor; if the execution result in the execution feedback information is successful, the disposal result meets the expected goal; if the execution result in the execution feedback information is failed, the disposal result does not meet the expected goal; if the disposal result meets the expected goal, the status of the transaction record is updated to disposed; if the disposal result does not meet the expected goal, the status of the transaction record is updated to disposal failure.

[0028] For example, the core of the transaction monitoring system is to collect and analyze transaction data in real time to identify potential risky transactions. The system first collects data reported by the transaction system, which contains key information such as transaction time, amount, and type. In order to effectively identify abnormal transaction patterns, the system uses sliding window technology to count transaction frequencies. For example, the system may set a 24-hour sliding window to count the number of transactions of a specific type for each account during this period. Assuming that an account has made 20 large transfers in the past 24 hours, while the average number of transfers per day does not exceed 5 under normal circumstances, this abnormally high-frequency transaction behavior will trigger a system warning. In addition to frequency analysis, the system also matches transaction descriptions through regular expressions to determine whether the transaction complies with the preset risk control rules. For example, the system may set rules to identify transaction descriptions containing sensitive words. If the transaction description matches these rules, the system will further use a decision tree model to evaluate the transaction risk level. The decision tree model may consider multiple factors, such as transaction amount, counterparty, transaction time, etc., to comprehensively evaluate transaction risks. The system uses the weighted average method to calculate the final transaction risk score, assigning 30% and 70% weights to the transaction frequency and risk level respectively. This weight distribution reflects the higher importance of risk level in the judgment. For example, if the frequency score of a transaction is 8 (out of 10 points) and the risk level is 9 (out of 10 points), the final risk score is: 8×0.3+9×0.7=8.7. Based on the calculated risk score, the system will select the corresponding disposal measures. For example, a risk score between 8-9 may trigger a restriction on trading, while a risk score above 9 may directly freeze the account. The disposal instructions generated by the system contain specific execution details, such as "freeze account 123456, transaction time 2023-05-01 14:30:00, transaction amount 50,000 yuan, transaction type: cross-border transfer". The trading system and risk control system perform corresponding operations according to the disposal instructions. For example, the trading system may immediately freeze the specified account and prohibit it from making any transactions; the risk control system may mark the transaction as pending manual review and notify relevant personnel for in-depth investigation. The system will record the execution results of these operations, including execution time, executor and other information. If the disposal measures are successfully executed, the system will update the transaction record status to "disposed". This ensures that the same transaction does not repeatedly trigger risk control measures. If the execution fails, the system will update the status to "failed to dispose" and may trigger a higher level of alarm to ensure that the problem is handled in a timely manner. This process is designed to quickly identify and handle high-risk transactions, while minimizing misjudgments through a multi-level judgment and disposal mechanism. It can effectively enhance the risk management capabilities of financial institutions, protect the security of customer assets, and assist institutions in complying with relevant regulatory requirements.

[0029] Step S107, after completing the handling of a single deviation transaction, continue to track the subsequent status changes of the transaction. If an abnormal change occurs, re-trigger the deviation transaction identification process, and optimize and adjust the transaction path model and rules based on the accumulated deviation transaction data.

[0030] The status data of a single deviation transaction after disposal is obtained, and the preset abnormal change threshold is used for judgment. If the status data exceeds the threshold, the deviation transaction identification process is triggered; if the deviation transaction identification process is triggered, all data information of the transaction is extracted from the transaction database, and merged with the historical deviation transaction data set to construct a cumulative deviation transaction data set; the support vector machine algorithm is used to train the cumulative deviation transaction data set to obtain a transaction path classification model; the rationality of the transaction path is judged through the transaction path classification model, and if there is an unreasonable path, the feature vector of the unreasonable path is extracted; the Apriori algorithm is used to mine association rules on the unreasonable path feature vector, obtain optimization rules, and determine the adjustment parameters of the rules; the adjusted rules are input into the transaction path classification model, and the rationality of the rules is judged again. If the rules are reasonable, the online rule library is updated.

[0031] For example, the transaction status monitoring program is an important part of the risk management of the financial system. The program obtains the status data of a single deviation transaction in real time and compares it with the preset abnormal change threshold to quickly identify potential risks. For example, a bank sets up a monitoring program to track the status changes of large transfer transactions in real time. Assuming that under normal circumstances, it usually takes 30 minutes from initiation to completion of a large transfer, the system will set the abnormal change threshold to 60 minutes. If a transaction is still not completed after 60 minutes, the system will automatically trigger the deviation transaction identification process. After the deviation transaction identification process is started, the system will extract all the information of the transaction from the transaction database, including the transaction amount, initiation time, and payee information. This information is merged with the historical deviation transaction data set to construct a more comprehensive cumulative deviation transaction data set. This approach can provide more sample data and help improve the accuracy of subsequent machine learning models. In this case, SVM is used to train a transaction path classification model. The goal of the model is to judge the rationality of the transaction path. For example, the model may consider the following features: transaction amount, transaction frequency, historical credit record of the counterparty, etc. By learning patterns in historical data, the SVM model can effectively distinguish between normal transaction paths and abnormal transaction paths. When the model identifies unreasonable transaction paths, the system extracts feature vectors of these paths. Feature vectors may include multiple dimensions, such as abnormal transaction amounts, high transaction frequency, and low counterparty credit ratings. These feature vectors provide a basis for subsequent association rule mining. In this scenario, the Apriori algorithm is used to analyze the feature vectors of unreasonable transaction paths to discover potential association rules. For example, the algorithm may find that the probability of the two features "transaction amount exceeds 1 million yuan" and "counterparty is a newly registered user" appearing at the same time is high, so a new rule can be generated: for transactions with an amount exceeding 1 million yuan and a counterparty being a newly registered user, additional manual review is required. Based on the association rules discovered by the Apriori algorithm, the system determines the corresponding adjustment parameters. These parameters may include rule weights, trigger thresholds, etc. For example, for the above rule, the system may set a higher weight and set the trigger threshold to 80% confidence. This means that only when the confidence of the rule exceeds 80% will it be considered for inclusion in the rule base. To ensure that the newly generated rules can indeed improve the accuracy of the system, the system will re-enter the adjusted rules into the transaction path classification model for verification. If the rule is judged to be reasonable after verification, it will be updated to the online rule base. This dynamic update mechanism ensures that the rule base can be continuously optimized as new transaction patterns emerge, improving the system's ability to identify new risks. Through this series of steps, financial institutions can continuously optimize their transaction monitoring systems. This not only improves the accuracy of the system in identifying abnormal transactions, but also adapts to the ever-changing financial environment.For example, in some cases, the trading behavior of many users has changed significantly. Traditional static rules may generate a large number of false positives. The system using this dynamic optimization method can quickly learn new trading patterns and reduce false positives while maintaining a high sensitivity to truly risky transactions.

[0032] The present invention provides a mobile payment supervision system based on aggregate payment transaction routing, which mainly includes: Deviation transaction identification module, used to obtain the actual transaction path and expected transaction path of each transaction, and determine whether there is a transaction deviation by comparing the difference between the actual transaction path and the expected transaction path. If the difference exceeds the preset threshold, the transaction is determined to be a deviation transaction and marked as a deviation transaction to be processed; The deviation type determination module is used to determine the deviation type of the identified deviation transactions using a classification algorithm. Based on the pre-established deviation type knowledge base, it comprehensively considers multi-dimensional features such as transaction amount, transaction time, and transaction object to determine the deviation type of each deviation transaction. The deviation cause analysis module is used to further use the association rule mining algorithm to mine the cause characteristics and association rules of different deviation types from the massive historical deviation transaction data after completing the deviation type determination, and establish a deviation cause knowledge base; The deviation cause determination module is used to obtain the corresponding cause characteristics and association rules from the deviation cause knowledge base for each deviation transaction according to its deviation type, and use the pattern matching algorithm to determine the specific cause of the deviation transaction; A disposal measure matching module is used to match applicable disposal measures from a pre-established disposal strategy knowledge base according to the type and cause of the deviation transaction; The disposal measure execution module is used to trigger the corresponding disposal process and monitor the execution progress and results of the disposal measures by realizing real-time information sharing and linkage with the trading system, risk control system, etc. after the disposal measures are determined; The continuous tracking and optimization module is used to continuously track the subsequent status changes of a single deviation transaction after completing the disposal of the transaction. If any abnormal changes occur, the deviation transaction identification process is re-triggered, and the transaction path model and rules are optimized and adjusted based on the accumulated deviation transaction data.

[0033] Based on the above embodiments of the present invention, relevant personnel can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the content in the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A mobile payment supervision method based on aggregated payment transaction routing, characterized in that: The method comprises: Obtain the actual transaction path and expected transaction path of each transaction, and determine whether there is a transaction deviation by comparing the difference between the actual transaction path and the expected transaction path. If the difference exceeds the preset threshold, the transaction is determined to be a deviation transaction and marked as a deviation transaction to be processed; A classification algorithm is used to determine the deviation type of the identified deviation transactions, and the deviation type of each deviation transaction is determined based on the pre-established deviation type knowledge base; After the deviation type is determined, the association rule mining algorithm is used to mine the cause characteristics and association rules of different deviation types from the historical deviation transaction data to establish a deviation cause knowledge base; According to the deviation type of each deviation transaction, the corresponding cause characteristics and association rules are obtained from the deviation cause knowledge base, and the specific cause of the deviation transaction is determined by using the pattern matching algorithm; According to the type and cause of deviation transactions, match the applicable disposal measures from the pre-established disposal strategy knowledge base; After determining the disposal measures, the corresponding disposal process is triggered through real-time information sharing and linkage with the trading system and risk control system, and the progress and results of the disposal measures are monitored; After completing the handling of a single deviation transaction, continue to track the subsequent status changes of the transaction. If any abnormal changes occur, re-trigger the deviation transaction identification process, and optimize and adjust the transaction path model and rules based on the accumulated deviation transaction data.

2. The method according to claim 1, characterized in that The actual transaction path and expected transaction path of each transaction are obtained, and by comparing the difference between the actual transaction path and the expected transaction path, whether there is a transaction deviation is determined. If the difference exceeds a preset threshold, the transaction is determined to be a deviation transaction and marked as a deviation transaction to be processed, including: Obtain the actual transaction path and expected transaction path of each transaction, and use the dynamic time warping algorithm to calculate the difference between the actual transaction path and the expected transaction path; Determine whether the difference exceeds a preset threshold. If so, mark the transaction as a pending deviation transaction. For deviation transactions marked as pending, obtain the detailed differences between the actual transaction path and the expected transaction path in terms of time, amount and key elements of the transaction object, and comprehensively determine the cause of the transaction deviation in combination with business rules and risk control strategies; According to the difference information and deviation reasons of the deviation transactions, the transaction time, amount and object characteristics are extracted, and the decision tree algorithm is used to classify the deviation transactions; For different types of deviation transactions classified by the decision tree algorithm, corresponding processing strategies are adopted to deal with them. If the deviation is caused by information entry errors, data correction is performed; If the deviation is caused by a system failure, the system troubleshooting and repair process is initiated; If the deviation is caused by abnormal operation, the compliance and rationality of the transaction will be judged in combination with business rules and authorization system; On the basis of deviation transaction processing, an anti-fraud model based on graph computing is adopted to extract transaction subject, time and amount elements from deviation transactions, build a transaction relationship graph, mine fraudulent transaction patterns, and identify and warn high-risk transactions in real time; By continuously evaluating the processing results of deviation transactions, dynamically optimizing preset thresholds and processing strategies, combining the recognition rate, processing accuracy and timeliness indicators of deviation transactions, and adjusting thresholds and iterating strategies, the management and control effects of deviation transactions are continuously improved.

3. The method according to claim 1, characterized in that The method of using a classification algorithm to determine the deviation type of the identified deviation transaction and determining the deviation type of each deviation transaction according to a pre-established deviation type knowledge base includes: Obtain transaction data, which includes transaction amount, transaction time and transaction object information; Clean the transaction data, remove duplicate and missing values, and convert the transaction data into a unified format; Select transaction amount, transaction time and transaction object from the cleaned transaction data as model training features; Normalize the model training features and scale the feature values ​​of the model training features to between 0 and 1; Divide the normalized model training features into training set and test set; The training set and random forest algorithm are used to train the deviation transaction determination model; The test set is input into the deviation transaction determination model to obtain the probability value of each transaction output by the deviation transaction determination model belonging to each deviation type; Obtain a preset probability threshold. If the probability value is greater than the probability threshold, the transaction is determined to belong to the corresponding deviation type. Obtain the pre-established deviation type rule library, which contains the feature rules and judgment conditions of each type of deviation transaction. The feature rules and judgment conditions are stored in the form of "feature-deviation type"; Match the transaction data determined as the deviation type with the feature rules and determination conditions in the deviation type rule base. If the match is successful, update the deviation type determination result of the transaction data to the deviation type corresponding to the feature rules and determination conditions; If the transaction is judged as a high-risk deviation type, a risk warning message will be generated and relevant personnel will be notified to handle it; storing the deviation type determination result of the transaction data together with the transaction data; Updating the deviation type rule base using the stored transaction data and the corresponding deviation type determination results; Use the updated deviation type rule library to reselect model training features from transaction data and perform normalization; The deviation transaction determination model is retrained using the reselected and normalized model training features and the updated deviation type determination results.

4. The method according to claim 1, characterized in that After the deviation type determination is completed, an association rule mining algorithm is used to mine the cause characteristics and association rules of different deviation types from the historical deviation transaction data, and a deviation cause knowledge base is established, including: Obtain historical deviation transaction data of determined deviation types, classify the historical deviation transaction data according to deviation types using the K-means clustering algorithm, and obtain sub-datasets of different deviation types; For each sub-dataset, PCA technology is used to extract the feature vectors related to the occurrence of deviations to obtain the feature data set; The Apriori algorithm is used to mine association rules on the feature data set to obtain the feature combinations that frequently occur for different deviation types; According to the feature combination, the correlation strength between the features is calculated. When the correlation strength exceeds the preset threshold, a strong correlation rule is determined; According to the strong association rules, the mapping relationship between deviation types and causes is established, and the deviation cause knowledge base is constructed; Acquire newly added deviation transaction data, use principal component analysis technology to extract features of the newly added deviation transaction data, and obtain features of the newly added deviation transaction; Input the newly added deviation transaction features into the decision tree model, and determine the deviation causes of the newly added deviation transaction data based on the strong association rules in the deviation cause knowledge base; The deviation causes of the newly added deviation transaction data are fed back to the deviation cause knowledge base, and the deviation cause knowledge base is updated.

5. The method according to claim 1, characterized in that According to the deviation type of each deviation transaction, the corresponding cause characteristics and association rules are obtained from the deviation cause knowledge base, and the specific cause of the deviation transaction is determined by using a pattern matching algorithm, including: Obtaining deviation transaction records and determining deviation types of the deviation transaction records; According to the deviation type, a set of cause features corresponding to the deviation type is obtained from a pre-established deviation cause knowledge base table, and the set of cause features includes time features, amount features, and transaction counterparty features; Obtain the transaction time of the deviation transaction record, match the transaction time with the time feature, and if the transaction time meets the preset time feature threshold, include the transaction time in the time feature matching result set; Obtain the transaction amount and counterparty information of the deviation transaction record, combine it with the historical transaction data, and use the Apriori algorithm to generate an association rule set. The association rule set is obtained by mining frequent item sets on the historical transaction data, setting support and confidence thresholds, and generating association rules that meet the conditions. Match the association rule set with the rule features in the cause feature set. If the match is successful, the rule features are classified into the rule feature matching result set; The time feature matching result set and the rule feature matching result set are input into a pre-trained decision tree model, and the decision tree model is used to determine the deviation cause of the deviation transaction record.

6. The method according to claim 1, characterized in that According to the type and cause of the deviation transaction, the applicable disposal measures are matched from the pre-established disposal strategy knowledge base, including: A1. Obtain deviation transaction records, extract the types and cause characteristics of deviation transactions, and construct deviation transaction feature vectors; A2. Use K-means clustering algorithm to cluster the feature vectors of deviation transactions and obtain the cluster centers of different disposal strategies; A3. Calculate the Euclidean distance between the current deviation transaction feature vector and each cluster center, and select the disposal strategy corresponding to the cluster center with the smallest distance as the candidate disposal strategy; A4. Extract association rules of candidate disposal strategies from the disposal strategy knowledge base, and use the association rule matching algorithm to determine whether the current deviation transaction meets the triggering conditions of the candidate disposal strategy; If the trigger condition is met, the candidate disposal strategy is added to the set of disposal strategies to be selected; Otherwise, the candidate disposal strategy is excluded, and the disposal strategy corresponding to the next cluster center with the smallest distance is selected as a new candidate disposal strategy, and steps A3-A4 are repeated until a disposal strategy that meets the trigger condition is found or all candidate disposal strategies are traversed; A5. Sort the strategies in the set of disposal strategies to be selected according to the preset priority rules, and select the disposal strategy with the highest priority as the final disposal measure; A6. Generate a detailed disposal plan based on the final disposal measures, combined with business rules and system logic, including disposal time, disposal object, disposal amount, and disposal reason; A7. Send the generated detailed disposal plan to the relevant business system for actual disposal operation, and feed back the disposal execution results to the disposal strategy knowledge base; A8. If the disposal result does not meet expectations, the association rules in the disposal strategy knowledge base are adjusted. By modifying the conditions and conclusions of the association rules, the matching accuracy of the disposal strategy is optimized and the disposal effect of the deviation transaction is improved; A9. The adjusted association rules will be applied to the subsequent deviation transaction disposal process to continuously optimize the performance of the disposal strategy.

7. The method according to claim 1, characterized in that After determining the disposal measures, the corresponding disposal process is triggered by achieving real-time information sharing and linkage with the trading system and risk control system, and the progress and results of the disposal measures are monitored, including: Obtain transaction data reported by the transaction system; For transaction data, the number of transactions of a certain type within a sliding window of fixed time length is counted as the transaction frequency; If the transaction frequency exceeds the preset frequency threshold, an alert is triggered; For transaction data, regular expressions are used to determine whether the transaction type complies with the preset risk control rules; If the preset risk control rules are met, the transaction risk level is determined through the decision tree model; If both the transaction risk level and transaction frequency exceed the preset thresholds, the transaction risk score is calculated using the weighted average method, where the weights of transaction frequency and transaction risk level are 3 and 7 respectively; If the transaction risk score exceeds the preset score threshold, the transaction disposal measures will be determined according to the range of the transaction risk score. The transaction disposal measures include freezing the account, restricting transactions and manual review. Generate a disposal instruction based on the transaction disposal measures, which includes the disposal measures, transaction account, transaction time, transaction amount and transaction type; According to the disposal instructions, the trading system and risk control system perform corresponding operations. Among them, the trading system performs account freezing and transaction restriction operations, and the risk control system performs manual review operations; The trading system and risk control system use the execution results as execution feedback information, which includes the execution results, execution time and executor; If the execution result in the execution feedback information is successful, the disposal result meets the expected goal; If the execution result in the execution feedback information is failure, the disposal result does not meet the expected goal; If the disposal result meets the expected goal, the status of the transaction record is updated to disposed; If the disposal result does not meet the expected goal, the status of the transaction record is updated to disposal failure.

8. The method according to claim 1, characterized in that After the handling of a single deviation transaction is completed, the subsequent status changes of the transaction are continuously tracked. If an abnormal change occurs, the deviation transaction identification process is re-triggered, and the transaction path model and rules are optimized and adjusted based on the accumulated deviation transaction data, including: Obtain the status data of a single deviation transaction after disposal, and use the preset abnormal change threshold to make a judgment. If the status data exceeds the threshold, the deviation transaction identification process is triggered; If the deviation transaction identification process is triggered, all data information of the transaction is extracted from the transaction database and merged with the historical deviation transaction data set to construct a cumulative deviation transaction data set; The support vector machine algorithm is used to train the cumulative deviation transaction data set to obtain the transaction path classification model; The rationality of the transaction path is judged through the transaction path classification model. If there is an unreasonable path, the feature vector of the unreasonable path is extracted; Apriori algorithm is used to mine association rules for unreasonable path feature vectors, obtain optimization rules, and determine the adjustment parameters of the rules; Input the adjusted rules into the transaction path classification model and judge the rationality of the rules again. If the rules are reasonable, update the online rule library.

9. A mobile payment supervision system based on aggregated payment transaction routing, characterized in that: The system comprises: Deviation transaction identification module, used to obtain the actual transaction path and expected transaction path of each transaction, and determine whether there is a transaction deviation by comparing the difference between the actual transaction path and the expected transaction path. If the difference exceeds the preset threshold, the transaction is determined to be a deviation transaction and marked as a deviation transaction to be processed; The deviation type determination module is used to determine the deviation type of the identified deviation transactions using a classification algorithm, and determine the deviation type of each deviation transaction based on a pre-established deviation type knowledge base; The deviation cause analysis module is used to mine the cause characteristics and association rules of different deviation types from the historical deviation transaction data after completing the deviation type determination, and establish a deviation cause knowledge base; The deviation cause determination module is used to obtain the corresponding cause characteristics and association rules from the deviation cause knowledge base for each deviation transaction according to its deviation type, and use the pattern matching algorithm to determine the specific cause of the deviation transaction; A disposal measure matching module is used to match applicable disposal measures from a pre-established disposal strategy knowledge base according to the type and cause of the deviation transaction; The disposal measure execution module is used to trigger the corresponding disposal process and monitor the execution progress and results of the disposal measures by realizing real-time information sharing and linkage with the trading system, risk control system, etc. after the disposal measures are determined; The continuous tracking and optimization module is used to continuously track the subsequent status changes of a single deviation transaction after completing the disposal of the transaction. If any abnormal changes occur, the deviation transaction identification process is re-triggered, and the transaction path model and rules are optimized and adjusted based on the accumulated deviation transaction data.

Images