Payment risk control platform decision flow execution method, apparatus and device based on threshold dynamic tuning, and medium

Abstract

The invention discloses a payment risk control platform decision flow execution method, device and equipment based on threshold dynamic tuning, and a medium, and relates to the technical field of intelligent decision and dynamic optimization, and the method comprises the steps: obtaining the current shopping order transaction data of a payment risk control platform; performing risk control decision-making on the current shopping order transaction data by using a current decision-making rule in the target decision-making flow to obtain current order decision-making data; the target decision flow is a shopping order risk control decision flow of the platform, and the current decision rule is a transaction risk control rule of the platform and comprises each current rule threshold; obtaining a decision flow operation index based on the current order decision data; matching the decision flow operation index with each preset threshold adjustment and optimization triggering condition, and determining a target adjustment and optimization model to obtain each next rule threshold; and obtaining a next decision rule based on the hot update mechanism and the next rule threshold so as to carry out risk control decision. And the threshold adaptability, the optimization scientificity and the service continuity in the risk control decision flow are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent decision-making and dynamic optimization, and particularly relates to a payment risk control platform decision flow execution method and device based on threshold dynamic optimization, equipment and a medium. BACKGROUND

[0002] In the transaction order risk control scene, the static rule threshold is usually used in the risk control decision flow, which is usually set based on artificial experience. However, in actual business, transaction data characteristics (such as order amount, transaction regional distribution) frequently fluctuate with the scene (such as promotion activities, user behavior changes), and the traditional decision flow only focuses on the decision result index and lacks the whole link monitoring of the rule running process and data distribution, which leads to the difficulty in accurately locating the abnormal source when the rule threshold does not match the actual business, and the threshold adjustment depends on the experience judgment, and the optimization lacks scientificity. In addition, the traditional threshold update needs to restart the decision flow engine, which is easy to interrupt the real-time transaction risk control business, causing the delay or failure of normal order processing, which cannot adapt to the dynamically changing transaction scene to ensure the accuracy of the risk control decision and the user experience, and it is also difficult to balance the business continuity and the risk control effect.

[0003] In summary, how to improve the threshold adaptability, optimization scientificity and business continuity in the risk control decision flow is a problem to be solved in the field. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a payment risk control platform decision flow execution method and device based on threshold dynamic optimization, which improves the threshold adaptability, optimization scientificity and business continuity in the risk control decision flow. The specific scheme is as follows:

[0005] In a first aspect, the present application discloses a payment risk control platform decision flow execution method based on threshold dynamic optimization, comprising:

[0006] obtaining current shopping order transaction data of a payment risk control platform;

[0007] performing risk control decision on the current shopping order transaction data by using a current decision rule in a target decision flow to obtain current order decision data; wherein the target decision flow is a shopping order risk control decision flow of the payment risk control platform, and the current decision rule is a transaction risk control rule of the payment risk control platform and includes each current rule threshold;

[0008] obtaining a decision flow running index corresponding to the current decision rule based on the current order decision data; wherein the decision flow running index includes a rule running state index, a feature data distribution index and a decision effect index;

[0009] match the decision flow running index with each preset threshold value tuning trigger condition, and determine a target tuning model based on a matching result, input each current rule threshold value in the current decision rule into the target tuning model to obtain each next rule threshold value; wherein the target tuning model comprises a tuning direction control item and a tuning amplitude calculation item;

[0010] write the next rule threshold value into the target decision flow based on a hot update mechanism to obtain a next decision rule, and use the next decision rule to make a risk control decision on a next shopping order transaction data.

[0011] Optionally, the rule running state index comprises a rule trigger rate, a rule average time consumption, the feature data distribution index comprises a feature mean value, a feature standard deviation and a distribution change rate, and the decision effect index comprises a decision accuracy rate, a decision precision rate and a threshold value sensitivity; the obtaining of the decision flow running index corresponding to the current decision rule based on the current order decision data comprises:

[0012] obtaining a rule trigger number, a rule total execution number and an execution time consumption of each rule execution of the current decision rule from the current order decision data, determining a rule trigger rate as a ratio of the rule trigger number to the rule total execution number, and determining a rule average time consumption according to the execution time consumption of each time;

[0013] determining a feature mean value, a feature standard deviation and a distribution change rate based on a feature value and a sample size in the current order decision data;

[0014] determining a decision accuracy rate as a ratio of a correct decision number to a total decision number in the current order decision data, determining a decision precision rate as a ratio of a true positive decision number to a positive decision number in the current order decision data, and determining a threshold value sensitivity as a ratio of a change amount of the decision accuracy rate in a first preset time window to a threshold value change amount in the current order decision data.

[0015] Optionally, the matching of the decision flow running index with each preset threshold value tuning trigger condition, and the determination of a target tuning model based on a matching result, the input of each current rule threshold value in the current decision rule into the target tuning model to obtain each next rule threshold value, comprises:

[0016] merging each sub result obtained by matching the decision flow running index with each preset threshold value tuning trigger condition respectively to obtain a matching result;

[0017] if it is determined according to the matching result that the decision flow running index only meets any one of the preset threshold value tuning trigger conditions, then determining that a target tuning mode of the current rule threshold value is a single model tuning mode;

[0018] If it is determined according to the matching result that the decision flow running index satisfies at least two of the preset threshold tuning trigger conditions, it is determined that the target tuning mode of the current rule threshold is a combined model tuning mode.

[0019] According to the target tuning mode and the matching result, a target tuning model is determined, and each current rule threshold in the current decision rule is input into the target tuning model to obtain each next rule threshold.

[0020] Optionally, the decision flow running index is matched with each preset threshold tuning trigger condition respectively to obtain each sub-result, including:

[0021] The decision accuracy in the decision flow running index is matched with a first preset threshold tuning trigger condition, and if the decision accuracy satisfies the first preset threshold tuning trigger condition, a first sub-result representing decision accuracy anomaly in decision effect is generated; wherein the first preset threshold tuning trigger condition is that the decision accuracy is lower than a preset accuracy threshold and the duration is greater than a first preset time length.

[0022] The rule triggering rate in the decision flow running index is matched with a second preset threshold tuning trigger condition, and if the rule triggering rate satisfies the second preset threshold tuning trigger condition, a second sub-result representing rule triggering imbalance is generated; wherein the second preset threshold tuning trigger condition is that the rule triggering rate is higher than a first preset triggering rate threshold or lower than a second preset triggering rate threshold.

[0023] The distribution change rate in the decision flow running index is matched with a third preset threshold tuning trigger condition, and if the distribution change rate satisfies the third preset threshold tuning trigger condition, a third sub-result representing feature distribution mutation is generated; wherein the third preset threshold tuning trigger condition is that the distribution change rate is higher than a preset change rate threshold and a KS test value of the distribution change rate is less than a preset test value.

[0024] The threshold sensitivity in the decision flow running index is matched with a fourth preset threshold tuning trigger condition, and if the threshold sensitivity satisfies the fourth preset threshold tuning trigger condition, a fourth sub-result representing threshold sensitivity anomaly in decision effect is generated; wherein the fourth preset threshold tuning trigger condition is that the threshold sensitivity is higher than a preset sensitivity threshold.

[0025] Optionally, according to the target tuning mode and the matching result, a target tuning model is determined, and each current rule threshold in the current decision rule is input into the target tuning model to obtain each next rule threshold, including:

[0026] If the target tuning mode is a single model tuning mode, a single target tuning model corresponding to a sub-result in the matching result is determined, and each current rule threshold in the current decision rule is input into the target tuning model to obtain each next rule threshold;

[0027] If the target tuning mode is a combined model tuning mode, a plurality of target tuning models corresponding to sub-results in the matching result are determined, and each current rule threshold in the current decision rule is input into the target tuning model, each initial rule threshold output by each target tuning model is weighted and summed based on a preset threshold weight to obtain each next rule threshold;

[0028] Wherein, the tuning direction in the tuning direction control item of the target tuning model is determined based on the deviation of the decision flow running index, and the tuning amplitude in the tuning amplitude calculation item is determined based on the deviation proportion of the decision flow running index, the deviation of the decision flow running index is the difference between the decision flow running index and the target index value, and the deviation proportion of the decision flow running index is the ratio of the absolute value of the deviation of the decision flow running index to the target index value.

[0029] Optionally, the writing of the next rule threshold into the target decision flow based on the hot update mechanism comprises:

[0030] Introducing the next rule threshold into a gray rule set of a decision flow engine to use the next rule threshold to make risk control decisions on a preset number of current shopping order transaction data as gray traffic to obtain a decision effect index of the gray traffic;

[0031] If the decision effect index of the gray traffic meets a preset check pass condition, the next rule threshold is written into the target decision flow based on the hot update mechanism;

[0032] If the decision effect index of the gray traffic does not meet the preset check pass condition, the preset threshold weight is adjusted, and the step of weighting and summing each initial rule threshold output by each target tuning model based on the preset threshold weight is re-executed.

[0033] Optionally, the writing of the next rule threshold into the target decision flow based on the hot update mechanism comprises:

[0034] The next rule threshold is smoothed using a preset smoothing factor to obtain a smoothed next rule threshold;

[0035] The smoothed next rule threshold is written into the target decision flow based on the hot update mechanism.

[0036] In a second aspect, the application discloses a payment risk control platform decision flow execution device based on threshold dynamic optimization, comprising:

[0037] a data acquisition module, configured to acquire current shopping order transaction data of a payment risk control platform;

[0038] a data decision module, configured to perform risk control decision on the current shopping order transaction data by using current decision rules in a target decision flow, to obtain current order decision data; wherein the target decision flow is a shopping order risk control decision flow of the payment risk control platform, and the current decision rules are transaction risk control rules of the payment risk control platform and comprise respective current rule thresholds;

[0039] an index acquisition module, configured to acquire decision flow running indexes corresponding to the current decision rules based on the current order decision data; wherein the decision flow running indexes comprise rule running state indexes, feature data distribution indexes and decision effect indexes;

[0040] a threshold optimization module, configured to match the decision flow running indexes with respective preset threshold optimization trigger conditions, and determine a target optimization model based on a matching result, input respective current rule thresholds in the current decision rules into the target optimization model, to acquire respective next rule thresholds; wherein the target optimization model comprises an optimization direction control item and an optimization amplitude calculation item;

[0041] a decision flow execution module, configured to write the next rule thresholds into the target decision flow based on a hot update mechanism, to obtain next decision rules, and perform risk control decision on next shopping order transaction data by using the next decision rules.

[0042] In a third aspect, the application discloses an electronic device, comprising:

[0043] a memory, configured to save a computer program;

[0044] a processor, configured to execute the computer program, to realize steps of the payment risk control platform decision flow execution method based on threshold dynamic optimization disclosed in the foregoing.

[0045] In a fourth aspect, the application discloses a computer readable storage medium, configured to store a computer program; wherein the computer program is executed by a processor to realize steps of the payment risk control platform decision flow execution method based on threshold dynamic optimization disclosed in the foregoing.

[0046] The beneficial effects of this application are as follows: This application obtains current shopping order transaction data from a payment risk control platform; utilizes the current decision rules in the target decision flow to make risk control decisions on the current shopping order transaction data to obtain current order decision data; wherein, the target decision flow is the shopping order risk control decision process of the payment risk control platform, and the current decision rules are the transaction risk control rules of the payment risk control platform and include each current rule threshold; based on the current order decision data, it obtains decision flow operation indicators corresponding to the current decision rules; wherein, the decision flow operation indicators include rule operation status indicators, feature data distribution indicators, and decision effect indicators; it matches the decision flow operation indicators with each preset threshold optimization trigger condition, and determines the target optimization model based on the matching results, inputs each current rule threshold in the current decision rules into the target optimization model to obtain each next rule threshold; wherein, the target optimization model includes optimization direction control items and optimization magnitude calculation items; based on a hot update mechanism, it writes the next rule threshold into the target decision flow to obtain the next decision rule, and uses the next decision rule to make risk control decisions on the next shopping order transaction data. Therefore, this application obtains current shopping order transaction data from the payment risk control platform and conducts risk control decisions based on the platform's exclusive shopping order risk control decision-making process and transaction risk control rules. Based on the current order decision data, it acquires comprehensive decision flow operation indicators covering rule operation status, feature data distribution, and decision effectiveness. This allows for a complete understanding of the operation process, core data feature changes, and decision quality of the payment risk control platform's shopping order risk control decision flow, providing precise data support for rule threshold optimization tailored to the payment shopping order scenario. By matching the decision flow operation indicators with preset threshold optimization trigger conditions and calling the target optimization model to calculate the next rule threshold, the optimization direction control item ensures that the threshold adjustment direction aligns with the core objectives of the payment risk control platform's shopping order risk control. The optimization magnitude calculation item accurately... By controlling the granularity of threshold adjustments, the subjectivity of manual optimization is avoided, improving the scientific rigor and accuracy of threshold optimization for shopping order risk control rules on the payment risk control platform. This ensures the accuracy of shopping order risk control decisions and reduces the false interception rate of normal orders. Based on a hot update mechanism, the next rule threshold is written into the target decision stream. This allows for rule updates without interrupting the real-time risk control service for shopping orders on the payment risk control platform, ensuring the continuity and stability of payment risk control operations. Simultaneously, the updated next decision rule is used to conduct subsequent shopping order risk control decisions, forming a dynamic closed loop of "decision-monitoring-optimization-update" adapted to the shopping order scenario of the payment risk control platform. This continuously adapts to the dynamic changes in shopping order transaction data and business scenarios, improving the overall adaptability and operational stability of the payment risk control platform's shopping order risk control decisions. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of 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 only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0048] Figure 1 This application discloses a flowchart of a decision flow execution method for a payment risk control platform based on dynamic threshold adjustment.

[0049] Figure 2 This is a schematic diagram of the decision flow execution device of a payment risk control platform based on threshold dynamic optimization disclosed in this application;

[0050] Figure 3 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0051] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0052] In transaction order risk control scenarios, the current risk control decision flow mostly adopts static rule thresholds, which are usually pre-set based on human experience. However, in actual business, transaction data characteristics (such as order amount and transaction geographical distribution) fluctuate frequently with the scenario (such as promotional activities and changes in user behavior). At the same time, the traditional decision flow only focuses on decision result indicators and lacks full-link monitoring of the rule operation process and data distribution. This makes it difficult to accurately locate the root cause of the anomaly when the rule threshold does not match the actual business. Moreover, manual adjustment of thresholds relies on experience judgment, and the optimization lacks scientificity. In addition, traditional threshold updates require restarting the decision flow engine, which can easily interrupt real-time transaction risk control business and cause delays or failures in normal order processing. It is neither able to adapt to dynamically changing transaction scenarios to ensure the accuracy of risk control decisions and user experience, nor can it balance business continuity and risk control effectiveness.

[0053] To address this, this application provides a decision flow execution scheme for a payment risk control platform based on dynamic threshold optimization, thereby improving the threshold adaptability, optimization scientificity, and business continuity in the risk control decision flow.

[0054] See Figure 1 As shown in the figure, this application discloses a decision flow execution method for a payment risk control platform based on dynamic threshold adjustment, including:

[0055] Step S11: Obtain the current shopping order transaction data from the payment risk control platform.

[0056] The payment risk control platform receives real-time, full-volume transaction payment data associated with the current shopping order from upstream systems such as e-commerce transaction systems and payment channels via pre-set multi-source data access interfaces. The acquired data must cover the core dimensions required for making risk control decisions for the shopping order, specifically including basic order information, user identity information, payment information, transaction environment information, and order product information. Basic order information includes order number, order generation time, total order amount, product category, and order payment timeliness requirements. User identity information includes the user's unique identifier on the payment risk control platform, user real-name authentication status, user's historical payment credit rating, and the user's commonly used shipping address and contact information. Payment information includes anonymized bank card number, payment channel type, and payment account. The payment risk control platform collects and processes the above data in real time, including balance status, payment password / verification code verification status, and bank card payment risk level. Transaction environment information includes the transaction terminal device model, device login IP address, IP address location, device fingerprint information, and network operator type. Order product information includes unit price, quantity, shipping and receiving locations, and risk category of the product. The platform performs real-time format verification and data cleaning on the above data, removing invalid, missing, and abnormal data to ensure data integrity, accuracy, and timeliness. This provides a reliable data foundation for subsequent precise risk control decisions based on target decision flows. Furthermore, the data access process must comply with payment industry data security compliance requirements, encrypting sensitive user information to protect user data privacy and transaction data security.

[0057] Step S12: Use the current decision rules in the target decision flow to make risk control decisions on the current shopping order transaction data to obtain the current order decision data; wherein, the target decision flow is the shopping order risk control decision process of the payment risk control platform, and the current decision rules are the transaction risk control rules of the payment risk control platform and include the threshold values ​​of each current rule.

[0058] The process of using the current decision rules in the target decision flow to make risk control decisions on the current shopping order transaction data to obtain the current order decision data is the core execution link of the payment risk control platform to achieve risk identification based on the exclusive shopping order risk control decision process. The target decision flow is a three-level risk control decision link pre-built by the payment risk control platform for shopping order scenarios, namely "pre-filtering - core rule judgment - bottom-line compliance verification". This link has been solidified in the platform's risk control engine and can adapt to the risk prevention and control needs of the entire process of shopping orders from order placement to payment. The current decision rules are an exclusive transaction risk control rule system configured by the payment risk control platform in combination with the transaction characteristics of shopping orders, and include various current rule thresholds. The current rule thresholds are, for example, the payment order amount verification threshold, the transaction IP region compliance threshold, the user payment behavior abnormality number threshold, the bank card payment risk level threshold, the high-risk product order verification threshold, and the matching degree threshold between the delivery address and the commonly used address, etc.

[0059] In actual decision-making, the payment risk control platform first maps the current shopping order transaction data, which has undergone format verification and cleaning, to the corresponding data input nodes in the target decision flow according to fields. Then, the decision-making process is initiated. First, it enters the pre-filtering layer, quickly filtering out low-risk orders based on basic rules such as the user's historical payment credit rating and the order's product category. Then, it enters the core rule judgment layer, comparing the order data with each current rule threshold. For example, it checks whether the order amount exceeds the payment order amount verification threshold, whether the transaction IP is within the geographical compliance threshold range, whether the number of abnormal payment behaviors by the user in a single day reaches the set threshold, whether the risk level of the bound bank card matches the threshold requirements, whether the product is a high-risk category requiring additional verification, and whether the matching degree between the delivery address and the frequently used address is lower than the preset threshold. If a match is found... Any core rule threshold marks the corresponding risk point, and then the order enters the bottom compliance verification layer to verify the compliance of the payment industry regulatory system, ensuring that it complies with regulatory requirements such as the spending limit for minors. Throughout the decision-making process, the platform records the execution status of each rule node, data comparison results, and rule hit status in real time. Finally, it integrates the judgment results of each level to generate the current order decision data. This data not only includes the final risk control judgment conclusion of the order, but also covers specific rule hit details, risk level classification results, decision link execution logs, and compliance verification records. This provides a complete decision process and result basis for the collection of subsequent decision flow operation indicators. Moreover, the entire decision-making process must meet the real-time requirements of the payment risk control platform for order processing, and the decision-making time for a single order shall not exceed a preset threshold to ensure the smoothness of the shopping order payment process.

[0060] Step S13: Based on the current order decision data, obtain the decision flow operation indicators corresponding to the current decision rule; wherein, the decision flow operation indicators include rule operation status indicators, feature data distribution indicators, and decision effect indicators.

[0061] This step involves data monitoring and quantitative analysis after the payment risk control platform completes the risk control decision for a shopping order. Based on the generated current order decision data, combined with the corresponding order's original transaction data and decision chain execution logs, the platform extracts and calculates decision flow operation indicators directly related to the current decision rule at a preset differentiated collection frequency. These indicators include rule operation status indicators, feature data distribution indicators, and decision effect indicators. Specifically, rule operation status indicators and decision effect indicators are collected every 100ms, and feature data distribution indicators are calculated every minute based on a sliding window (window size 5 minutes). All indicators are structured and stored in conjunction with the decision flow ID, rule node ID, and timestamp. In other words, indicators are categorized and stored in the indicator engine by timestamp + rule ID. Specifically, the platform can retain the raw data from the most recent 24 hours (for optimization) and aggregated data from the past 30 days (for trend analysis). This provides accurate, comprehensive, and relevant data support for subsequent threshold optimization trigger judgments, tailored to the payment risk control shopping order scenario.

[0062] In this embodiment, the rule operation status indicators include rule trigger rate and average rule execution time; the feature data distribution indicators include feature mean, feature standard deviation, and distribution change rate; and the decision performance indicators include decision accuracy, decision precision, and threshold sensitivity. The step of obtaining the decision flow operation indicators corresponding to the current decision rule based on the current order decision data includes: obtaining the rule trigger count, total rule execution count, and execution time of each rule execution from the current order decision data; determining the rule trigger rate as the ratio of the rule trigger count to the total rule execution count; determining the average rule execution time based on the execution time of each rule execution; determining the feature mean, feature standard deviation, and distribution change rate based on the feature values ​​and sample size in the current order decision data; determining the decision accuracy as the ratio of the number of correct decisions to the total number of decisions in the current order decision data; determining the decision precision as the ratio of the number of true positive decisions to the number of positive decisions in the current order decision data; and determining the threshold sensitivity as the ratio of the change in decision accuracy within the first preset time window to the change in the threshold in the current order decision data.

[0063] The platform conducts precise calculations based on current order decision data and associated original transaction data and decision execution logs for three types of decision flow operation indicators.

[0064] For the rule trigger rate in the rule operation status indicators, the platform first extracts the rule trigger count (i.e., the number of times the order hits the corresponding rule) and the total number of rule executions (i.e., the total number of times the order flows through the corresponding rule) of each current decision rule from the current order decision data. The rule trigger rate is calculated by the ratio of the two. The platform then extracts the specific time consumption data for each rule execution from the current order decision data and calculates the average rule execution time (in milliseconds) by averaging the rule execution time for this batch of orders. This quantifies the actual trigger frequency and execution efficiency of each risk control rule. The rule trigger rate... Average time for rules The calculation formula is:

[0065] ;

[0066] ;

[0067] For the feature data distribution indicators such as feature mean, feature standard deviation, and distribution change rate, the platform will retrieve the specific feature values ​​of core features such as order amount, delivery address matching degree, and transaction IP geographic correlation degree from the current order decision data. Based on the sample size m of this batch of orders, the characteristic mean is obtained through statistical calculation. With characteristic standard deviation Then, the current feature mean and standard deviation are compared with the corresponding feature statistics of the historical benchmark window to calculate the feature distribution change rate. This reflects the distribution patterns and fluctuations of the core transaction characteristics of shopping orders; among them, the characteristic mean... Characteristic standard deviation and the rate of change of characteristic distribution The calculation formula is as follows:

[0068] ;

[0069] ;

[0070] ;

[0071] For the decision-making effectiveness metrics, namely decision accuracy (A), decision precision (P), and threshold sensitivity (S), the platform filters the current order decision data to determine the number of correct decisions (i.e., the number of orders whose judgments are consistent with the actual risk situation) and the total number of decisions. The ratio of these two values ​​determines the decision accuracy. Then, it extracts the number of true positive decisions (i.e., the number of orders judged as risky and actually being risky) and the number of positive decisions (i.e., all orders judged as risky). The ratio of these two values ​​determines the decision precision. Simultaneously, a first preset time window is defined, and the absolute change in decision accuracy within this time window is statistically analyzed. This is then combined with the change in the corresponding rule threshold in the current order decision data during the same period, and the ratio of these two values ​​is used to calculate the threshold sensitivity. This assesses the impact of threshold adjustments on decision-making effectiveness. The calculation formulas for decision accuracy (A), decision precision (P), and threshold sensitivity (S) are as follows:

[0072] ;

[0073] ;

[0074] ;

[0075] In the formula, The change in decision accuracy. This represents the change in the threshold.

[0076] Step S14: Match the decision flow operation index with each preset threshold tuning trigger condition, and determine the target tuning model based on the matching result. Input each current rule threshold in the current decision rule into the target tuning model to obtain each next rule threshold; wherein, the target tuning model includes tuning direction control terms and tuning magnitude calculation terms.

[0077] In this embodiment, the step of matching the decision flow operation index with each preset threshold tuning trigger condition, determining the target tuning model based on the matching result, and inputting each current rule threshold in the current decision rule into the target tuning model to obtain each next rule threshold includes: merging the sub-results obtained by matching the decision flow operation index with each preset threshold tuning trigger condition to obtain a matching result; if the matching result determines that the decision flow operation index only satisfies any one of the preset threshold tuning trigger conditions, then the target tuning mode of the current rule threshold is determined to be a single model tuning mode; if the matching result determines that the decision flow operation index satisfies at least two of the preset threshold tuning trigger conditions, then the target tuning mode of the current rule threshold is determined to be a combined model tuning mode; determining the target tuning model based on the target tuning mode and the matching result, and inputting each current rule threshold in the current decision rule into the target tuning model to obtain each next rule threshold.

[0078] The acquired rule operation status indicators (rule trigger rate, average rule time), feature data distribution indicators (feature mean, feature standard deviation, distribution change rate), and decision performance indicators (decision accuracy, decision precision, threshold sensitivity) are compared one by one with the preset threshold optimization trigger conditions. The preset threshold optimization trigger conditions set clear abnormal threshold ranges for each type of indicator. Then, the sub-results obtained by matching each indicator with the corresponding conditions are integrated to form a complete matching result.

[0079] The platform will then determine the optimization mode based on the matching result. If the matching result shows that the decision flow operation index only meets any one of the preset threshold optimization trigger conditions, such as only the decision accuracy is lower than the target threshold while other indicators are normal, or only the feature distribution change rate exceeds the standard while the rule operation status and decision effect indicators are normal, then the target optimization mode for the current rule threshold is determined to be the single model optimization mode. At this time, the scenario-specific optimization model corresponding to this type of abnormal index will be matched. If the matching result shows that the decision flow operation index meets at least two preset threshold optimization trigger conditions at the same time, such as the decision accuracy not meeting the standard and the feature distribution change occurring simultaneously, or the average rule time exceeding the standard and the threshold sensitivity being abnormal, then the target optimization mode is determined to be the combined model optimization mode. At this time, the multi-scenario fusion combined optimization model will be activated. The platform determines the corresponding target optimization model based on the selected target optimization mode and specific matching results. In the single-model optimization mode, the optimization model only includes optimization direction control items and optimization magnitude calculation items for a single abnormal scenario. In the combined model optimization mode, the threshold values ​​of each current rule in the current decision rule are input into the selected target optimization model. The model first determines the direction of threshold adjustment based on the optimization direction control items, and then precisely controls the adjustment granularity through the optimization magnitude calculation items. The combined model optimization mode will further weight and fuse the candidate thresholds calculated in multiple scenarios, and finally output the next rule threshold corresponding to each rule, providing a scientific, accurate and relevant threshold basis for subsequent decision flow rule updates that fits the payment risk control shopping order scenario.

[0080] The following section elaborates on the judgment logic for matching the tuning trigger conditions.

[0081] The decision flow operation indicators are matched with each preset threshold optimization trigger condition to obtain sub-results, including: matching the decision accuracy rate in the decision flow operation indicators with a first preset threshold optimization trigger condition; if the decision accuracy rate meets the first preset threshold optimization trigger condition, a first sub-result representing an abnormal decision accuracy rate in the decision effect is generated; wherein, the first preset threshold optimization trigger condition is that the decision accuracy rate is lower than a preset accuracy rate threshold and the duration is greater than a first preset duration; matching the rule trigger rate in the decision flow operation indicators with a second preset threshold optimization trigger condition; if the rule trigger rate meets the second preset threshold optimization trigger condition, a second sub-result representing an imbalance in rule triggering is generated; wherein, the second preset threshold optimization trigger condition is that the rule trigger rate is higher than... The first preset trigger rate threshold is lower than the second preset trigger rate threshold; the distribution change rate in the decision flow operation index is matched with the third preset threshold optimization trigger condition. If the distribution change rate meets the third preset threshold optimization trigger condition, a third sub-result representing a sudden change in the feature distribution is generated; wherein, the third preset threshold optimization trigger condition is that the distribution change rate is higher than the preset change rate threshold and the KS test value of the distribution change rate is less than the preset test value; the threshold sensitivity in the decision flow operation index is matched with the fourth preset threshold optimization trigger condition. If the threshold sensitivity meets the fourth preset threshold optimization trigger condition, a fourth sub-result representing an abnormal threshold sensitivity in the decision effect is generated; wherein, the fourth preset threshold optimization trigger condition is that the threshold sensitivity is higher than the preset sensitivity threshold.

[0082] The decision flow operation indicators are verified one by one according to the four preset special trigger conditions, and corresponding abnormal sub-results are generated, as shown in Table 1:

[0083] Table 1 Decision Logic

[0084]

[0085] First, the decision accuracy data from the decision flow operation metrics is retrieved and precisely matched with the first preset threshold optimization trigger condition. This first preset threshold optimization trigger condition is that the decision accuracy A is lower than the risk control decision accuracy threshold preset by the payment risk control platform for shopping order scenarios. Furthermore, if the duration of the low accuracy state is longer than the first preset duration (e.g., 3 collection cycles), and the decision accuracy meets both of these requirements, the first sub-result representing the abnormal decision accuracy in the decision effect dimension will be generated immediately to mark the abnormal situation where the risk control decision quality fails to meet the business objectives.

[0086] Next, the platform extracts the rule trigger rate data from the decision flow operation metrics and compares it with the second preset threshold optimization trigger condition, which sets the rule trigger rate. The bidirectional abnormal interval, that is, the rule trigger rate is higher than the first preset trigger rate threshold set for the risk control rules of shopping orders. Or below the second preset trigger rate threshold ,in, As long as the rule trigger rate falls into any abnormal range, a second sub-result is generated to represent the imbalance of rule triggering in the rule operation state dimension, thereby achieving accurate identification of abnormal rule execution frequency.

[0087] The platform then obtains the feature distribution change rate data from the decision flow operation indicators and verifies it against the third preset threshold optimization trigger condition. This third preset threshold optimization trigger condition is a dual judgment criterion, requiring both the distribution change rate and the feature distribution change rate. The mean change rate of the feature is higher than the distribution fluctuation threshold preset by the payment risk control platform for the core transaction features of shopping orders. Furthermore, it is required that the KS test value corresponding to the change in this distribution is obtained. If the value is less than the preset test value (e.g., 0.05), only if both conditions are met simultaneously will a third sub-result be generated to represent the abrupt change in the feature distribution dimension of the feature data distribution, thus completing the determination of abnormal patterns in the transaction data features.

[0088] Finally, the platform retrieves the threshold sensitivity data from the decision flow operation indicators and matches it with the fourth preset threshold optimization trigger condition. This fourth preset threshold optimization trigger condition is that the threshold sensitivity S is higher than the sensitivity threshold preset by the payment risk control platform. , specifically If the threshold sensitivity reaches the standard, a fourth sub-result representing the threshold sensitivity anomaly in the decision effect dimension is generated, thus completing the marking of the decision rule stability anomaly.

[0089] All sub-results will be linked to the corresponding abnormal data, judgment criteria, and timestamps for unified storage, providing clear support for the determination of subsequent optimization modes and models.

[0090] The following section elaborates on the use of tuning models to optimize thresholds.

[0091] In this embodiment, the step of determining the target optimization model based on the target optimization mode and the matching result, and inputting the current rule thresholds of the current decision rule into the target optimization model to obtain the next rule thresholds, includes: if the target optimization mode is a single model optimization mode, then determining a single target optimization model corresponding to the sub-result in the matching result, and inputting the current rule thresholds of the current decision rule into the target optimization model to obtain the next rule thresholds; if the target optimization mode is a combined model optimization mode, then determining multiple target optimization models corresponding to the sub-results in the matching result, and inputting the current rule thresholds of the current decision rule into the target optimization model to obtain the next rule thresholds; Each current rule threshold is input into the target optimization model. Based on preset threshold weights, the initial rule thresholds output by each target optimization model are weighted and summed to obtain the next rule threshold. The optimization direction in the optimization direction control item of the target optimization model is determined based on the deviation of the decision flow operation index, and the optimization magnitude in the optimization magnitude calculation item is determined based on the deviation ratio of the decision flow operation index. The deviation of the decision flow operation index is the difference between the decision flow operation index and the target index value, and the deviation ratio of the decision flow operation index is the ratio of the absolute value of the deviation of the decision flow operation index to the target index value.

[0092] If the target optimization mode is single-model optimization mode, the platform will first identify the unique abnormal sub-result in the matching results (such as the first sub-result with only abnormal decision accuracy, the second sub-result with only rule trigger imbalance, etc.), and then determine the single target optimization model corresponding to the abnormal sub-result. For example, abnormal decision accuracy corresponds to the decision effect optimization model, rule trigger imbalance corresponds to the rule running state optimization model, abrupt change in feature distribution corresponds to the feature data distribution optimization model, and abnormal threshold sensitivity corresponds to the decision effect threshold stability optimization model. Subsequently, the platform will adjust the threshold values ​​of each current rule in the current decision rule (such as the payment order amount). Inputting thresholds such as verification thresholds and transaction IP region compliance thresholds into a single target optimization model will cause the model to directly output the corresponding next rule thresholds based on its built-in optimization logic. The optimization direction determined by the optimization direction control items of all target optimization models is based on the deviation of the decision flow operation index. The deviation of the decision flow operation index is the difference between the actual collected decision flow operation index value and the target index value preset by the platform. The optimization magnitude in the optimization magnitude calculation item is determined based on the deviation ratio of the decision flow operation index. The deviation ratio is the ratio of the absolute value of the decision flow operation index deviation to the target index value.

[0093] The decision-effects optimization model is as follows:

[0094] ;

[0095] In the formula, For symbolic functions: if (Insufficient accuracy) (Relax the threshold to improve accuracy); If (If the accuracy rate is too high, it may be overly strict.) (Tighten the threshold); Indicates the threshold for the next rule. Indicates the current rule threshold. This indicates the accuracy of decisions based on current order decision data. This represents the target indicator value, i.e., the target decision accuracy rate, and the deviation of the decision accuracy rate. . This represents the adjustment coefficient (i.e., the control factor for the magnitude of a single optimization). (Default is 0.5). Step is the minimum tuning step size set according to the threshold type, such as monetary threshold (Step=100) yuan, or fractional threshold (Step=5) fen.

[0096] The runtime state-based tuning model is as follows:

[0097] ;

[0098] In the formula, where, For symbolic functions: if (If the trigger rate is too high and the rules are too lenient), then (Increase the threshold to reduce the trigger rate); if (If the trigger rate is too low or the rules are too strict), then... (Lower the threshold to increase the trigger rate). This indicates the rule trigger rate based on the current order decision data. This represents the deviation between the target metric value, i.e., the target trigger rate, and the rule trigger rate. .

[0099] The feature data distribution optimization model is as follows:

[0100] ;

[0101] In the formula, For symbolic functions: if (If the characteristic mean increases), then (Simultaneously increase the threshold to adapt to distribution changes); if (If the mean of the feature decreases), then (Simultaneously reduce the threshold). This represents the mean of features obtained from a baseline window (such as a historical stable period) based on current order decision data. This represents the target indicator value, i.e., the target characteristic mean, which is also the characteristic mean of the benchmark window (such as the historical stable period), and the deviation of the characteristic mean. .

[0102] If the target optimization mode is a combined model optimization mode, the platform will first identify multiple abnormal sub-results in the matching results, then determine multiple target optimization models corresponding to each sub-result. Simultaneously, the current rule thresholds from the current decision rules will be input into these target optimization models. Each model will output its corresponding initial rule threshold. The platform will then retrieve preset threshold weights (based on the business priority settings of the shopping order scenario on the payment risk control platform) and perform a weighted summation calculation on each initial rule threshold according to these weights to obtain a preliminary next rule threshold. If necessary, business constraint verification will also be performed on this preliminary threshold. The optimization direction determined by the optimization direction control items of all target optimization models is based on the decision flow. For example, if the matching results include a first sub-result, a second sub-result, and a third sub-result, then the decision effect optimization model, the running state optimization model, and the feature data distribution optimization model need to be determined as target optimization models. Then, these target optimization models will be used to calculate the initial rule thresholds. And determine the threshold weights of these initial rule thresholds. It is understandable. Therefore, the formula for obtaining the threshold of the next rule is:

[0103] ;

[0104] Set business constraints:

[0105] ;

[0106] in, The threshold business constraint range is set to avoid optimization exceeding reasonable boundaries.

[0107] Step S15: Write the next rule threshold into the target decision stream based on the hot update mechanism to obtain the next decision rule, and use the next decision rule to make risk control decisions on the next shopping order transaction data.

[0108] In this embodiment, writing the next rule threshold into the target decision stream based on the hot update mechanism includes: smoothing the next rule threshold using a preset smoothing factor to obtain a smoothed next rule threshold; and writing the smoothed next rule threshold into the target decision stream based on the hot update mechanism.

[0109] Smoothing is applied to the next rule threshold calculated using a single model or a combination of models to prevent significant threshold fluctuations from causing turbulence in risk control decisions. Specifically, the platform retrieves a preset smoothing factor. ,and The threshold of the next rule is adjusted using a preset smoothing factor. Smoothing is performed to obtain the next rule threshold after smoothing. The formula is:

[0110] ;

[0111] The preset smoothing factor is a coefficient between 0 and 1, typically set to 0.3, based on the business stability requirements of shopping order risk control scenarios, and can be dynamically adjusted according to business periods. A weighted fusion smoothing process is applied to the initial next rule threshold. By retaining some of the weight of the current rule threshold, the abrupt change in the new threshold is reduced, ensuring that the threshold adjustment is gradual and stable. Simultaneously, the smoothing process sets specific constraints for different types of rule thresholds. After the smoothing process is completed, the platform writes the smoothed next rule threshold into the target decision stream based on a hot update mechanism.

[0112] In this embodiment, the step of writing the next rule threshold into the target decision flow based on the hot update mechanism includes: introducing the next rule threshold into the gray-scale rule set of the decision flow engine, so as to use the next rule threshold to make risk control decisions on a preset number of current shopping order transaction data as gray-scale traffic, so as to obtain the decision effect index of the gray-scale traffic; if the decision effect index of the gray-scale traffic meets the preset verification pass condition, then the next rule threshold is written into the target decision flow based on the hot update mechanism; if the decision effect index of the gray-scale traffic does not meet the preset verification pass condition, then the preset threshold weight is adjusted, and the process jumps back to the step of weighted summation of the initial rule thresholds output by each of the target optimization models based on the preset threshold weight.

[0113] To ensure the security and effectiveness of the new threshold before it takes full effect, the payment risk control platform will first conduct a gray-scale verification process. Specifically, the platform will specifically introduce the next rule threshold (which may be smoothed) into the gray-scale rule set of the decision flow engine. At the same time, it will select a preset number of orders (usually 5% of the total traffic and covering orders from different user levels, product categories, and payment channels) from the real-time access to current shopping order transaction data as gray-scale traffic. Then, it will use the next rule threshold in the gray-scale rule set to conduct independent risk control decisions on this portion of gray-scale traffic. After the decision is completed, the platform will collect the decision effect indicators corresponding to the gray-scale traffic, specifically including the decision accuracy and stability of the gray-scale orders, and calculate the improvement in decision accuracy and stability. The formula for calculating the improvement in decision accuracy is as follows:

[0114] ;

[0115] The formula for calculating the stability improvement is:

[0116] ;

[0117] Subsequently, the improvements in decision accuracy and stability were compared with preset verification conditions. These preset verification conditions are multi-dimensional compliance standards set by the payment risk control platform based on shopping order scenarios, with specific requirements... If all the decision performance indicators of the gray-scale traffic meet these preset verification conditions, the new threshold is deemed to be fully effective. The platform then uses a hot update mechanism to write the next rule threshold into the formal rule set of the target decision flow, achieving seamless full-scale implementation of the new threshold. If the decision performance indicators of the gray-scale traffic do not meet the preset verification conditions, such as in cases where the decision accuracy is not up to standard or the false interception rate exceeds the standard, the current preset threshold weight or preset adjustment coefficient is determined. Unable to balance the optimization needs of multiple scenarios, the platform will adjust the preset threshold weights based on the specific anomalies in the gray-scale decision-making effect. For example, if the decision-making effect indicators do not meet the standards, the weight of the corresponding optimization model for decision-making effect will be increased; if the rule trigger rate is abnormal, the weight of the corresponding optimization model for rule running status will be increased. After the weight adjustment is completed, the platform will jump back to the step of weighted summation of the initial rule thresholds output by each target optimization model based on the preset threshold weights, recalculate the new initial rule thresholds, and then carry out smoothing, gray-scale verification, and other processes in sequence until the decision-making effect indicators of the gray-scale traffic meet the preset verification conditions. This ensures that the next rule threshold finally written into the target decision flow not only meets the business objectives of shopping order risk control, but also has sufficient stability and security, avoiding large-scale risk control decision anomalies caused by direct full-scale activation.

[0118] The hot update mechanism is implemented based on the stateless architecture of the payment risk control platform's distributed configuration center and risk control engine. The platform first structurally encapsulates the next rule threshold according to the rule node ID and threshold type, generating a unique version identifier. Then, it pushes the threshold update command to all risk control engine nodes through the incremental synchronization protocol of the configuration center. Upon receiving the command, each risk control engine node loads the new threshold during the business gap between the completion of risk control decisions for the current batch of shopping orders. During the loading process, the original rule threshold is retained until the new threshold is loaded. The entire update process does not require restarting the risk control engine or interrupting the real-time risk control decision service for shopping orders. The threshold update time for a single node does not exceed a preset time limit. The platform monitors the threshold status of each node in real time. The target decision flow then integrates the new threshold to generate the next decision rule adapted to the shopping order scenario. This rule inherits the three-level link architecture of the original decision flow and optimizes the threshold parameters of each rule node. Subsequently, the payment risk control platform integrates the real-time access to the next shopping order transaction data into the updated target decision flow and uses the newly generated next decision rule to carry out risk control decisions. This enters a new round of "indicator monitoring - optimization judgment - threshold update" closed loop, thereby realizing the dynamic iteration of the payment risk control platform's shopping order risk control decision rules. This ensures that risk control decisions are always adapted to changes in transaction data and business scenarios, balancing decision accuracy and business continuity.

[0119] The beneficial effects of this application are as follows: This application obtains current shopping order transaction data from a payment risk control platform; utilizes the current decision rules in the target decision flow to make risk control decisions on the current shopping order transaction data to obtain current order decision data; wherein, the target decision flow is the shopping order risk control decision process of the payment risk control platform, and the current decision rules are the transaction risk control rules of the payment risk control platform and include each current rule threshold; based on the current order decision data, it obtains decision flow operation indicators corresponding to the current decision rules; wherein, the decision flow operation indicators include rule operation status indicators, feature data distribution indicators, and decision effect indicators; it matches the decision flow operation indicators with each preset threshold optimization trigger condition, and determines the target optimization model based on the matching results, inputs each current rule threshold in the current decision rules into the target optimization model to obtain each next rule threshold; wherein, the target optimization model includes optimization direction control items and optimization magnitude calculation items; based on a hot update mechanism, it writes the next rule threshold into the target decision flow to obtain the next decision rule, and uses the next decision rule to make risk control decisions on the next shopping order transaction data. Therefore, this application obtains current shopping order transaction data from the payment risk control platform and conducts risk control decisions based on the platform's exclusive shopping order risk control decision-making process and transaction risk control rules. Based on the current order decision data, it acquires comprehensive decision flow operation indicators covering rule operation status, feature data distribution, and decision effectiveness. This allows for a complete understanding of the operation process, core data feature changes, and decision quality of the payment risk control platform's shopping order risk control decision flow, providing precise data support for rule threshold optimization tailored to the payment shopping order scenario. By matching the decision flow operation indicators with preset threshold optimization trigger conditions and calling the target optimization model to calculate the next rule threshold, the optimization direction control item ensures that the threshold adjustment direction aligns with the core objectives of the payment risk control platform's shopping order risk control. The optimization magnitude calculation item accurately... By controlling the granularity of threshold adjustments, the subjectivity of manual optimization is avoided, improving the scientific rigor and accuracy of threshold optimization for shopping order risk control rules on the payment risk control platform. This ensures the accuracy of shopping order risk control decisions and reduces the false interception rate of normal orders. Based on a hot update mechanism, the next rule threshold is written into the target decision stream. This allows for rule updates without interrupting the real-time risk control service for shopping orders on the payment risk control platform, ensuring the continuity and stability of payment risk control operations. Simultaneously, the updated next decision rule is used to conduct subsequent shopping order risk control decisions, forming a dynamic closed loop of "decision-monitoring-optimization-update" adapted to the shopping order scenario of the payment risk control platform. This continuously adapts to the dynamic changes in shopping order transaction data and business scenarios, improving the overall adaptability and operational stability of the payment risk control platform's shopping order risk control decisions.

[0120] The following explanation uses the decision-making process of ride-hailing trip safety risk control in travel services as an example: The ride-hailing platform first obtains core data such as basic trip data (route planning, departure / arrival time) and driver / passenger behavior data (driving trajectory, real-time communication records); then, using the preset trip safety risk control decision-making process and current safety risk judgment rules (including thresholds for route deviation and late-night trip identification), it conducts safety risk control decisions for the trip, generating trip safety decision data; subsequently, based on the decision data, it obtains decision-making process operation indicators such as safety incident occurrence rate, false alarm rate, and user complaint rate; it matches these indicators with preset optimization trigger conditions, determines the target optimization model, and inputs the current threshold to obtain the next risk judgment threshold that adapts to the characteristics of the time period and region; finally, it writes the new threshold into the decision-making process through a hot update mechanism to generate new safety risk control rules for subsequent trip safety risk identification, achieving dynamic threshold adaptation for different time periods and regions.

[0121] The following explanation uses the decision-making process of intelligent security access control in a smart park as an example to illustrate this application: The park security system first collects access verification data such as facial recognition information, license plate data, and visitor registration records for personnel / vehicles; then, based on the intelligent security access control decision-making process and the current access judgment rules (including thresholds for facial matching degree and visitor qualification verification), it completes the access risk control decision for personnel / vehicles, forming access decision data; next, it extracts decision-making process operation indicators such as false recognition rate, false recognition rate, and passage efficiency based on the decision data; after matching the indicators with the optimization trigger conditions, it determines the target optimization model, inputs the current threshold to calculate the next access judgment threshold; and writes the new threshold into the decision-making process through hot updating to generate new security access rules for subsequent access verification of personnel and vehicles in the park, taking into account the security and passage needs of different time periods and personnel types.

[0122] The following explanation uses the decision flow execution process of order risk control in an e-commerce scenario as an example: The e-commerce platform first obtains data such as the user's device ID, IP address, account information, and historical transaction and operation records when placing an order; then, it uses the order risk control decision flow and current anti-fraud risk control rules (including thresholds such as the number of orders placed from the same IP, logistics address risk, and review duplication rate) to conduct risk control decisions on the order and generate order risk control decision data; subsequently, it obtains decision flow operation indicators such as the proportion of fraudulent orders and the frequency of orders placed from abnormal IPs based on the decision data; after matching the indicators with preset trigger conditions, it determines the target optimization model, inputs the current threshold to obtain the next risk control threshold; and writes the new threshold into the decision flow through a hot update mechanism to generate new order risk control rules for subsequent order risk control review, especially enabling precise adjustment of the threshold during special periods such as major promotions.

[0123] See Figure 2As shown in the figure, this application discloses a decision flow execution device for a payment risk control platform based on dynamic threshold adjustment, comprising:

[0124] Data acquisition module 11 is used to acquire the current shopping order transaction data of the payment risk control platform;

[0125] The data decision module 12 is used to make risk control decisions on the current shopping order transaction data using the current decision rules in the target decision flow, so as to obtain the current order decision data; wherein, the target decision flow is the shopping order risk control decision process of the payment risk control platform, and the current decision rules are the transaction risk control rules of the payment risk control platform and include the threshold values ​​of each current rule;

[0126] The indicator acquisition module 13 is used to acquire decision flow operation indicators corresponding to the current decision rule based on the current order decision data; wherein, the decision flow operation indicators include rule operation status indicators, feature data distribution indicators, and decision effect indicators;

[0127] The threshold tuning module 14 is used to match the decision flow operation index with each preset threshold tuning trigger condition, and determine the target tuning model based on the matching result. The current rule threshold in the current decision rule is input into the target tuning model to obtain each next rule threshold. The target tuning model includes a tuning direction control term and a tuning magnitude calculation term.

[0128] The decision flow execution module 15 is used to write the next rule threshold into the target decision flow based on the hot update mechanism to obtain the next decision rule, and to use the next decision rule to make risk control decisions on the next shopping order transaction data.

[0129] Furthermore, embodiments of this application also provide an electronic device. Figure 3 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0130] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Specifically, it may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the threshold-based dynamic optimization payment risk control platform decision flow execution method disclosed in any of the foregoing embodiments.

[0131] In this embodiment, the power supply 23 is used to provide operating voltage for various hardware devices on the electronic device; the communication interface 24 can create a data transmission channel between the electronic device and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0132] The processor 21 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 21 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 21 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0133] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored on it include operating system 221, computer program 222 and data 223, etc., and the storage method can be temporary storage or permanent storage.

[0134] The operating system 221 manages and controls the various hardware devices and computer programs 222 on the electronic device to enable the processor 21 to perform calculations and processing on the massive amounts of data 223 in the memory 22. The operating system can be Windows, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of executing the threshold-based dynamic optimization payment risk control platform decision flow execution method disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the electronic device from external devices, as well as data collected by its own input / output interface 25.

[0135] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned payment risk control platform decision flow execution method based on threshold dynamic optimization. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0136] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0137] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly in hardware, software modules executed by a processor, or a combination of both. The software module may be located in random access memory (RAM), memory, read-only memory (ROM), electrically programmable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), register, hard disk, removable disk, CD-ROM (Compact Disc Read-Only Memory), or any other form of storage medium known in the art.

[0138] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0139] The foregoing has provided a detailed description of the decision flow execution method, apparatus, equipment, and medium for a payment risk control platform based on threshold dynamic optimization provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only intended to help understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A decision flow execution method for a payment risk control platform based on threshold dynamic optimization, characterized in that, include: Obtain current shopping order transaction data from the payment risk control platform; The current shopping order transaction data is used to make risk control decisions using the current decision rules in the target decision flow to obtain the current order decision data; wherein, the target decision flow is the shopping order risk control decision process of the payment risk control platform, and the current decision rules are the transaction risk control rules of the payment risk control platform and include the threshold values ​​of each current rule; Based on the current order decision data, decision flow operation indicators corresponding to the current decision rule are obtained; wherein, the decision flow operation indicators include rule operation status indicators, feature data distribution indicators, and decision effect indicators; The decision flow operation indicators are matched with each preset threshold tuning trigger condition, and the target tuning model is determined based on the matching results. The current rule thresholds in the current decision rule are input into the target tuning model to obtain the next rule thresholds. The target tuning model includes tuning direction control terms and tuning magnitude calculation terms. The next rule threshold is written into the target decision stream based on the hot update mechanism to obtain the next decision rule, and the next decision rule is used to make risk control decisions on the next shopping order transaction data.

2. The decision flow execution method for payment risk control platform based on threshold dynamic optimization according to claim 1, characterized in that, The rule operation status indicators include rule trigger rate and average rule execution time; the feature data distribution indicators include feature mean, feature standard deviation, and distribution change rate; the decision performance indicators include decision accuracy, decision precision, and threshold sensitivity; the step of obtaining the decision flow operation indicators corresponding to the current decision rule based on the current order decision data includes: The number of rule triggers, the total number of rule executions, and the execution time of each rule execution are obtained from the current order decision data. The ratio of the number of rule triggers to the total number of rule executions is determined as the rule trigger rate. The average rule execution time is determined based on the execution time of each execution. The feature mean, feature standard deviation, and distribution change rate are determined based on the feature values ​​and sample size in the current order decision data. The ratio of the number of correct decisions to the total number of decisions in the current order decision data is determined as the decision accuracy rate. The ratio of the number of true positive decisions to the number of positive decisions in the current order decision data is determined as the decision precision rate. The ratio of the change in the decision accuracy rate in the first preset time window to the change in the threshold in the current order decision data is determined as the threshold sensitivity.

3. The decision flow execution method for payment risk control platform based on threshold dynamic optimization according to claim 1, characterized in that, The process of matching the decision flow operation indicators with each preset threshold tuning trigger condition, determining the target tuning model based on the matching results, and inputting each current rule threshold in the current decision rule into the target tuning model to obtain each next rule threshold includes: The sub-results obtained by matching the decision flow operation indicators with each preset threshold optimization trigger condition are merged to obtain the matching result; If, based on the matching results, it is determined that the decision flow operation index only satisfies any one of the preset threshold tuning trigger conditions, then the target tuning mode of the current rule threshold is determined to be the single-model tuning mode. If, based on the matching results, it is determined that the decision flow operation index satisfies at least two of the preset threshold tuning trigger conditions, then the target tuning mode of the current rule threshold is determined to be the combined model tuning mode. The target optimization model is determined based on the target optimization mode and the matching result. The threshold values ​​of each current rule in the current decision rule are input into the target optimization model to obtain the threshold values ​​of each next rule.

4. The decision flow execution method for payment risk control platform based on threshold dynamic optimization according to claim 3, characterized in that, The decision flow operation indicators are matched with each preset threshold tuning trigger condition to obtain each sub-result, including: The decision accuracy rate in the decision flow operation index is matched with the first preset threshold optimization trigger condition. If the decision accuracy rate meets the first preset threshold optimization trigger condition, a first sub-result representing the abnormal decision accuracy rate in the decision effect is generated. The first preset threshold optimization trigger condition is that the decision accuracy rate is lower than the preset accuracy rate threshold and the duration is greater than the first preset duration. The rule trigger rate in the decision flow operation index is matched with the second preset threshold optimization trigger condition. If the rule trigger rate meets the second preset threshold optimization trigger condition, a second sub-result representing the rule trigger imbalance is generated. The second preset threshold optimization trigger condition is that the rule trigger rate is higher than the first preset trigger rate threshold or lower than the second preset trigger rate threshold. The distribution change rate in the decision flow operation index is matched with the third preset threshold optimization trigger condition. If the distribution change rate meets the third preset threshold optimization trigger condition, a third sub-result representing the abrupt change in the feature distribution is generated. The third preset threshold optimization trigger condition is that the distribution change rate is higher than the preset change rate threshold and the KS test value of the distribution change rate is less than the preset test value. The threshold sensitivity in the decision flow operation index is matched with the fourth preset threshold tuning trigger condition. If the threshold sensitivity meets the fourth preset threshold tuning trigger condition, a fourth sub-result representing the abnormal threshold sensitivity in the decision effect is generated. The fourth preset threshold tuning trigger condition is that the threshold sensitivity is higher than the preset sensitivity threshold.

5. The decision flow execution method for payment risk control platform based on threshold dynamic optimization according to claim 4, characterized in that, The step of determining the target optimization model based on the target optimization mode and the matching result, and inputting the threshold values ​​of each current rule in the current decision rule into the target optimization model to obtain the threshold values ​​of each next rule, includes: If the target tuning mode is a single model tuning mode, then determine the single target tuning model corresponding to the sub-result in the matching result, and input the current rule thresholds in the current decision rule into the target tuning model to obtain the next rule thresholds; If the target optimization mode is a combined model optimization mode, then multiple target optimization models corresponding to the sub-results in the matching results are determined, and the threshold values ​​of each current rule in the current decision rule are input into the target optimization model. Based on the preset threshold weights, the initial rule threshold values ​​output by each target optimization model are weighted and summed to obtain the next rule threshold value. In the target optimization model, the optimization direction in the optimization direction control item is determined based on the deviation of the decision flow operation index, and the optimization magnitude in the optimization magnitude calculation item is determined based on the deviation ratio of the decision flow operation index. The deviation of the decision flow operation index is the difference between the decision flow operation index and the target index value, and the deviation ratio of the decision flow operation index is the ratio of the absolute value of the deviation of the decision flow operation index to the target index value.

6. The decision flow execution method for payment risk control platform based on dynamic threshold adjustment according to claim 5, characterized in that, The step of writing the next rule threshold into the target decision stream based on the hot update mechanism includes: The next rule threshold is introduced into the gray rule set of the decision flow engine to make risk control decisions on a preset number of current shopping order transaction data as gray traffic, so as to obtain the decision effect index of the gray traffic. If the decision performance index of the grayscale traffic meets the preset verification pass condition, then the next rule threshold is written into the target decision stream based on the hot update mechanism; If the decision performance index of the grayscale traffic does not meet the preset verification pass condition, the preset threshold weight is adjusted, and the process jumps back to the step of weighted summation of the initial rule thresholds output by each of the target optimization models based on the preset threshold weight.

7. The decision flow execution method for payment risk control platform based on dynamic threshold adjustment according to any one of claims 1 to 6, characterized in that, The step of writing the next rule threshold into the target decision stream based on the hot update mechanism includes: The next rule threshold is smoothed using a preset smoothing factor to obtain the smoothed next rule threshold. The next rule threshold after the smoothing process is written into the target decision stream based on the hot update mechanism.

8. A decision flow execution device for a payment risk control platform based on threshold dynamic optimization, characterized in that, include: The data acquisition module is used to acquire the current shopping order transaction data from the payment risk control platform; The data decision module is used to make risk control decisions on the current shopping order transaction data using the current decision rules in the target decision flow, so as to obtain the current order decision data; wherein, the target decision flow is the shopping order risk control decision process of the payment risk control platform, and the current decision rules are the transaction risk control rules of the payment risk control platform and include the threshold values ​​of each current rule; The indicator acquisition module is used to acquire decision flow operation indicators corresponding to the current decision rule based on the current order decision data; wherein, the decision flow operation indicators include rule operation status indicators, feature data distribution indicators, and decision effect indicators; The threshold tuning module is used to match the decision flow operation indicators with each preset threshold tuning trigger condition, and determine the target tuning model based on the matching results. The current rule thresholds in the current decision rule are input into the target tuning model to obtain the next rule thresholds. The target tuning model includes tuning direction control terms and tuning magnitude calculation terms. The decision flow execution module is used to write the next rule threshold into the target decision flow based on a hot update mechanism to obtain the next decision rule, and to use the next decision rule to make risk control decisions on the next shopping order transaction data.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor is configured to execute the computer program to implement the steps of the decision flow execution method for a payment risk control platform based on threshold dynamic tuning as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer programs; wherein, when the computer programs are executed by a processor, they implement the steps of the decision flow execution method for a payment risk control platform based on dynamic threshold tuning as described in any one of claims 1 to 7.

Images