A cross-platform transaction flow analysis and precision marketing automation management system

Abstract

The application discloses a cross-platform transaction flow analysis and precision marketing automation management system, relates to the field of intelligent marketing, and collects original transaction flow data of multiple heterogeneous transaction platforms and washes and standardizes the original transaction flow data, extracts user continuous and discrete transaction behavior characteristics to construct a multi-dimensional portrait vector, determines initial weight parameters based on marketing target type identification, executes mixed attribute clustering to generate user clustering results, receives user response data after marketing execution, identifies attribute dominant features based on response rate differences and dynamically corrects weight parameters to iteratively optimize clustering results, and returns response data through closed-loop feedback to drive weight continuous evolution, so that adaptive clustering driven by marketing targets and closed-loop optimization based on actual marketing effects are realized, and the matching precision of user clustering and marketing targets and the marketing conversion rate are significantly improved.

Description

Technical Field

[0001] This application relates to the field of intelligent marketing technology, and in particular to a cross-platform transaction flow analysis and precision marketing automated management system. Background Technology

[0002] Currently, with the rapid development of internet finance and e-commerce, user transactions are widely distributed across multiple heterogeneous transaction platforms, such as third-party payment platforms, online banking channels, and e-commerce platforms. Existing marketing management systems are typically built for a single platform, collecting transaction data within that platform to extract user characteristics such as spending amount and frequency, constructing user profiles, and executing marketing strategies. Some systems have introduced clustering algorithms (such as K-Means and K-Prototypes) to segment users and generate differentiated marketing plans based on the segmentation results. However, these systems generally adopt static or fixed-weight clustering strategies, meaning that regardless of changes in marketing objectives, the same attribute weights are used to segment users.

[0003] The aforementioned existing technologies have the following shortcomings: First, transaction data from a single platform cannot fully reflect users' cross-platform spending power and behavioral preferences, resulting in insufficient completeness and accuracy of user profiles. Second, existing clustering methods use fixed weights for distance calculations on continuous attributes (such as spending amount) and discrete attributes (such as product category preference), failing to adaptively adjust the clustering focus according to different marketing objectives (such as category penetration, average order value increase, user retention, etc.), leading to low matching degree between clustering results and marketing goals. Finally, there is a lack of a mechanism to provide closed-loop feedback of user response data after marketing execution to optimize clustering parameters, making it impossible to dynamically improve the clustering strategy based on actual marketing results. This prevents the clustering model from continuously evolving with business practices, making it difficult to improve marketing accuracy. Summary of the Invention

[0004] The purpose of this application is to provide a cross-platform transaction flow analysis and precision marketing automated management system to solve the aforementioned problems existing in the prior art.

[0005] To achieve the above objectives, this application provides a cross-platform transaction flow analysis and precision marketing automated management system, comprising: The cross-platform data acquisition module is used to collect raw transaction flow data from multiple heterogeneous trading platforms; The data cleaning and standardization module, connected to the cross-platform data acquisition module, is used to perform format conversion and cleaning processing on the raw transaction log data to generate a standardized transaction dataset in a unified format. The multi-dimensional user profile building module, connected to the data cleaning and standardization module, is used to extract user transaction behavior features based on standardized transaction datasets. These user transaction behavior features include continuous attribute features and discrete attribute features. The marketing feedback adaptive clustering module, connected to the multi-dimensional user profile building module, is used to receive the target type identifier of the current marketing campaign, determine the initial weight parameters for mixed attribute clustering based on the preset target-weight mapping relationship, perform mixed attribute clustering on user transaction behavior characteristics based on the initial weight parameters, generate user grouping results, and receive user response data after execution, identify the dominant attribute features based on the response rate differences of each user group, and dynamically adjust the weight parameters according to the dominant attribute features to generate updated user grouping results. The marketing strategy generation module is connected to the marketing feedback adaptive clustering module to generate personalized marketing plans based on user segmentation results. The automated execution module connects to the marketing strategy generation module to execute personalized marketing plans; The closed-loop feedback module, connected to the automated execution module and the marketing feedback adaptive clustering module, is used to send user response data back to the marketing feedback adaptive clustering module to drive the iterative optimization of weight parameters.

[0006] Preferably, the cross-platform data acquisition module includes: The data permission verification submodule is used to verify the validity of data access authorization tokens of each platform before data collection, and to automatically trigger a refresh mechanism when the token expires. The incremental synchronization submodule is used to collect transaction records that have been added or changed since the last synchronization time, based on the timestamp field of the transaction flow of each platform.

[0007] Preferably, the data cleaning and standardization module includes: The format conversion unit is used to map data fields from different heterogeneous trading platforms to a unified data dictionary, and to perform time zone alignment for transaction times from each platform; the data dictionary includes transaction time, transaction amount, counterparty, transaction type, and commodity category; The anomaly detection unit is used to identify abnormal transaction data, including missing timestamps, zero or negative amounts, and cross-platform time conflict records. For the identified abnormal data, corresponding processing is performed according to the anomaly type: for records with missing timestamps, the timestamps of adjacent valid records are interpolated to fill the gaps; for records with zero or negative amounts, the transactions are marked as invalid and removed; and for records with cross-platform time conflicts, the timestamps are corrected based on time zone alignment rules. The cross-platform duplicate identification unit is used to identify duplicate records of the same transaction on different platforms based on joint matching of transaction amount, transaction time window and counterparty identifier; for the identified duplicate records, one complete record is retained, and the remaining records are marked as cross-platform redundant data and removed. The missing value processing unit is used to clean records that still have missing fields after the above processing by using the corresponding strategies from mean filling, nearest neighbor filling, or whole record removal.

[0008] Preferably, the multi-dimensional user profile construction module includes: The continuous attribute extraction unit is used to calculate users' continuous transaction behavior indicators based on standardized transaction datasets. The continuous transaction behavior indicators include: the duration of the most recent consumption interval, the frequency of consumption within the statistical period, the total consumption amount within the statistical period, the average amount of a single transaction, the standard deviation of transaction amount, and the ratio of fund inflow to outflow. The discrete attribute extraction unit is used to identify users' discrete transaction behavior preferences based on a standardized transaction dataset. Discrete transaction behavior preferences include: high-frequency transaction category ranking, preferred transaction channels, distribution of active transaction periods, and concentration of counterparty types. The profile vector generation unit is used to vectorize and concatenate continuous transaction behavior indicators with discrete transaction behavior preferences to generate multi-dimensional user profile vectors.

[0009] Preferably, the marketing feedback adaptive clustering module includes: The initial clustering unit is used to receive the target type identifier of the current marketing campaign, determine the initial weight parameters of the mixed attribute clustering based on the preset target-weight mapping relationship, and perform the first mixed attribute clustering on the user transaction behavior characteristics based on the initial weight parameters to generate the initial user grouping results; The response rate calculation unit is used to calculate the marketing response rate of each user group based on the collected user response data after the marketing campaign is executed, and to identify user groups with response rates that are significantly higher or lower than the average level. The attribute-dominant discrimination unit is used to determine the attribute-dominant type of user groups with abnormal response rates by comparing the variance of continuous attributes with the purity of discrete attributes. The weight update unit, connected to the attribute dominance discrimination unit and the closed-loop feedback module, is used to determine the weight adjustment direction based on the attribute dominance type and update the weight parameters based on the preset learning rate. The clustering iteration unit, connected to the initial clustering unit and the weight update unit, is used to: output the initial user clustering results during the first run; and in subsequent iterations, re-execute the mixed attribute clustering based on the updated weight parameters output by the weight update unit to generate updated user clustering results.

[0010] Preferably, the hybrid attribute clustering is based on the K-Prototypes algorithm, and introduces dynamic weight parameters to weight and fuse the distances of continuous attributes and discrete attributes. The hybrid distance metric formula is as follows: ; in, Euclidean distance for continuous properties Hamming distance for discrete attributes For the first t The dynamic weight parameters for each round of iterations are initially determined by the target type identifier and are iteratively adjusted based on user response data after the marketing campaign is executed.

[0011] Preferably, the formula for calculating the purity of the discrete attribute is: ; in, For user groups The number of samples, d The total number of feature dimensions. p The number of dimensions for continuous attributes. This is a matching indicator function for discrete attributes. It takes a value of 1 when the discrete attribute value matches, and a value of 0 otherwise.

[0012] Preferably, the weight update unit is configured as follows: When the proportion of discrete attributes dominating in a high-response-rate user group exceeds a preset threshold, increase the weighting parameter. To enhance the influence of discrete attributes in clustering; When the proportion of continuous attributes dominating in a high-response-rate user group exceeds a preset threshold, reduce the weighting parameter. This is to enhance the influence of continuous attributes in clustering.

[0013] Preferably, the target-weight mapping relationship includes: The category penetration target corresponds to the first initial weight value, which is used to strengthen the weight of discrete attributes to gather users with similar category preferences. The value of the first initial weight value ranges from 1.2 to 1.5. The target for increasing average order value corresponds to the initial value of the second weight, which is used to strengthen the weight of continuous attributes to gather users with similar spending power. The value of the initial value of the second weight ranges from 0.3 to 0.5. The initial value of the third weight for user retention targets is used to balance the influence of continuous and discrete attributes. The value of the initial value of the third weight ranges from 0.8 to 1.2. The initial value of the first weight is greater than the initial value of the third weight, and the initial value of the third weight is greater than the initial value of the second weight.

[0014] Preferably, the marketing feedback adaptive clustering module further includes a convergence judgment unit, which is used to: stop iterative optimization and output the current user grouping result as the final grouping result when the change of the weight parameter in multiple consecutive iterations is lower than a preset convergence threshold.

[0015] Therefore, this application adopts the aforementioned cross-platform transaction flow analysis and precision marketing automated management system. Through the collaboration of the cross-platform data acquisition module and the data cleaning and standardization module, it achieves unified access and standardized processing of transaction flow data from multiple heterogeneous transaction platforms, solving the problem of user profile distortion caused by data silos on a single platform and improving the completeness and accuracy of user profiles. By introducing target type identifier-driven initial weight parameters and a dynamic correction mechanism through the marketing feedback adaptive clustering module, the clustering weights can be adaptively adjusted according to the marketing target type, solving the problem of fixed-weight clustering being disconnected from marketing objectives and improving the matching degree between clustering results and marketing strategies. Through the closed-loop feedback module, user response data after marketing execution is fed back to the marketing feedback adaptive clustering module, driving the weight parameters to iteratively optimize based on actual marketing results. This solves the problem of the clustering model lacking a business feedback evolution mechanism, realizing the continuous evolution of clustering strategies and improving marketing accuracy.

[0016] The technical solution of this application will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] Figure 1 This is an overall architecture diagram of a cross-platform transaction flow analysis and precision marketing automated management system according to this application; Figure 2 This is a flowchart of the marketing feedback adaptive clustering process in the embodiments of this application; Figure 3 This is a schematic diagram illustrating the construction of a multi-dimensional user profile in an embodiment of this application. Detailed Implementation

[0018] The following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0019] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning as understood by a person of ordinary skill in the art to which this application pertains.

[0020] The terms "comprising" or "including," as used in this application, mean that the element preceding the term encompasses the element listed after the term, and do not exclude the possibility of encompassing other elements as well. The terms "inner," "outer," "upper," and "lower," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. When the absolute position of the described object changes, the relative positional relationship may also change accordingly. In this application, unless otherwise expressly specified and limited, the term "attached," etc., should be interpreted broadly. For example, it can refer to a fixed connection, a detachable connection, or an integral part; it can refer to a direct connection or an indirect connection through an intermediate medium; it can refer to the internal communication of two elements or the interaction relationship between two elements. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0021] Example 1: A cross-platform transaction flow analysis and precision marketing automated management system, such as Figures 1-3 As shown, it includes: The cross-platform data acquisition module is used to collect raw transaction flow data from multiple heterogeneous trading platforms; The cross-platform data acquisition module includes: The data permission verification submodule is used to verify the validity of data access authorization tokens of each platform before data collection, and to automatically trigger a refresh mechanism when the token expires. Specifically, before the cross-platform data acquisition module executes its acquisition tasks, it verifies the validity of the data access authorization tokens for each heterogeneous trading platform. This submodule maintains an authorization credential repository for each platform, storing at least one type of authorization information, including OAuth Token, API Key, digital certificate, and signing key, for each platform's API access requirements.

[0022] When a data collection task is triggered or when a preset time interval has elapsed since the last verification, this submodule initiates a token validity probe request to each platform, parsing the returned status code or error message to determine whether the current token is valid. When a token is detected to be expired, revoked, or about to reach its expiration threshold, a token refresh mechanism is automatically triggered: for the OAuth 2.0 protocol, a Refresh Token is used to obtain a new Access Token; for a custom API Key mechanism, the platform-provided key rotation interface is called or a temporary credential is re-applied for. Upon successful refresh, the new token is updated to the authorization credential repository, and the refresh timestamp and log information are recorded. If multiple refreshes fail consecutively, an alarm message is generated to notify operations personnel, and the platform is marked as "temporarily unavailable." Simultaneously, locally cached historical data is enabled as a fallback solution to ensure the continuity of the overall data collection task.

[0023] In multi-platform concurrent data collection scenarios, token refresh requests from different platforms are asynchronously decoupled to avoid blocking data collection on other platforms due to token refresh on a single platform.

[0024] Through the above mechanism, the data permission verification submodule realizes automated authorization management in the cross-platform data collection process, effectively solving problems such as scattered token management across multiple platforms, frequent manual intervention, and collection interruption due to authorization failure, and providing stable and continuous data input guarantee for the subsequent data cleaning and standardization modules.

[0025] The incremental synchronization submodule is used to collect transaction records that have been added or changed since the last synchronization time, based on the timestamp field of the transaction flow of each platform.

[0026] Based on the timestamp field of transaction records from each platform, this submodule collects transaction records that have been added or changed since the last synchronization time. It maintains an independent synchronization status record for each connected platform, including the timestamp of the last successful synchronization and the starting cursor for the current synchronization.

[0027] During task scheduling, this submodule reads the synchronization status record of the target platform, constructs incremental query conditions starting from the last successful synchronization timestamp, and initiates a data request to the platform. The collected transaction records are written to a standardized cache queue after time-series consistency verification. Data with verification errors enters the exception retry queue and is processed first in the next synchronization cycle.

[0028] This submodule is also configured to dynamically allocate acquisition thread resources based on the data update frequency and interface rate limiting strategy of each platform in multi-platform concurrent acquisition scenarios. When acquisition fails due to an anomaly on a certain platform, the current task is paused and the breakpoint state is retained. After the anomaly is resolved, the acquisition can resume from the breakpoint to avoid data loss.

[0029] Through the above mechanism, the incremental synchronization submodule achieves efficient incremental collection of cross-platform transaction flow, providing complete and continuous data input for the subsequent data cleaning and standardization modules.

[0030] The data cleaning and standardization module, connected to the cross-platform data acquisition module, is used to perform format conversion and cleaning processing on the raw transaction log data to generate a standardized transaction dataset in a unified format. The data cleaning and standardization module includes: The format conversion unit is used to map data fields from different heterogeneous trading platforms to a unified data dictionary, and to perform time zone alignment for transaction times from each platform; the data dictionary includes transaction time, transaction amount, counterparty, transaction type, and commodity category; The anomaly detection unit is used to identify abnormal transaction data, including missing timestamps, zero or negative amounts, and cross-platform time conflict records. For the identified abnormal data, corresponding processing is performed according to the anomaly type: for records with missing timestamps, the timestamps of adjacent valid records are interpolated to fill the gaps; for records with zero or negative amounts, the transactions are marked as invalid and removed; and for records with cross-platform time conflicts, the timestamps are corrected based on time zone alignment rules. The cross-platform duplicate identification unit is used to identify duplicate records of the same transaction on different platforms based on joint matching of transaction amount, transaction time window and counterparty identifier; for the identified duplicate records, one complete record is retained, and the remaining records are marked as cross-platform redundant data and removed. The missing value processing unit is used to clean records that still have missing fields after the above processing by using the corresponding strategies from mean filling, nearest neighbor filling, or whole record removal.

[0031] The multi-dimensional user profile building module, connected to the data cleaning and standardization module, is used to extract user transaction behavior features based on standardized transaction datasets. These user transaction behavior features include continuous attribute features and discrete attribute features. Multi-dimensional user profile building module, such as Figure 3 As shown, it includes: The continuous attribute extraction unit is used to calculate users' continuous transaction behavior indicators based on standardized transaction datasets. The continuous transaction behavior indicators include: the duration of the most recent consumption interval, the frequency of consumption within the statistical period, the total consumption amount within the statistical period, the average amount of a single transaction, the standard deviation of transaction amount, and the ratio of fund inflow to outflow. The discrete attribute extraction unit is used to identify users' discrete transaction behavior preferences based on a standardized transaction dataset. Discrete transaction behavior preferences include: high-frequency transaction category ranking, preferred transaction channels, distribution of active transaction periods, and concentration of counterparty types. The profile vector generation unit is used to vectorize and concatenate continuous transaction behavior indicators with discrete transaction behavior preferences to generate multi-dimensional user profile vectors.

[0032] Marketing feedback adaptive clustering module, such as Figure 2 As shown, it connects to the multi-dimensional user profile building module to receive the target type identifier of the current marketing campaign, determine the initial weight parameters for hybrid attribute clustering based on the preset target-weight mapping relationship, perform hybrid attribute clustering on user transaction behavior features based on the initial weight parameters, generate user grouping results, and receive user response data after execution, identify the dominant attribute features based on the response rate differences of each user group, and dynamically adjust the weight parameters according to the dominant attribute features to generate updated user grouping results. The specific process of the marketing feedback adaptive clustering module is as follows: Figure 2 As shown.

[0033] The marketing feedback adaptive clustering module includes: The initial clustering unit is used to receive the target type identifier of the current marketing campaign, determine the initial weight parameters of the mixed attribute clustering based on the preset target-weight mapping relationship, and perform the first mixed attribute clustering on the user transaction behavior characteristics based on the initial weight parameters to generate the initial user grouping results; The target-weight mapping relationship includes: The category penetration target corresponds to the first initial weight value, which is used to strengthen the weight of discrete attributes to gather users with similar category preferences. The value of the first initial weight value ranges from 1.2 to 1.5. The target for increasing average order value corresponds to the initial value of the second weight, which is used to strengthen the weight of continuous attributes to gather users with similar spending power. The value of the initial value of the second weight ranges from 0.3 to 0.5. The initial value of the third weight for user retention targets is used to balance the influence of continuous and discrete attributes. The value of the initial value of the third weight ranges from 0.8 to 1.2. The initial value of the first weight is greater than the initial value of the third weight, and the initial value of the third weight is greater than the initial value of the second weight.

[0034] Hybrid attribute clustering is based on the K-Prototypes algorithm and introduces dynamic weight parameters to weight and fuse the distances of continuous and discrete attributes. The formula for its hybrid distance metric is as follows: ; in, Euclidean distance for continuous properties Hamming distance for discrete attributes For the first tThe dynamic weight parameters for each round of iterations are initially determined by the target type identifier and are iteratively adjusted based on user response data after the marketing campaign is executed.

[0035] In this embodiment, the distance metric D1 for continuous attributes uses Euclidean distance, and its calculation formula is as follows: ; in, The number of dimensions for continuous attributes. For the first i The sample at the th l The values ​​of a continuous attribute For the first j The cluster centers at the in l The values ​​are taken from continuous attributes. In this embodiment, the continuous attributes include the most recent consumption interval, consumption frequency within the statistical period, total consumption amount within the statistical period, average amount per transaction, standard deviation of transaction amount, and capital inflow / outflow ratio, totaling 6 dimensions, hence p=6.

[0036] The distance metric for discrete attributes is Hamming distance, which is calculated using the following formula: ; in, d The total number of feature dimensions. Matching indicator function for discrete attributes: ; In this embodiment, the discrete attributes include high-frequency trading category ranking, preferred trading channels, distribution of active trading periods, and concentration of counterparty types, totaling four dimensions. d - p =4, total dimensions d =10.

[0037] The above and Substituting into the mixed distance metric formula, we get: ; in, For the first t The dynamic weight parameters of the round iteration, their initial values The target type of the current marketing campaign is determined from the preset target-weight mapping relationship, and iteratively corrected based on user response data after the marketing campaign is executed.

[0038] The response rate calculation unit is used to calculate the marketing response rate of each user group based on the collected user response data after the marketing campaign is executed, and to identify user groups with response rates that are significantly higher or lower than the average level. Specifically, the response rate calculation unit is used to quantitatively evaluate the marketing effectiveness of each user group based on the user response data returned by the closed-loop feedback module after the marketing campaign is executed. This unit first analyzes the user response data, extracts key behavioral indicators, including click behavior, conversion behavior, and transaction closed-loop data after the marketing message is delivered, and calculates the marketing response rate for each user group. The marketing response rate is the proportion of users in a user group who have performed the target conversion behavior to the total number of users reached in that group.

[0039] This unit maintains independent response statistics records for each user group, including group identifier, number of users reached, number of users who clicked, number of converted users, and response rate. After calculating the response rate, this unit further calculates the average response rate of the entire user group as a baseline, and uses a preset deviation threshold (such as a multiple of the standard deviation or a fixed percentage) to identify user groups with response rates significantly higher than the average (high response group) and user groups with response rates significantly lower than the average (low response group). The high response group and the low response group constitute the analysis objects of the attribute-dominant discrimination unit, used to subsequently determine the attribute-dominant type and weight adjustment direction of each group.

[0040] In multi-round marketing iteration scenarios, historical response rate data is accumulated and response rate trends are calculated. When the response rate of a user group shows a monotonous decline or abnormal fluctuations for several consecutive rounds, the group is marked as a stability anomaly, and the weight adjustment process is triggered first. Through the above mechanism, the response rate calculation unit realizes the quantitative evaluation of marketing effectiveness and anomaly identification, providing a data-driven decision-making basis for the dynamic weight adjustment of the marketing feedback adaptive clustering module.

[0041] The attribute-dominant discrimination unit is used to determine the attribute-dominant type of user groups with abnormal response rates by comparing the variance of continuous attributes with the purity of discrete attributes. The formula for calculating the purity of discrete attributes is: ; in, For user groups The number of samples, d The total number of feature dimensions. p The number of dimensions for continuous attributes. This is a matching indicator function for discrete attributes. It takes a value of 1 when the discrete attribute value matches, and a value of 0 otherwise.

[0042] The system identifies user groups with abnormal response rates, extracts the user transaction behavior feature vectors for all samples in that group, and calculates the variance of continuous attributes and the purity of discrete attributes. The variance of continuous attributes reflects the dispersion of continuous indicators (such as spending amount and frequency) within the group; a larger variance indicates greater differences in spending power among users within the group. The purity of discrete attributes reflects the consistency of discrete indicators (such as category preference and channel preference) within the group; higher purity indicates more similar behavioral preferences among users within the group.

[0043] The variance of continuous attributes is normalized and compared with the purity of discrete attributes: if the purity of discrete attributes is significantly higher than the variance of continuous attributes, the user group is determined to be dominated by discrete attributes, indicating that the users in this group exhibit high / low responses due to convergent preferences such as product category and channel; if the variance of continuous attributes is significantly higher than the purity of discrete attributes, the user group is determined to be dominated by continuous attributes, indicating that the users in this group exhibit high / low responses due to differences in purchasing power. The discrimination result is output to the weight update unit to determine the adjustment direction of the weight parameters.

[0044] The weight update unit, connected to the attribute dominance discrimination unit and the closed-loop feedback module, is used to determine the weight adjustment direction based on the attribute dominance type and update the weight parameters based on the preset learning rate. When the proportion of discrete attributes dominating in a high-response-rate user group exceeds a preset threshold, increase the weighting parameter. To enhance the influence of discrete attributes in clustering; When the proportion of continuous attributes dominating in a high-response-rate user group exceeds a preset threshold, reduce the weighting parameter. This is to enhance the influence of continuous attributes in clustering.

[0045] The specific update formula for the weight parameters is as follows: ; in: Let be the weight parameters for the (t+1)th iteration (the next iteration). The preset learning rate is used to control the step size of a single weight adjustment. In this embodiment... The value is 0.1 (the range is from 0.05 to 0.2). To adjust the direction sign, the following rules apply: When it is necessary to increase the weight =+1; When it is necessary to reduce the weight =-1; When neither is triggered (i.e., the proportion of both dominant types does not exceed the threshold), = 0, keeping the weight parameters unchanged.

[0046] Updated weight parameters Constrained within a preset effective range (e.g., [0.1, 5.0]) to prevent extreme values ​​from causing clustering failure.

[0047] The clustering iteration unit, connected to the initial clustering unit and the weight update unit, is used to: output the initial user clustering results during the first run; and in subsequent iterations, re-execute the mixed attribute clustering based on the updated weight parameters output by the weight update unit to generate updated user clustering results.

[0048] The convergence judgment unit stops iterative optimization and outputs the current user grouping result as the final grouping result when the change of the weight parameter in multiple consecutive iterations is lower than the preset convergence threshold.

[0049] When the change in weight parameters is lower than the preset convergence threshold in multiple consecutive iterations (e.g., | - When | < 0.01), the iteration terminates; when the weight parameters are detected to be oscillating continuously at the boundary of the effective range, the learning rate is automatically reduced to promote stable convergence. The updated weight parameters are output to the clustering iteration unit to drive the next round of hybrid attribute clustering.

[0050] The marketing strategy generation module is connected to the marketing feedback adaptive clustering module to generate personalized marketing plans based on user segmentation results. Specifically, this module receives user segmentation results from the marketing feedback adaptive clustering module. These results include at least the group identifier, group member list, group center vector, and statistical information on the sample size of each user group. The marketing strategy generation module has a pre-set marketing strategy library, which maintains differentiated marketing action configurations for different user group types. For example, for high-response-rate user groups dominated by discrete attributes (such as users with concentrated preferences for high-frequency transaction categories), the marketing strategy generation module is configured to generate precise category recommendation schemes, including pushing coupons, limited-time discounts, or new product trial information related to the user group's preferred categories; for high-response-rate user groups dominated by continuous attributes (such as users with high spending amounts and high-frequency transactions), the marketing strategy generation module is configured to generate average order value enhancement schemes, including pushing discount coupons, bundled offers, or exclusive membership benefits; for user groups with low response rates, the marketing strategy generation module is configured to generate wake-up or activation schemes, including pushing general coupons, sign-in rewards, or limited-time event notifications. The marketing strategy generation module also supports A / B testing configuration, allowing multiple candidate marketing plans to be generated for the same user group and allocated for execution according to a preset ratio, so as to select the optimal plan through actual response data. The generated personalized marketing plans are output to the automated execution module in a structured data format (such as JSON). The plan includes at least the target user group identifier, marketing content template ID, reach channels (such as app push, SMS, email, in-app message), and planned execution time window, thus providing clear execution instructions for the automated execution module.

[0051] The automated execution module connects to the marketing strategy generation module to execute personalized marketing plans; Specifically, this module receives the structured marketing plan output by the marketing strategy generation module, parses the target user group identifier, marketing content template ID, reach channel and execution time window, and calls the corresponding channel API through the internally maintained channel adapter pool (supporting App push, SMS, email and in-site message) to complete the task distribution.

[0052] The closed-loop feedback module, connected to the automated execution module and the marketing feedback adaptive clustering module, is used to send user response data back to the marketing feedback adaptive clustering module to drive the iterative optimization of weight parameters.

[0053] Specifically, after receiving user response data, the module calculates the marketing response rate (number of converted users / number of reached users) for each user group and pushes the aggregated response data to the response rate calculation unit through an asynchronous message queue, triggering a cascading operation of attribute-driven discrimination, weight update, and group iteration.

[0054] To further verify the technical effectiveness of this system, the applicant conducted a comparative experiment with an existing fixed-weight K-Prototypes clustering system. The experimental data came from 90 consecutive days of anonymized transaction records from a third-party payment platform, with a sample size of approximately 100,000 users. The experimental group used this system for five rounds of closed-loop iterative optimization, while the control group used a fixed weight (α=1.0) for a single clustering and executed the same marketing campaign. The experimental results show that after the third iteration, the marketing conversion rate of the high-response-rate user group in the experimental group increased by approximately 15.2% compared to the control group; after the fifth iteration, the variance of the response rate within the cluster decreased by approximately 32% compared to the control group, indicating that this system can effectively converge to a better clustering strategy. In summary, this system significantly outperforms existing technologies in both clustering accuracy and marketing conversion effectiveness.

[0055] Therefore, this application adopts the aforementioned cross-platform transaction flow analysis and precision marketing automated management system. It integrates multi-source heterogeneous transaction data through a cross-platform data collection and standardization module, extracts continuous and discrete transaction behavior features through a multi-dimensional user profile construction module, adaptively adjusts clustering weights based on target type through a marketing feedback adaptive clustering module and dynamically iterates and optimizes the clustering results based on user response data through a closed-loop feedback module, and continuously corrects the weight parameters by transmitting marketing effects back through a closed-loop feedback module. This achieves unified integration of cross-platform user data, adaptive clustering driven by marketing objectives, and closed-loop iterative optimization based on actual marketing effects, significantly improving the matching accuracy between user clustering and marketing objectives, and the return on investment of marketing activities.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and not to limit them. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of this application, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of this application.

Claims

1. A cross-platform transaction flow analysis and precision marketing automated management system, characterized in that, include: The cross-platform data acquisition module is used to collect raw transaction flow data from multiple heterogeneous trading platforms. The data cleaning and standardization module, connected to the cross-platform data acquisition module, is used to perform format conversion and cleaning processing on the raw transaction log data to generate a standardized transaction dataset in a unified format. The multi-dimensional user profile building module, connected to the data cleaning and standardization module, is used to extract user transaction behavior features based on standardized transaction datasets. These user transaction behavior features include continuous attribute features and discrete attribute features. The marketing feedback adaptive clustering module, connected to the multi-dimensional user profile building module, is used to receive the target type identifier of the current marketing campaign and determine the initial weight parameters of the mixed attribute clustering based on the preset target-weight mapping relationship. Based on the initial weight parameters, hybrid attribute clustering is performed on the user transaction behavior characteristics to generate user grouping results; In addition, it receives user response data after execution, identifies the dominant features of attributes based on the differences in response rates of each user group, and dynamically adjusts the weight parameters according to the dominant features of attributes to generate updated user grouping results. The marketing strategy generation module is connected to the marketing feedback adaptive clustering module to generate personalized marketing plans based on user segmentation results. The automated execution module connects to the marketing strategy generation module to execute personalized marketing plans; The closed-loop feedback module, connected to the automated execution module and the marketing feedback adaptive clustering module, is used to send user response data back to the marketing feedback adaptive clustering module to drive the iterative optimization of weight parameters.

2. The cross-platform transaction flow analysis and precision marketing automated management system according to claim 1, characterized in that, The cross-platform data acquisition module includes: The data permission verification submodule is used to verify the validity of data access authorization tokens of each platform before data collection, and to automatically trigger a refresh mechanism when the token expires. The incremental synchronization submodule is used to collect transaction records that have been added or changed since the last synchronization time, based on the timestamp field of the transaction flow of each platform.

3. The cross-platform transaction flow analysis and precision marketing automated management system according to claim 2, characterized in that, The data cleaning and standardization module includes: The format conversion unit is used to map data fields from different heterogeneous trading platforms to a unified data dictionary, and to perform time zone alignment for transaction times from each platform; the data dictionary includes transaction time, transaction amount, counterparty, transaction type, and commodity category; The anomaly detection unit is used to identify abnormal transaction data, including missing timestamps, zero or negative amounts, and cross-platform time conflict records. For the identified abnormal data, corresponding processing is performed according to the anomaly type: for records with missing timestamps, the timestamps of adjacent valid records are interpolated to fill the gaps; for records with zero or negative amounts, the transactions are marked as invalid and removed; and for records with cross-platform time conflicts, the timestamps are corrected based on time zone alignment rules. The cross-platform duplicate identification unit is used to identify duplicate records of the same transaction on different platforms based on joint matching of transaction amount, transaction time window and counterparty identifier; for the identified duplicate records, one complete record is retained, and the remaining records are marked as cross-platform redundant data and removed. The missing value processing unit is used to clean records that still have missing fields after the above processing by using the corresponding strategies from mean filling, nearest neighbor filling, or whole record removal.

4. The cross-platform transaction flow analysis and precision marketing automated management system according to claim 1, characterized in that, The multi-dimensional user profile construction module includes: The continuous attribute extraction unit is used to calculate users' continuous transaction behavior indicators based on standardized transaction datasets. The continuous transaction behavior indicators include: the duration of the most recent consumption interval, the frequency of consumption within the statistical period, the total consumption amount within the statistical period, the average amount of a single transaction, the standard deviation of transaction amount, and the ratio of fund inflow to outflow. The discrete attribute extraction unit is used to identify users' discrete transaction behavior preferences based on a standardized transaction dataset. Discrete transaction behavior preferences include: high-frequency transaction category ranking, preferred transaction channels, distribution of active transaction periods, and concentration of counterparty types. The profile vector generation unit is used to vectorize and concatenate continuous transaction behavior indicators with discrete transaction behavior preferences to generate multi-dimensional user profile vectors.

5. The cross-platform transaction flow analysis and precision marketing automated management system according to claim 1, characterized in that, The marketing feedback adaptive clustering module includes: The initial clustering unit is used to receive the target type identifier of the current marketing campaign, determine the initial weight parameters of the mixed attribute clustering based on the preset target-weight mapping relationship, and perform the first mixed attribute clustering on the user transaction behavior characteristics based on the initial weight parameters to generate the initial user grouping results; The response rate calculation unit is used to calculate the marketing response rate of each user group based on the collected user response data after the marketing campaign is executed, and to identify user groups with response rates that are significantly higher or lower than the average level. The attribute-dominant discrimination unit is used to determine the attribute-dominant type of user groups with abnormal response rates by comparing the variance of continuous attributes with the purity of discrete attributes. The weight update unit, connected to the attribute dominance discrimination unit and the closed-loop feedback module, is used to determine the weight adjustment direction based on the attribute dominance type and update the weight parameters based on the preset learning rate. The clustering iteration unit, connected to the initial clustering unit and the weight update unit, is used to: output the initial user clustering results during the first run; and in subsequent iterations, re-execute the mixed attribute clustering based on the updated weight parameters output by the weight update unit to generate updated user clustering results.

6. The cross-platform transaction flow analysis and precision marketing automated management system according to claim 5, characterized in that, The hybrid attribute clustering is based on the K-Prototypes algorithm and introduces dynamic weight parameters to weight and fuse continuous attribute distances and discrete attribute distances. The hybrid distance metric formula is as follows: ； in, Euclidean distance for continuous properties Hamming distance for discrete attributes For the first t The dynamic weight parameters for each round of iterations are initially determined by the target type identifier and are iteratively adjusted based on user response data after the marketing campaign is executed.

7. The cross-platform transaction flow analysis and precision marketing automated management system according to claim 5, characterized in that, The formula for calculating the purity of the discrete attribute is: ； in, For user groups The number of samples, d The total number of feature dimensions. p The number of dimensions for continuous attributes. This is a matching indicator function for discrete attributes. It takes a value of 1 when the discrete attribute value matches, and a value of 0 otherwise.

8. The cross-platform transaction flow analysis and precision marketing automated management system according to claim 5, characterized in that, The weight update unit is configured as follows: When the proportion of discrete attributes dominating in a high-response-rate user group exceeds a preset threshold, increase the weighting parameter. To enhance the influence of discrete attributes in clustering; When the proportion of continuous attributes dominating in a high-response-rate user group exceeds a preset threshold, reduce the weighting parameter. This is to enhance the influence of continuous attributes in clustering.

9. The cross-platform transaction flow analysis and precision marketing automated management system according to claim 1, characterized in that, The target-weight mapping relationship includes: The category penetration target corresponds to the first initial weight value, which is used to strengthen the weight of discrete attributes to gather users with similar category preferences. The value of the first initial weight value ranges from 1.2 to 1.

5. The target for increasing average order value corresponds to the initial value of the second weight, which is used to strengthen the weight of continuous attributes to gather users with similar spending power. The value of the initial value of the second weight ranges from 0.3 to 0.

5. The initial value of the third weight for user retention targets is used to balance the influence of continuous and discrete attributes. The value of the initial value of the third weight ranges from 0.8 to 1.

2. Among them, the initial value of the first weight is greater than the initial value of the third weight, and the initial value of the third weight is greater than the initial value of the second weight.

10. The cross-platform transaction flow analysis and precision marketing automated management system according to claim 5, characterized in that, The marketing feedback adaptive clustering module also includes a convergence judgment unit, which is used to: stop iterative optimization and output the current user grouping result as the final grouping result when the change of the weight parameter in multiple consecutive iterations is lower than the preset convergence threshold.

Images