A method for promoting online payment rights based on big data

Abstract

The present invention relates to the technical field of network payment rights promotion based on big data, and discloses a network payment rights promotion method based on big data. First, through the definition of the original data record set, the operation data such as user login, transaction, interaction, etc. are structured to solve the problems of messy data sources and inconsistent formats. Then, data normalization processing converts all records into values ​​between 0 and 1 to avoid the influence of different operation numerical dimension differences on the stability of the algorithm. Deduplication and elimination of abnormal records filter repeated operations at the same time, retaining only those with larger normalized values, and setting a threshold to eliminate abnormal data to reduce redundancy and noise. Finally, the cleaned high-quality data set provides a standardized data source for user stratification, incentive design, A / B testing and other links, ensuring that subsequent analysis is accurate and effective. The overall process optimizes the network payment rights promotion method through data standardization, cleaning and quality improvement.

Description

Technical Field

[0001] The present invention relates to the technical field of network payment rights promotion, and in particular to a network payment rights promotion method based on big data. Background Art

[0002] In the current online payment ecosystem, user benefit promotion has become a crucial strategy for boosting user activity and payment conversion rates. With the rapid development of big data technology, data-analysis-based benefit promotion methods have gradually replaced traditional static benefit allocation models. However, existing promotion methods still suffer from numerous issues, primarily inefficient user data utilization, inaccurate user stratification strategies, delayed incentive program adjustments, and imperfect monitoring and feedback mechanisms, resulting in suboptimal overall promotion results.

[0003] Existing methods for promoting online payment benefits primarily rely on pre-set rules to allocate user benefits. For example, users are categorized into different levels based on static metrics such as registration time and transaction count, and fixed benefit plans are provided to users of different levels. However, this approach fails to fully account for the changing nature of user behavior, making it difficult for promotional plans to align with users' real needs. Regarding data processing, existing methods often employ simple data collection and cleaning strategies, such as directly removing outliers or normalizing data based on fixed thresholds. However, these approaches are subject to high noise interference when dealing with large amounts of user data and fail to fully leverage the correlations between data. Regarding user stratification and lifecycle management, existing technologies typically use fixed-period statistics on user activity, such as categorizing user types based on 30 days of transaction data. However, this approach struggles to dynamically adapt to changes in user behavior. Regarding the design of benefit incentive programs, existing methods primarily set fixed incentive amounts based on user stratification results. For example, new users may receive cash back on their first payment and receive additional points rewards for active users. These static, rule-based incentive strategies struggle to address the diverse needs of individual users and may result in poor incentive effectiveness for some users. In terms of real-time monitoring and A / B testing, the existing method is to randomly divide users into two groups and compare the effects of different equity schemes to adjust the experimental strategy. However, this method makes it difficult to adjust the experimental strategy in real time when user behavior characteristics are changing. In terms of promotion effect evaluation, the existing methods are mainly based on a single profit calculation formula, such as calculating the promotion return rate by the ratio of incentive cost to transaction revenue. However, this calculation method fails to fully consider the long-term value of users. In addition, in terms of model optimization, the existing methods usually adopt fixed parameter adjustment strategies, such as optimizing incentive parameters based on gradient descent. However, this method may have the problem of excessively high computational overhead when faced with large-scale user data, and lacks an adaptive adjustment mechanism during the parameter update process.

[0004] In summary, existing methods for promoting online payment benefits still have numerous shortcomings in data processing, user stratification, benefits incentives, effectiveness evaluation, and parameter optimization. To address these issues, this case aims to propose a more precise, efficient, and dynamically adaptable big data-based benefits promotion method to improve the effectiveness of user benefits promotion, optimize promotion costs, and enhance the user experience of the overall payment ecosystem. Summary of the Invention

[0005] The present invention provides a method for promoting network payment rights and interests based on big data, which promotes the solution of the problems mentioned in the above background technology.

[0006] The present invention provides the following technical solution: a method for promoting online payment benefits based on big data, comprising:

[0007] S1. Data collection and preprocessing:

[0008] S11. Original data record set definition:

[0009] Let user set ;

[0010] in, For users unique number; is the total number of users;

[0011] Set the original record set to: ;

[0012] in, For users In the Records of the first operation; For users The total number of operations;

[0013] Each record For users Detailed data in one operation, specifically in the form of:

[0014] ;

[0015] in, The timestamp when the record is generated; The operation type code is:

[0016] when When , it indicates login operation; when When , it indicates a transaction operation; when When , it indicates interactive operation;

[0017] To record the corresponding original values, specifically:

[0018] when hour, ;when When getting the user In the The transaction amount in this transaction is recorded as , ;when When getting the user The interaction count generated in this interaction is recorded as , ;

[0019] S12. Data normalization:

[0020] Normalize the raw values ​​of all records:

[0021] make, ; ;

[0022] in, is the minimum value of the original value recorded; is the maximum value of the original value recorded;

[0023] For each record Calculate the normalized value: ;in, is the normalized value;

[0024] The updated record is recorded as: ;

[0025] S13, remove duplicate and abnormal records;

[0026] S14. Output cleaned data set:

[0027] Finally, the cleaned data set is generated:

[0028] ;

[0029] S2, user stratification and life cycle stage division;

[0030] S3. Design of customized equity incentive plan;

[0031] S4, real-time monitoring, A / B testing and dynamic adjustment;

[0032] S5. Effect evaluation and continuous optimization.

[0033] Optionally, the deduplication and removal of abnormal records specifically include:

[0034] S131, deduplication processing:

[0035] For the same user Any two records of and If satisfied: and , set the selection function:

[0036] ;

[0037] in, For users No. records; and For users No. Timestamp and operation type in each record;

[0038] S132. Eliminate abnormal records:

[0039] Set the abnormal threshold, denoted as ;

[0040] For each record If satisfied: , then delete the record .

[0041] Optionally, the user stratification and life cycle stage division specifically includes:

[0042] Set the statistical period to 30 days, recorded as ;

[0043] For each user , in the statistical period Calculate the following indicators:

[0044] ;

[0045] ;

[0046] ;

[0047] ;

[0048] ;

[0049] in, is the current moment, in days; For users Number of logins; For users Number of transactions; For users Number of interactions; For users The last login interval of For users Number of days of registration; For users the time of registration;

[0050] Calculate users through active scoring algorithm Activity rating:

[0051] ;

[0052] in, For users Activity score;

[0053] According to the number of days the user has registered and activity score , set up the user Life cycle stage for:

[0054] ;

[0055] in, For new users; For active users; For dormant users; For lost users.

[0056] Optionally, the customized equity incentive plan design specifically includes:

[0057] For each user , according to the life cycle stage Calculate incentive value , and set incentive costs ;make ;

[0058] like , set the new user incentive algorithm, and calculate the incentive value according to the new user incentive algorithm , specifically: ;

[0059] like , set the active user incentive algorithm, and calculate the incentive value based on the active user incentive algorithm , specifically: ;

[0060] like , set the dormant user incentive algorithm, and calculate the incentive value based on the dormant user incentive algorithm , specifically: ;

[0061] like , set the lost user incentive algorithm, and calculate the incentive value based on the lost user incentive algorithm , specifically: ;

[0062] Send the incentive value to the user terminal and record the incentive log.

[0063] Optionally, the real-time monitoring, A / B testing, and dynamic adjustment specifically include:

[0064] For a collection of users in the same lifecycle stage , and randomly divided them into two groups: and grouping ; Group The total number of users in , grouping The total number of users in ;

[0065] Set separately incentive schemes;

[0066] Group use incentive schemes;

[0067] Group use incentive schemes;

[0068] Record users who successfully complete transactions as converted users;

[0069] Set the time window, denoted as ;

[0070] exist Get group The total number of users who converted, recorded as ;

[0071] exist Get group The total number of users who converted, recorded as ;

[0072] Calculation Grouping Conversion rate , specifically: ;

[0073] Calculation Grouping Conversion rate , specifically: ;

[0074] Set the difference indicator algorithm and calculate the difference indicator according to the difference indicator algorithm , specifically:

[0075] ;

[0076] like , then select Group incentive schemes;

[0077] like , then select Group incentive schemes;

[0078] like , then maintain the current incentive plan.

[0079] Optionally, the effect evaluation and continuous optimization specifically include:

[0080] During the statistical period, for all users Record incentive costs and generate income ;

[0081] Which generates income The revenue value generated by user conversion behavior;

[0082] Calculate the overall rate of return, denoted as , specifically:

[0083] ;

[0084] Set the user conversion prediction algorithm, specifically:

[0085] ;

[0086] in, For users In motivation The predicted probability of conversion; is a parameter vector, where represents the excitation sensitivity coefficient, represents the excitation smoothing parameter;

[0087] Calculate the average predicted conversion rate for all users , specifically:

[0088] ;

[0089] Set the evaluation function to: ;

[0090] in, is the conversion rate weight; is the incentive cost weight;

[0091] Update each parameter in the parameter vector;

[0092] Output and ;

[0093] in is the updated parameter vector.

[0094] Optionally, each parameter in the update parameter vector specifically includes:

[0095] S51, yes Find the partial derivative, specifically:

[0096] ;

[0097] S52, yes Find the partial derivative, specifically:

[0098] ;

[0099] Set the learning rate ;

[0100] S53, the update formula is:

[0101] ;

[0102] ;

[0103] in, is the excitation sensitivity parameter at the beginning of the current iteration step The original value of , which is the value determined at the end of the previous iteration, or the initial setting in the first iteration value; Indicates that after the current iteration step is updated, the excitation sensitivity parameter The new value of is the excitation smoothing parameter at the beginning of the current iteration step The original value of value; is the excitation smoothing parameter after updating in the current iteration step The new value of

[0104] Set the convergence threshold, denoted as ;

[0105] like , return to step S51, repeat steps S51 to S53, iteratively update until the convergence condition is met .

[0106] The present invention has the following beneficial effects:

[0107] 1. Through the raw data record set definition step, the system structuredly defines various user operation data, including logins, transactions, and interactions. Each record includes a unique user ID, a timestamp when the record was generated, an operation type code, and a raw value. This resolves the issue of disparate raw data sources and inconsistent formats, laying a standardized foundation for subsequent data processing and analysis. By clearly distinguishing different operation types—logins are fixed at 1.0, transactions are assigned a specific amount, and interactions are assigned a specific count—this ensures accuracy and consistency during the data collection phase, thereby improving data usability. Through the data normalization step, the raw values ​​of all records are unified into the same numerical range. By calculating the minimum and maximum values, each record is converted to a normalized value between 0 and 1. This resolves the issue of algorithm incompatibility caused by significant differences in the raw values ​​of different operation types, such as transaction amounts versus fixed login values ​​or interaction counts. This enables comparability between different data types and reduces the impact of extreme values ​​on algorithm stability, thereby enhancing the accuracy of subsequent data mining and analysis. Through the deduplication and outlier removal steps, the system filters duplicate records of the same user performing the same operation at the same time, retaining records with higher normalized values ​​and eliminating outliers below a preset threshold. This not only effectively removes data redundancy and reduces noise interference with subsequent analysis, but also ensures data integrity and accuracy. Through these steps, the system ensures that the input data is high-quality, free of duplication and anomalies, providing a solid foundation for subsequent user stratification, equity incentives, and effectiveness evaluation. Through the output cleaned data set step, the normalized, deduplicated, and outlier-removed data set is output, providing a standardized, high-quality data source for subsequent user stratification and lifecycle stage classification. Overall, this data collection and preprocessing process addresses issues such as disorganized raw data, high noise levels, and inconsistent formats. It achieves data standardization, cleansing, and quality improvement, ensuring that subsequent processes can perform efficient calculations and make precise decisions based on accurate and unified data, ultimately improving the effectiveness and user experience of the entire online payment equity promotion method.

[0108] 2. Through the deduplication process, the system filters out duplicate records for the same user, addressing the data redundancy issue caused by multiple records of the same type generated by the user at the same time. During this process, the system compares the normalized values ​​of any two records and, using a selection function, retains only the record with the larger value. This prevents duplicate statistics or abnormally high data counts caused by multiple duplicate records. This ensures that only one valid record for the same user is retained at each moment in the dataset, ensuring data uniqueness and accuracy, and providing clean and accurate input for subsequent data analysis and algorithm calculations. Furthermore, during the outlier removal process, the system sets an outlier threshold. A lower threshold indicates more accurate records, while a higher threshold indicates less accurate records. For each record whose normalized value falls below the preset threshold, it is considered an outlier or noise and is removed. This measure effectively addresses the issue of low-quality records in the data and prevents outliers from interfering with overall data statistics and model training. Outlier removal ensures data stability and consistency, enabling subsequent data mining, user behavior analysis, and predictive model calculations to be conducted on a high-quality, reliable dataset.

[0109] 3. By setting a 30-day statistical cycle and calculating multiple key metrics for each user during this period, this step addresses the incomplete user behavior information problem inherent in traditional user stratification, which arises from inconsistent time spans or incomplete data collection. Specifically, by counting data such as the number of logins, transactions, interactions, the interval between last logins, and the number of days since registration within a 30-day period, the system comprehensively and dynamically captures user behavioral characteristics and accurately calculates user lifecycle information by comparing the current time with the registration time. Furthermore, an activity scoring algorithm is used to score users. By adding 1 to the formula to ensure the denominator is non-zero, this effectively reduces the scores of users who have not logged in for a long time, addressing the issue of high or low scores due to long periods of inactivity. Based on the number of days since registration and the activity score, the system further categorizes users into four distinct lifecycle stages: new users, active users, dormant users, and churned users. This step not only eliminates the ambiguous user categories inherent in traditional stratification methods, but also ensures that user groups within each stage have clear behavioral characteristics and value performance. This allows marketing and equity promotion strategies to develop differentiated incentives for users at different lifecycle stages, thereby improving promotion accuracy and conversion rates. Overall, this step, through a comprehensive process encompassing data collection, metric statistics, activity scoring, and stratification, addresses issues such as incomplete user information, strong data heterogeneity, and unclear stratification. Ultimately, it enables the precise capture and scientific management of user behavior, providing a solid data foundation for the subsequent optimization of personalized equity incentives and promotion strategies.

[0110] 4. Through the integrated design of customized equity incentive schemes, the system calculates incentive values ​​for each user based on their lifecycle stage and sets corresponding incentive costs, effectively addressing the single, untargeted incentives inherent in traditional equity promotions. Specifically, the system first stratifies users into distinct categories: new, active, dormant, and churned. It then designs specialized incentive algorithms for each user category. The new user incentive algorithm sets a high initial incentive value, encouraging new users to complete transactions quickly during their first online payment experience, thereby increasing initial conversion rates. This step addresses the issue of new users struggling to complete their first transaction due to a lack of trust in the platform, ensuring the platform can quickly build an active user base. The active user incentive algorithm incorporates an incentive growth mechanism for active users, dynamically adjusting incentive values ​​based on their activity ratings, closely aligning incentives with their recent behavior. This step addresses the issue of under- or over-incentivizing active users, ensuring that incentives maintain user engagement while avoiding resource waste, effectively boosting long-term user stickiness and repeat purchase rates. The dormant user incentive algorithm provides compensatory incentives to dormant users, enabling them to be reactivated. This step addresses the issue of dormant users being unable to return to the platform due to a lack of timely incentives, helping to improve overall user retention. The churned user incentive algorithm designs retention incentive strategies for churned users, offering higher incentive amounts to entice them back to the platform, thereby mitigating the risk of churn due to insufficient incentives. Finally, the system sends the calculated incentive value to the user's terminal and records the incentive log, enabling real-time tracking and effectiveness monitoring of equity distribution. Through this series of steps, the system not only addresses issues such as mismatched user incentives at each stage, uncontrolled incentive costs, and low user conversion rates, but also significantly improves overall user engagement and payment conversion rates on the platform. It also provides a detailed basis for subsequent data analysis and promotion effectiveness evaluation, ultimately enhancing the competitiveness and operational efficiency of the entire online payment ecosystem.

[0111] 5. Through real-time monitoring, A / B testing, and dynamic adjustment, the system first randomly divides users at the same lifecycle stage into two groups, Group A and Group B. This addresses the issue of being unable to directly compare the effectiveness of different incentive schemes when user characteristics are similar. This random grouping ensures a balanced number of users in each group, making subsequent comparative analysis statistically significant and reflecting the differences between the two incentive schemes in practice. Next, the system sets up two different incentive schemes for each group. Within a preset time window (which can be flexibly adjusted based on the actual testing duration), the system counts the number of users in each group who complete transactions and successfully convert, and then calculates the conversion rate for each group. In this way, the system dynamically captures users' real-world responses to different incentive schemes, addressing the issue of traditional incentives being fixed and lacking real-time feedback. After calculating the conversion rate for each group, the system further uses a difference index algorithm to calculate the difference in conversion rate between the two groups. This difference index reflects the difference in effectiveness between the two incentive schemes within the same user population. Based on preset positive and negative thresholds, the system can automatically determine which incentive plan is better: if the difference index is greater than the positive threshold, it means that the incentive effect of Group A's plan is significantly better than that of Group B. In this case, the system selects the incentive plan for Group A; if the difference index is less than the negative threshold, the plan for Group B is selected; if the difference index is in the middle range, the current plan remains unchanged. This dynamic adjustment mechanism solves the problem of the inability to optimize and switch incentive plans in a timely manner during traditional promotions, ensuring that the system can quickly adjust strategies based on real-time data feedback, thereby continuously improving user conversion rates and platform operating efficiency. Overall, through a series of steps such as random grouping, conversion rate statistics, difference index calculation, and automatic selection of the optimal incentive plan based on preset thresholds, the system not only solves the problems of inaccurate evaluation of incentive plan effects, delayed feedback, and fixed strategies, but also realizes dynamic optimization of incentive plans. This method enables the system to continuously monitor the performance of each incentive plan and automatically adjust the promotion strategy based on real-time data, thereby effectively improving the overall user conversion rate and equity promotion effect.

[0112] 6. Through the effectiveness evaluation and continuous optimization steps, the system first records each user's incentive cost and generated revenue during the statistical period. Generated revenue refers to the actual revenue generated by the user's conversion behavior. This method accurately calculates the overall rate of return—the ratio between incentive cost and generated revenue. This solves the difficulty of quantifying the effectiveness of promotion investment in traditional methods and allows for a more intuitive understanding of the overall promotion results. Next, the system implements a user conversion prediction algorithm that calculates the predicted probability of each user converting under the incentive, thereby quantifying their conversion behavior into a specific value. This overcomes the problem of relying solely on historical data to accurately predict future conversion trends and provides a scientific basis for subsequent adjustments to promotion strategies. The system further calculates the average predicted conversion rate for all users and constructs an evaluation function based on this. This function comprehensively considers the weighting of conversion rate and incentive cost, reflecting both the importance of high conversion rates and the negative effects of incentive costs, thereby achieving a balance between the cost-effectiveness and effectiveness of the incentive program. By establishing this evaluation function, the system overcomes the problem of traditional promotion strategies failing to balance user conversion results and cost control, and provides real-time feedback on the actual effectiveness of the promotion program. Finally, the system continuously optimizes the user conversion prediction model and evaluation function by updating each parameter in the parameter vector and utilizing proprietary algorithms such as gradient optimization, thereby achieving the goal of continuously optimizing the promotion strategy. Overall, this step, through data recording, prediction model construction, evaluation function setting, and dynamic parameter updates, effectively addresses issues such as inaccurate promotion effectiveness evaluation, delayed feedback, and rigid strategies. Ultimately, it enables real-time monitoring and continuous improvement of online payment benefit promotion programs, thereby increasing promotion effectiveness and user conversion rates.

[0113] 7. Through steps S51 to S53, the system automatically updates and optimizes each parameter in the parameter vector, addressing the issue of fixed parameter settings and a lack of dynamic adaptive adjustment in traditional methods. First, in step S51, by taking the partial derivative of the evaluation function, the system quantifies the sensitivity of each parameter to the overall objective function, such as conversion prediction and cost control indicators, thereby resolving the mismatch between parameters and system performance. Next, in step S52, the partial derivatives of the parameters are taken again to further refine the contribution of the parameters to different components, ensuring that the comprehensive impact of each variable on the system's performance is fully considered during parameter updates. Next, a learning rate is set. Using the update formula in S53, the previously calculated gradient information is multiplied by the learning rate, and the parameter values ​​are gradually adjusted, achieving dynamic parameter updates. By setting a convergence threshold, the system can determine whether the current parameter update has reached the optimal point, addressing the issues of oscillation or slow convergence that may occur during parameter iteration. If the convergence condition is not met, the system returns to S51 and repeats the entire gradient solution and parameter update process until the convergence condition is met, ultimately outputting the updated parameter vector. This updated parameter vector can more accurately reflect the optimal combination of incentive sensitivity and incentive smoothness, allowing the user conversion prediction model and evaluation function to more accurately predict user conversion probabilities and control incentive costs. Therefore, the partial derivative calculation step in S51 resolves the issue of ambiguous mapping between parameters and evaluation functions; further partial derivative calculations in S52 address the issues of local extreme values ​​and error accumulation during parameter updates; and through the update formula and convergence judgment in S53, the system achieves dynamic parameter optimization, enabling the entire incentive strategy to continuously self-adjust, ultimately achieving the effect of optimizing user conversion rates and reducing incentive costs. Overall, this parameter update mechanism enables the online payment benefit promotion method to possess the capabilities of adaptive learning and continuous optimization, greatly improving the system's flexibility in responding to market changes and user behavior, ensuring the real-time effectiveness of the promotion plan and optimal resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0114] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0115] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0116] Example, see Figure 1 , a method for promoting network payment rights based on big data, comprising:

[0117] S1. Data collection and preprocessing:

[0118] S11. Original data record set definition:

[0119] Let user set ;

[0120] in, For users unique number; is the total number of users;

[0121] Set the original record set to: ;

[0122] in, For users In the Records of the first operation; For users The total number of operations;

[0123] Each record For users Detailed data in one operation, specifically in the form of:

[0124] ;

[0125] in, The timestamp when the record is generated; The operation type code is:

[0126] when When , it indicates login operation; when When , it indicates a transaction operation; when When , it indicates interactive operation;

[0127] To record the corresponding original values, specifically:

[0128] when hour, ;when When getting the user In the The transaction amount in this transaction is recorded as , ;when When getting the user The interaction count generated in this interaction is recorded as , ; For example, when the user 3 clicks means 3 interactions. ;

[0129] S12. Data normalization:

[0130] Normalize the raw values ​​of all records:

[0131] make, ; ;

[0132] in, is the minimum value of the original value recorded; is the maximum value of the original value recorded;

[0133] For each record Calculate the normalized value: ;in, This is a normalized value; it eliminates the dimensionality impact between different operation types (such as transaction amount and number of clicks) to facilitate unified processing.

[0134] The updated record is recorded as: ;

[0135] S13, remove duplicate and abnormal records;

[0136] S14. Output cleaned data set:

[0137] Finally, the cleaned data set is generated:

[0138] ;

[0139] S2, user stratification and life cycle stage division;

[0140] S3. Design of customized equity incentive plan;

[0141] S4, real-time monitoring, A / B testing and dynamic adjustment;

[0142] S5. Effect evaluation and continuous optimization.

[0143] Through the raw data record set definition step, the system structuredly defines various user operation data, including logins, transactions, and interactions. Each record includes a unique user ID, a timestamp when the record was generated, an operation type code, and the original value. This resolves the issue of disparate raw data sources and inconsistent formats, laying a standardized foundation for subsequent data processing and analysis. By clearly distinguishing different operation types, accuracy and consistency are ensured during the data collection phase, thereby improving data usability. Through the data normalization step, the raw values ​​of all records are unified into the same numerical range. Specifically, by calculating the minimum and maximum values, each record value is converted to a normalized value between 0 and 1. This resolves the issue of incompatibility in subsequent algorithms caused by significant differences in the raw values ​​of different operation types, such as transaction amounts versus fixed login values ​​or interaction counts. It enables comparability between different data sets and reduces the impact of extreme values ​​on algorithm stability, thereby enhancing the accuracy of subsequent data mining and analysis. Through the deduplication and outlier removal steps, the system filters duplicate records of the same user performing the same operation at the same time, retaining records with higher normalized values ​​and eliminating outliers below a preset threshold. This not only effectively removes data redundancy and reduces noise interference with subsequent analysis, but also ensures data integrity and accuracy. Through these steps, the system ensures that the input data is high-quality, free of duplication and anomalies, providing a solid foundation for subsequent user stratification, equity incentives, and effectiveness evaluation. Through the output cleaned data set step, the normalized, deduplicated, and outlier-removed data set is output, providing a standardized, high-quality data source for subsequent user stratification and lifecycle stage classification. Overall, this data collection and preprocessing process addresses issues such as disorganized raw data, high noise levels, and inconsistent formats. It achieves data standardization, cleansing, and quality improvement, ensuring that subsequent processes can perform efficient calculations and make precise decisions based on accurate and unified data, ultimately improving the effectiveness and user experience of the entire online payment equity promotion method.

[0144] The deduplication and removal of abnormal records specifically include:

[0145] S131, deduplication processing:

[0146] For the same user Any two records of and If satisfied: and , set the selection function:

[0147] ;

[0148] in, For users No. records; and For users No. Timestamp and operation type in each record;

[0149] Only the record with the larger normalized value at that moment is retained;

[0150] S132. Eliminate abnormal records:

[0151] Set the abnormal threshold, denoted as ; The smaller the anomaly threshold, the more accurate the record, and vice versa;

[0152] For each record If satisfied: , then delete the record .

[0153] Through the deduplication process, the system filters out duplicate records for the same user, addressing the data redundancy issue caused by multiple records of the same type generated by the user at the same time. During this process, the system compares the normalized values ​​of any two records and, using a selection function, retains only the record with the larger value. This prevents duplicate statistics or abnormally high data counts caused by multiple duplicate records. This ensures that only one valid record for the same user is retained at each moment in the dataset, ensuring data uniqueness and accuracy, and providing clean and accurate input for subsequent data analysis and algorithm calculations. Furthermore, during the outlier removal process, the system sets an outlier threshold. A lower threshold indicates more accurate records, while a higher threshold indicates less accurate records. For each record whose normalized value falls below the preset threshold, it is considered an outlier or noise and is removed. This measure effectively addresses the issue of low-quality records in the data and prevents outliers from interfering with overall data statistics and model training. Outlier removal ensures data stability and consistency, enabling subsequent data mining, user behavior analysis, and predictive model calculations to be conducted on a high-quality, reliable dataset.

[0154] The user stratification and life cycle stage division specifically include:

[0155] Set the statistical period to 30 days, recorded as ;

[0156] For each user , in the statistical period Calculate the following indicators:

[0157] ;

[0158] ;

[0159] ;

[0160] ;

[0161] ;

[0162] in, is the current moment, in days; For users Number of logins; For users Number of transactions; For users Number of interactions; For users The last login interval of For users Number of days of registration; For users the time of registration;

[0163] Calculate users through active scoring algorithm Activity rating:

[0164] ;

[0165] Here, 0.33 represents the contribution of logins to the activity score. While login is a basic user action, its guiding role in subsequent transactions and interactions is relatively indirect, so it is given a medium weight. 0.44 represents the direct contribution of transactions to user activity. Transactions are directly related to revenue and user stickiness, so they are given the highest weight in the score. 0.23 represents the impact of interactions on the score. While interactions can reflect user engagement, their influence is relatively indirect, so the value is lower.

[0166] in, For users Activity score; Used to reduce the score of users who have not logged in for a long time. +1 ensures that the denominator is not zero;

[0167] According to the number of days the user has registered and activity score , set up the user Life cycle stage for:

[0168] ;

[0169] in, For new users; For active users; For dormant users; For lost users.

[0170] By setting a 30-day statistical cycle and calculating multiple key metrics for each user during this period, this step addresses the incomplete user behavior information often encountered in traditional user stratification due to inconsistent time spans or incomplete data collection. Specifically, by analyzing data such as logins, transactions, interactions, the time between last logins, and days since registration within a 30-day period, the system comprehensively and dynamically captures user behavioral characteristics. By comparing the current time with the time of registration, the system accurately calculates user lifecycle information. Furthermore, an activity scoring algorithm is used to assign scores to users. By adding 1 to the formula to ensure the denominator is non-zero, this effectively reduces the scores of users who have been inactive for a long time, addressing the issue of overly high or low scores due to long periods of inactivity. Based on the number of days since registration and the activity score, the system further categorizes users into four distinct lifecycle stages: new users, active users, dormant users, and churned users. This step not only eliminates the ambiguity inherent in traditional stratification methods but also provides clear behavioral characteristics and value propositions for each user group. This allows marketing and promotion strategies to tailor incentives to users at different lifecycle stages, improving promotional precision and conversion rates. Overall, this step solves problems such as incomplete user information, strong data heterogeneity, and unclear stratification through a complete set of processes including data collection, indicator statistics, active scoring, and stratification. It ultimately achieves accurate capture and scientific management of user behavior, providing a solid data foundation for subsequent optimization of personalized rights incentives and promotion strategies.

[0171] The customized equity incentive plan design specifically includes:

[0172] For each user , according to the life cycle stage Calculate incentive value , and set incentive costs ;make ;

[0173] like , set the new user incentive algorithm, and calculate the incentive value according to the new user incentive algorithm , specifically: ;

[0174] here The initial incentive benchmark for new users is 10 points, which means that newly registered users receive higher rewards to lower the threshold for first-time users. 1.17 indicates that the daily incentive decay rate is 1.17 points.

[0175] like , set the active user incentive algorithm, and calculate the incentive value based on the active user incentive algorithm , specifically: ;

[0176] Here, 5 indicates that the base incentive for active users is 5 points, reflecting that users at this stage already have a certain behavioral foundation and do not need a too high initial reward. 0.81 indicates that the incentive increases by 0.81 points for every 5 points above the baseline activity score. 15 indicates that the incentive cap is 15 points to prevent excessively high incentives from causing uncontrolled costs.

[0177] like , set the dormant user incentive algorithm, and calculate the incentive value based on the dormant user incentive algorithm , specifically: ;

[0178] Here, 8 indicates that the base incentive for dormant users is 8 points, a slightly higher benchmark than for active users to encourage reactivation. 1.52 indicates that for every 5 points below the baseline, the compensatory incentive increases by 1.52 points. The higher coefficient reflects that dormant users are more difficult to convert and require stronger incentives. 20 indicates that the incentive cap is 20 points.

[0179] like , set the lost user incentive algorithm, and calculate the incentive value based on the lost user incentive algorithm , specifically: ;

[0180] Here, 12 indicates that the base incentive for churned users is 12 points, giving a higher initial reward to retain long-term inactive users. 2.03 indicates an incentive-retention coefficient of 2.03, reflecting that the incentive effect for churned users requires greater compensation to achieve a stronger retention effect. 2 is the active score threshold for determining churned users, and 25 indicates that the incentive cap for churned users is 25 points.

[0181] Send the incentive value to the user terminal and record the incentive log.

[0182] Through the integrated design of customized equity incentive plans, the system calculates incentive values ​​for each user based on their lifecycle stage and sets corresponding incentive costs, effectively addressing the single, untargeted incentives inherent in traditional equity promotions. Specifically, the system first stratifies users into distinct categories: new, active, dormant, and churned. It then designs specialized incentive algorithms for each user category. The new user incentive algorithm sets a high initial incentive value, encouraging new users to complete transactions quickly during their first online payment experience, thereby increasing initial conversion rates. This step addresses the issue of new users struggling to complete their first transaction due to a lack of trust in the platform, ensuring the platform can quickly build an active user base. The active user incentive algorithm incorporates an incentive growth mechanism for active users, dynamically adjusting incentive values ​​based on their activity ratings and closely aligning incentives with their recent behavior. This step addresses the issue of under- or over-incentivization for active users, ensuring that incentives maintain user engagement while avoiding resource waste, effectively improving long-term user stickiness and repeat purchase rates. The dormant user incentive algorithm provides compensatory incentives to inactive users, reactivating them. This step addresses the issue of dormant users unable to return to the platform due to a lack of timely incentives, helping to improve overall user retention. The churned user incentive algorithm designs retention incentive strategies for churned users, offering higher incentive amounts to entice them to return to the platform's services, thereby addressing the risk of churn due to insufficient incentives. Finally, the system sends the calculated incentive value to the user's terminal and records the incentive log, enabling real-time tracking and effectiveness monitoring of equity issuance. Through this series of steps, the system not only addresses issues such as mismatched user incentives at each stage, uncontrolled incentive costs, and low user conversion rates, but also significantly improves the platform's overall user engagement and payment conversion rate. It also provides a detailed basis for subsequent data analysis and promotion effectiveness evaluation, ultimately enhancing the competitiveness and operational efficiency of the entire online payment ecosystem.

[0183] The real-time monitoring, A / B testing, and dynamic adjustment specifically include:

[0184] For a collection of users in the same lifecycle stage , to satisfy the same All users of and grouping ; Group The total number of users in , grouping The total number of users in ; and meet ,in For collection The total number of users in ;

[0185] Set separately incentive schemes; The incentive schemes can be set by actual users. Here, only the number of converted users of the two incentive schemes is used;

[0186] Group use incentive schemes;

[0187] Group use incentive schemes;

[0188] Record users who successfully complete transactions as converted users;

[0189] Set the time window, denoted as ; Time window The duration of A / B testing can be set according to the user's specific requirements;

[0190] exist Get group The total number of users who converted, recorded as ;

[0191] exist Get group The total number of users who converted, recorded as ;

[0192] Calculation Grouping Conversion rate , specifically: ;

[0193] Calculation Grouping Conversion rate , specifically: ;

[0194] Set the difference indicator algorithm and calculate the difference indicator according to the difference indicator algorithm , specifically:

[0195] ;

[0196] like , then select Group incentive schemes;

[0197] like , then select Group incentive schemes;

[0198] like , then maintain the current incentive scheme; if the current group incentive scheme, continue to use group Incentive scheme; if the current group incentive scheme, continue to use group incentive scheme.

[0199] Through real-time monitoring, A / B testing, and dynamic adjustments, the system first randomly divides users at the same lifecycle stage into two groups, Group A and Group B. This addresses the difficulty of directly comparing the effectiveness of different incentive schemes when user characteristics are similar. This random grouping ensures a balanced population of users within each group, enabling statistically significant subsequent comparative analysis and reflecting the differences between the two incentive schemes in real-world applications. Next, the system sets up two different incentive schemes for each group. Within a preset time window (which can be flexibly adjusted based on the actual testing duration), the system counts the number of users in each group who complete transactions and successfully convert, and then calculates the conversion rate for each group. This approach dynamically captures users' real-world responses to different incentive schemes, addressing the issue of traditional incentives being fixed and lacking real-time feedback. After calculating the conversion rate for each group, the system further uses a difference index algorithm to calculate the difference in conversion rate between the two groups. This difference index reflects the difference in effectiveness between the two incentive schemes within the same user population. Based on preset positive and negative thresholds, the system can automatically determine which incentive plan is better: if the difference index is greater than the positive threshold, it means that the incentive effect of Group A's plan is significantly better than that of Group B. In this case, the system selects the incentive plan for Group A; if the difference index is less than the negative threshold, the plan for Group B is selected; if the difference index is in the middle range, the current plan remains unchanged. This dynamic adjustment mechanism solves the problem of the inability to optimize and switch incentive plans in a timely manner during traditional promotions, ensuring that the system can quickly adjust strategies based on real-time data feedback, thereby continuously improving user conversion rates and platform operating efficiency. Overall, through a series of steps such as random grouping, conversion rate statistics, difference index calculation, and automatic selection of the optimal incentive plan based on preset thresholds, the system not only solves the problems of inaccurate evaluation of incentive plan effects, delayed feedback, and fixed strategies, but also realizes dynamic optimization of incentive plans. This method enables the system to continuously monitor the performance of each incentive plan and automatically adjust the promotion strategy based on real-time data, thereby effectively improving the overall user conversion rate and equity promotion effect.

[0200] The effect evaluation and continuous optimization specifically include:

[0201] During the statistical period, for all users Record incentive costs and generate income ;

[0202] Which generates income The revenue value generated by user conversion behavior;

[0203] Calculate the overall rate of return, denoted as , specifically:

[0204] ;

[0205] Set the user conversion prediction algorithm, specifically:

[0206] ;

[0207] in, For users In motivation The predicted probability of conversion; is a parameter vector, where represents the excitation sensitivity coefficient, represents the excitation smoothing parameter;

[0208] Calculate the average predicted conversion rate for all users , specifically:

[0209] ;

[0210] Set the evaluation function to: ;

[0211] in, is the conversion rate weight, reflecting the importance of conversion rate; is the incentive cost weight, reflecting the negative effect of incentive cost; and It can be obtained through historical data fitting, cost-profit analysis and management strategy formulation;

[0212] Update each parameter in the parameter vector.

[0213] For example, there are two users and , the user incentive cost is and , user-generated revenue Yuanhe Element; calculate the total ; Assume the initial parameters are ; Calculate the predicted conversion rate for each user, , ;right , ; Calculate the average predicted conversion rate, ; Set weight , , .

[0214] Through performance evaluation and continuous optimization, the system first records each user's incentive cost and generated revenue during the statistical period. Generated revenue refers to the actual revenue generated by a user's conversion behavior. This method accurately calculates the overall rate of return—the ratio between incentive cost and generated revenue. This overcomes the difficulty of quantifying the effectiveness of promotional investment in traditional methods and allows for a more intuitive understanding of overall promotional results. Next, the system implements a user conversion prediction algorithm that calculates the predicted probability of each user converting under an incentive, thereby quantifying their conversion behavior into a specific value. This overcomes the problem of relying solely on historical data to accurately predict future conversion trends and provides a scientific basis for subsequent adjustments to promotional strategies. The system further calculates the average predicted conversion rate for all users and constructs an evaluation function based on this. This function comprehensively considers the weighting of conversion rate and incentive cost, reflecting both the importance of high conversion rates and the negative effects of incentive costs, thereby achieving a balance between the cost-effectiveness and effectiveness of incentive programs. By establishing this evaluation function, the system addresses the problem of traditional promotional strategies failing to balance user conversion results and cost control, and provides real-time feedback on the actual effectiveness of promotional programs. Finally, the system continuously optimizes the user conversion prediction model and evaluation function by updating each parameter in the parameter vector and utilizing proprietary algorithms such as gradient optimization, thereby achieving the goal of continuously optimizing the promotion strategy. Overall, this step, through data recording, prediction model construction, evaluation function setting, and dynamic parameter updates, effectively addresses issues such as inaccurate promotion effectiveness evaluation, delayed feedback, and rigid strategies. Ultimately, it enables real-time monitoring and continuous improvement of online payment benefit promotion programs, thereby increasing promotion effectiveness and user conversion rates.

[0215] Each parameter in the update parameter vector specifically includes:

[0216] S51, yes Find the partial derivative, specifically:

[0217] ;

[0218] S52, yes Find the partial derivative, specifically:

[0219] ;

[0220] Set the learning rate ;

[0221] S53, the update formula is:

[0222] ;

[0223] ;

[0224] in, is the excitation sensitivity parameter at the beginning of the current iteration step The original value of , which is the value determined at the end of the previous iteration, or the initial setting in the first iteration value; Indicates that after the current iteration step is updated, the excitation sensitivity parameter The new value of , which reflects the adjustment along the gradient direction in the current iteration The result after is the excitation smoothing parameter at the beginning of the current iteration step The original value of value; is the excitation smoothing parameter after updating in the current iteration step The new value of , which means that in this round of iteration, the gradient is adjusted value;

[0225] Set the convergence threshold, denoted as ;

[0226] like , return to step S51, repeat steps S51 to S53, iteratively update until the convergence condition is met ;

[0227] Output and ;

[0228] in is the updated parameter vector.

[0229] For example, to calculate the gradient , ;right , ; Set the learning rate , then update, ; ; If the convergence condition is met, the iteration is stopped and the optimal parameter vector is finally output .

[0230] Through steps S51 to S53, the system automatically updates and optimizes each parameter in the parameter vector, addressing the fixed parameter settings and lack of dynamic adaptive adjustment found in traditional methods. First, in step S51, by taking the partial derivative of the evaluation function, the system quantifies the sensitivity of each parameter to the overall objective function, such as conversion prediction and cost control indicators, thereby resolving the mismatch between parameters and system performance. Next, in step S52, the partial derivatives of the parameters are taken again to further refine the contribution of the parameters to different components, ensuring that the comprehensive impact of each variable on the system's performance is fully considered during parameter updates. Next, a learning rate is set. Using the update formula in S53, the previously calculated gradient information is multiplied by the learning rate, and the parameter values ​​are gradually adjusted, achieving dynamic parameter updates. By setting a convergence threshold, the system can determine whether the current parameter update has reached the optimal point, addressing potential oscillations or slow convergence during parameter iteration. If convergence conditions are not met, the system returns to S51 and repeats the entire gradient solution and parameter update process until convergence conditions are met, ultimately outputting the updated parameter vector. This updated parameter vector can more accurately reflect the optimal combination of incentive sensitivity and incentive smoothness, allowing the user conversion prediction model and evaluation function to more accurately predict user conversion probabilities and control incentive costs. Therefore, the partial derivative calculation step in S51 resolves the issue of ambiguous mapping between parameters and evaluation functions; further partial derivative calculations in S52 address the issues of local extreme values ​​and error accumulation during parameter updates; and through the update formula and convergence judgment in S53, the system achieves dynamic parameter optimization, enabling the entire incentive strategy to continuously self-adjust, ultimately achieving the effect of optimizing user conversion rates and reducing incentive costs. Overall, this parameter update mechanism enables the online payment benefit promotion method to possess the capabilities of adaptive learning and continuous optimization, greatly improving the system's flexibility in responding to market changes and user behavior, ensuring the real-time effectiveness of the promotion plan and optimal resource utilization.

[0231] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0232] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for promoting network payment rights based on big data, characterized in that: include: S1. Data collection and preprocessing: S11. Original data record set definition: Let user set ; in, For users unique number; is the total number of users; Set the original record set to: ; in, For users In the Records of the first operation; For users The total number of operations; Each record For users Detailed data in one operation, specifically in the form of: ； in, The timestamp when the record is generated; The operation type code is: when When , it indicates login operation; when When , it indicates a transaction operation; when When , it indicates interactive operation; To record the corresponding original values, specifically: when hour, ;when When getting the user In the The transaction amount in this transaction is recorded as , ;when When getting the user The interaction count generated in this interaction is recorded as , ; S12. Data normalization: Normalize the raw values ​​of all records: make, ; ; in, is the minimum value of the original value recorded; is the maximum value of the original value recorded; For each record Calculate the normalized value: ;in, is the normalized value; The updated record is recorded as: ; S13, remove duplicate and abnormal records; S14. Output cleaned data set: Finally, the cleaned data set is generated: ； S2, user stratification and life cycle stage division; The user stratification and life cycle stage division specifically include: Set the statistical period to 30 days, recorded as ; For each user , in the statistical period Calculate the following indicators: ； ； ； ； ； in, is the current moment, in days; For users Number of logins; For users Number of transactions; For users Number of interactions; For users The last login interval of For users Number of days of registration; For users the time of registration; Calculate users through active scoring algorithm Activity rating: ； in, For users Activity score; According to the number of days the user has registered and activity score , set up the user Life cycle stage for: ； in, For new users; For active users; For dormant users; For lost users; S3. Design of customized equity incentive plan; S4, real-time monitoring, A / B testing and dynamic adjustment; S5. Effect evaluation and continuous optimization.

2. A method for promoting network payment benefits based on big data according to claim 1, characterized in that: The deduplication and removal of abnormal records specifically include: S131, deduplication processing: For the same user Any two records of and If satisfied: and , set the selection function: ； in, For users No. records; and For users No. Timestamp and operation type in each record; S132. Eliminate abnormal records: Set the abnormal threshold, denoted as ; For each record If satisfied: , then delete the record .

3. The method for promoting network payment rights based on big data according to claim 1, characterized in that: The customized equity incentive plan design specifically includes: For each user , according to the life cycle stage Calculate incentive value , and set incentive costs ;make ; like , set the new user incentive algorithm, and calculate the incentive value according to the new user incentive algorithm , specifically: ; like , set the active user incentive algorithm, and calculate the incentive value based on the active user incentive algorithm , specifically: ; like , set the dormant user incentive algorithm, and calculate the incentive value based on the dormant user incentive algorithm , specifically: ; like , set the lost user incentive algorithm, and calculate the incentive value based on the lost user incentive algorithm , specifically: ; Send the incentive value to the user terminal and record the incentive log.

4. The method for promoting network payment rights based on big data according to claim 3, characterized in that: The real-time monitoring, A / B testing, and dynamic adjustment specifically include: For a collection of users in the same lifecycle stage , and randomly divided them into two groups: and grouping ; Group The total number of users in , grouping The total number of users in ; Set separately incentive schemes; Group use incentive schemes; Group use incentive schemes; Record users who successfully complete transactions as converted users; Set the time window, denoted as ; exist Get group The total number of users who converted, recorded as ; exist Get group The total number of users who converted, recorded as ; Calculation Grouping Conversion rate , specifically: ; Calculation Grouping Conversion rate , specifically: ; Set the difference indicator algorithm and calculate the difference indicator according to the difference indicator algorithm , specifically: ； like , then select Group incentive schemes; like , then select Group incentive schemes; like , then maintain the current incentive plan.

5. The method for promoting network payment rights based on big data according to claim 4, characterized in that: The effect evaluation and continuous optimization specifically include: During the statistical period, for all users Record incentive costs and generate income ; which generates income The revenue value generated by user conversion behavior; Calculate the overall rate of return, denoted as , specifically: ； Set the user conversion prediction algorithm, specifically: ； in, For users In motivation The predicted probability of conversion; is a parameter vector, where represents the excitation sensitivity coefficient, represents the excitation smoothing parameter; Calculate the average predicted conversion rate for all users , specifically: ； Set the evaluation function to: ; in, is the conversion rate weight; is the incentive cost weight; Update each parameter in the parameter vector; Output and ; in is the updated parameter vector.

6. The method for promoting network payment rights based on big data according to claim 5, characterized in that: Each parameter in the update parameter vector specifically includes: S51, yes Find the partial derivative, specifically: ； S52, yes Find the partial derivative, specifically: ； Set the learning rate ; S53, the update formula is: ； ； in, is the excitation sensitivity parameter at the beginning of the current iteration step The original value of , which is the value determined at the end of the previous iteration, or the initial setting in the first iteration value; Indicates that after the current iteration step is updated, the excitation sensitivity parameter The new value of is the excitation smoothing parameter at the beginning of the current iteration step The original value of value; is the excitation smoothing parameter after updating in the current iteration step The new value of Set the convergence threshold, denoted as ; like , return to step S51, repeat steps S51 to S53, iteratively update until the convergence condition is met .

Images