Marketing activity management method, device, computer equipment and storage medium

Abstract

The embodiments of the present application disclose a marketing activity management method, apparatus, computer equipment and storage medium. The method includes: when the refund information of the target equity order in the equity order pool is obtained, the target order actual consumption index of the target equity order is updated according to the refund information, and the target customer actual consumption index corresponding to the target customer identifier is updated according to the updated target order actual consumption index; according to the target equity issuance rules and the preset AI anomaly monitoring model, etc., it is determined whether the target customer is a risk customer; if the target customer is a risk customer, the issued equity of the target customer is recovered; according to the preset customer stratification algorithm, the target customer level corresponding to the target customer is determined; according to the preset differentiated compensation strategy, the target compensation strategy corresponding to the target customer level is determined; according to the target compensation strategy, the target customer is executed according to the target compensation strategy. By implementing the method of the embodiment of the present application, the customer retention rate can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to marketing activity management methods, devices, computer equipment, and storage media. Background Art

[0002] Current marketing campaign management systems primarily implement the following process for benefit recovery and customer retention: First, the system detects unusual consumer behavior and benefit usage through regular batch processing (optimized to a roughly one-hour cycle), flagging suspicious accounts based on pre-set static threshold rules. These flagged anomalies are then manually reviewed and confirmed by operations personnel, who then manually execute the benefit recovery process. This typically involves freezing accounts, revoking benefits, or adjusting points. The entire process, from the time the unusual behavior is detected to the completion of benefit recovery, can take several hours or even longer.

[0003] When it comes to customer retention, the system primarily relies on a standardized complaint handling process. When a customer complains about dissatisfaction with their benefits redemption, customer service staff will provide a uniform payment or compensation based on a pre-set compensation plan, such as bonus points or coupons. These compensation plans are often standardized, one-size-fits-all measures that fail to differentiate based on customer value. This results in high-value customers receiving the same retention experience as regular customers.

[0004] Current marketing campaign management systems still have shortcomings in terms of reclaiming benefits and retaining customers. First, there is a significant time lag in reclaiming benefits. It can take hours or even longer from the time abusive behavior occurs to when the benefits are actually reclaimed. This results in significant losses before abusive behavior is stopped. Second, customer retention relies on standardized, one-size-fits-all compensation measures, failing to intelligently differentiate based on customer value. High-value customers receive the same experience as regular customers, severely impacting the retention rate of key customers (such as high-value and potential customers). Summary of the Invention

[0005] The embodiments of the present application provide a marketing activity management method, apparatus, computer equipment, and storage medium, which are intended to address the problems in the prior art of long intervals between the occurrence of abnormal behavior and the completion of actual equity recovery, and the use of one-size-fits-all compensation measures when retaining customers, which affects the retention rate of key customers.

[0006] In a first aspect, an embodiment of the present application provides a marketing activity management method, which includes:

[0007] The method is applied to a marketing activity management system, wherein a rights order pool is maintained in the marketing activity management system, wherein the rights order pool includes rights orders respectively associated with a plurality of customer identifiers, wherein the rights orders are associated with issued rights, and the method comprises:

[0008] When refund information of a target equity order in the equity order pool is obtained, the target order actual consumption index of the target equity order is updated according to the refund information, and the target customer actual consumption index corresponding to the target customer identifier is updated according to the updated target order actual consumption index, where the target customer identifier is the customer identifier corresponding to the target equity order;

[0009] Determining whether the target customer corresponding to the target customer identifier is a risky customer based on the target customer's actual consumption index, the target equity issuance rule, the first multi-dimensional user feature data corresponding to the target customer identifier, and a preset AI anomaly monitoring model, wherein the target equity issuance rule is the equity issuance rule associated with the issued equity corresponding to the target equity order;

[0010] If the target customer is a risky customer, the issued rights and interests of the target customer will be recovered;

[0011] Determining a target customer level corresponding to the target customer based on the second multi-dimensional user feature data corresponding to the target customer identifier and a preset customer stratification algorithm;

[0012] Determining a target compensation strategy corresponding to the target customer level according to a preset differentiated compensation strategy;

[0013] Execute equity compensation processing on the target customer according to the target compensation strategy.

[0014] In a second aspect, an embodiment of the present application further provides a marketing activity management device, the marketing activity management device being deployed in a marketing activity management system, the marketing activity management system maintaining a rights order pool, the rights order pool including rights orders associated with multiple customer identifiers, the rights orders being associated with issued rights, the marketing activity management device comprising:

[0015] A transceiver unit, configured to obtain refund information of a target equity order in the equity order pool;

[0016] A processing unit is configured to, when refund information of a target equity order in the equity order pool is obtained, update the target order actual consumption index of the target equity order according to the refund information, and update the target customer actual consumption index corresponding to the target customer identifier according to the updated target order actual consumption index, where the target customer identifier is the customer identifier corresponding to the target equity order; determine whether the target customer corresponding to the target customer identifier is a risk customer based on the target customer actual consumption index, the target equity issuance rule, the first multi-dimensional user feature data corresponding to the target customer identifier, and a preset AI anomaly monitoring model, where the target equity issuance rule is the equity issuance rule associated with the issued equity corresponding to the target equity order; if the target customer is a risk customer, recover the issued equity of the target customer; determine the target customer level corresponding to the target customer based on the second multi-dimensional user feature data corresponding to the target customer identifier and a preset customer stratification algorithm; determine the target compensation strategy corresponding to the target customer level based on a preset differentiated compensation strategy; and perform equity compensation processing on the target customer based on the target compensation strategy.

[0017] In a third aspect, an embodiment of the present application further provides a computer device, which includes a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above method when executing the computer program.

[0018] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, wherein the storage medium stores a computer program, wherein the computer program includes program instructions, and the program instructions can implement the above method when executed by a processor.

[0019] The embodiment of the present application provides a marketing activity management method, apparatus, computer equipment and storage medium. The method is applied to a marketing activity management system, wherein a rights order pool is maintained in the marketing activity management system, wherein the rights order pool includes rights orders associated with multiple customer identifiers, and the rights orders are associated with issued rights. The method includes: when the refund information of the target rights order in the rights order pool is obtained, the target order actual consumption index of the target rights order is updated according to the refund information, and the target customer actual consumption index corresponding to the target customer identifier is updated according to the updated target order actual consumption index, wherein the target customer identifier is the customer identifier corresponding to the target rights order; according to the target customer actual consumption index, the target rights issuance rule, and the target rights issuance rule, the target customer actual consumption index is updated according to the target customer actual consumption index. The first multi-dimensional user feature data corresponding to the target customer identifier and a preset AI anomaly monitoring model are used to determine whether the target customer corresponding to the target customer identifier is a risky customer. The target equity issuance rule is the equity issuance rule associated with the issued equity corresponding to the target equity order. If the target customer is a risky customer, the issued equity of the target customer is recovered. The target customer tier corresponding to the target customer is determined based on the second multi-dimensional user feature data corresponding to the target customer identifier and a preset customer stratification algorithm. The target compensation strategy corresponding to the target customer tier is determined based on a preset differentiated compensation strategy. Equity compensation processing is performed on the target customer based on the target compensation strategy. On the one hand, when this solution monitors the existence of refund information for equity orders, it conducts risk detection on the corresponding customer. When the customer is detected as a risky customer, risk recovery is performed, thereby reducing the interval between the occurrence of abnormal behavior and the actual equity recovery, and preventing the abuse of equity. On the other hand, this solution intelligently stratifies risky customers and sets differentiated compensation strategies for customers of different tiers. Different compensation strategies are implemented for customers of different tiers. Compared with the adoption of a one-size-fits-all compensation measure, this solution can improve customer retention rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0021] Figure 1 A schematic diagram of the structure of a marketing activity management system provided in an embodiment of the present application;

[0022] Figure 2 A flowchart of a marketing activity management method provided in an embodiment of the present application;

[0023] Figure 3 A schematic diagram of a sub-process of the marketing activity management method provided in an embodiment of the present application;

[0024] Figure 4 A schematic diagram of a sub-process of the marketing activity management method provided in an embodiment of the present application;

[0025] Figure 5 A schematic block diagram of a marketing activity management device provided in an embodiment of the present application;

[0026] Figure 6 A schematic block diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0028] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0029] It should also be understood that the terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0030] It should be further understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0031] The embodiments of the present application provide a marketing activity management method, apparatus, computer equipment, and storage medium.

[0032] The executor of the marketing activity management method may be the marketing activity management apparatus provided in the embodiment of the present application, or a computer device integrated with the marketing activity management apparatus, wherein the marketing activity management apparatus may be implemented in the form of hardware or software, and the computer device may be a terminal or a server. The marketing activity management system provided in the present application is composed of the computer device.

[0033] In some embodiments, see Figure 1 The marketing activity management system includes a consumption indicator monitoring module, a risk intelligent identification module, and a customer intelligent retention module. The consumption indicator monitoring module is mainly used to collect payment and refund data at the millisecond level through a streaming computing engine, and dynamically update the customer's actual consumption indicators (actual payment amount minus refund amount, number of valid items minus number of refunds), ensuring that the indicator data is synchronized with customer behavior in real time. The risk intelligent identification module is mainly used to automatically trigger dual precision management when the customer's consumption indicators do not meet the activity rules and the risk score exceeds the threshold: withdrawing unused benefits (coupons, points, etc.), and using natural language processing (NLP) technology to identify and block related marketing messages in real time to avoid resource waste. The customer intelligent retention module is mainly used to automatically implement different compensation strategies for customers at different levels through customer stratification algorithms and differentiated compensation strategies.

[0034] Figure 2 This is a flow chart of the marketing activity management method provided by the embodiment of this application. Figure 2 As shown, the method includes the following steps S110-S160, wherein step S110 is executed based on the consumption indicator monitoring module in the marketing activity management system, steps S120-S130 are executed based on the risk intelligent identification module in the marketing activity management system, and steps S140-S160 are executed based on the customer intelligent retention module in the marketing activity management system. An equity order pool is maintained in the marketing activity management system, and the equity order pool includes equity orders associated with multiple customer identifiers, and the equity orders are associated with issued equity.

[0035] S110. When the refund information of the target equity order in the equity order pool is obtained, the target order actual consumption index of the target equity order is updated according to the refund information, and the target customer actual consumption index corresponding to the target customer identifier is updated according to the updated target order actual consumption index, where the target customer identifier is the customer identifier corresponding to the target equity order.

[0036] In this embodiment, after determining the actual consumption index of the target order, the consumption index monitoring module also writes the actual consumption index of the target order and the actual consumption index of the target customer into the ClickHouse table of the equity order pool (the core data storage structure in the ClickHouse column database), and sends consumption index update information (corresponding to the refund information) to the risk intelligent identification model. The consumption index update information carries the target customer identifier. The ClickHouse table stores the actual consumption index of each equity order, order attribute information, and the actual consumption index of the customer corresponding to each customer identifier.

[0037] Specifically, in some embodiments, the consumption index monitoring module provided by the present application can be used to monitor the actual consumption indicators of all orders (actual consumption amount and actual number of consumption items, etc.), and store the obtained actual consumption indicators of each order in the ClickHouse table. Since the orders in the equity order pool are associated with the customer's issued equity, in order to avoid the abuse of equity (such as wool plucking), this embodiment needs to perform risk monitoring on all orders in the equity order pool. Specifically, when it is monitored that an order in the equity order pool involves a refund, at this time, the consumption index monitoring module needs to send information to the risk intelligent identification module (that is, after the target customer's actual consumption indicator corresponding to the target customer identifier is updated according to the updated target order actual consumption indicator, the method further includes: through The consumption index monitoring module writes the actual consumption index of the target order and the actual consumption index of the target customer into the ClickHouse table of the equity order pool, and sends consumption index update information to the risk intelligent identification model. The consumption index update information carries the target customer identifier. The ClickHouse table stores the actual consumption index of each equity order, order attribute information, and the actual consumption index of each customer identifier. The risk intelligent identification module identifies whether the customer corresponding to the order involving a refund is a risky customer. If the customer is a risky customer, the customer's equity needs to be recovered, and the risky customers are stratified through the customer intelligent retention module. Finally, differentiated compensation operations are performed on customers of different levels based on the stratification results.

[0038] Among them, the consumption indicator monitoring module provided in this embodiment can deploy Flink (an open source distributed stream processing framework) streaming computing tasks based on the customer data platform (CDP) fact model, and achieve millisecond-level updates of customers' actual consumption indicators (core consumption indicators such as the customer's actual payment amount and the number of items), providing a real-time data basis for the decision-making of the subsequent risk intelligent identification module.

[0039] Specifically, the consumption metrics monitoring module provided in this embodiment monitors consumption metrics through the following structure and process. First, the module executes the data collection and merging process. By constructing a Flink task flow and subscribing to both the payment and refund Kafka topics, it merges the collected payment and refund streams. After receiving the information stream, it defines business rules (e.g., 1) payment order types not participating in marketing activities include gift card orders, distribution purchase orders, gift orders, and 0-yuan lottery group purchase orders; 2) refund types that do not require processing, such as refunds due to overdue payment, product exchanges, and 0-yuan lottery group purchase refunds). Machine learning algorithms (such as isolation forests and local outlier factors (LOF)) are then used to detect anomalies in payment and refund data, identifying abnormal payment or refund data, such as abnormal amounts or time periods. This abnormal data may interfere with the normal synchronization process. AI technology is used to preemptively detect, flag, or filter out abnormal data, ensuring data quality for subsequent synchronization. Secondly, we use sequential pattern mining algorithms (such as Prefix-Projected Pattern Growth (PrefixSpan)) to analyze payment and refund data according to time series or other business-related sequences. This allows us to identify common payment and refund sequence patterns, such as payment before refund, and periodic payments and refunds. Based on these patterns, we can better predict and schedule the timing of simultaneous processing of payment and refund data, improving processing efficiency.

[0040] Next, data completion (such as completing order attribute information) and status verification are performed. A framework for preventing data loss is built using asynchronous remote procedure calls (RPCs) and an intelligent retry mechanism. Threshold settings are more intelligent (for example, by collecting RPC call performance data, predicting call times, and dynamically adjusting timeout thresholds based on the predicted results). When a call to an external service encounters temporary failures such as network jitter, retries are performed according to a preset retry strategy (such as retrying three times at a fixed interval). This prevents data loss caused by brief network issues and prevents task flows from being blocked due to prolonged wait times. Dubbo (an open source distributed service framework) interfaces are used to query real-time order status (such as order attribute information (item details), shipping amount, actual payment amount, etc.) and order refund information (such as the latest refund status). By comparing and verifying the data returned by the interface with the currently processed data, missing information is supplemented and inconsistencies are corrected to ensure data integrity and accuracy.

[0041] Finally, we built a CDP fact model rule calculation engine. The first step was to define the model output fields. This clearly defined the output fields of the CDP fact model (such as the actual order payment minus the refund amount, the number of items ordered minus the number of refunds, and other key consumer indicator fields), providing clear output targets for subsequent calculations. The second step was to define rules: For example, the 1) item number deduction rule stipulates that the number of items paid is only deducted when all items are refunded. For example, if five items of item A were originally purchased and a partial refund was made, the number of items paid is not deducted. Only when the full refund is made does the corresponding number of items be deducted. The calculation engine executes this rule by comparing the number of refunded items with the original number of items purchased. 2) The actual amount paid deduction rule: The actual amount paid for an order is calculated as the original payment minus the refund amount (including shipping costs); the actual amount paid for an item is calculated as the original price minus the refund amount (excluding shipping costs). This amount deduction rule is implemented by programming the corresponding mathematical logic within the calculation engine. The third step involves intelligent computing logic generation. This involves collecting historical computing logic and data samples to construct a training dataset. Deep learning models (such as Long Short-Term Memory (LSTM) and Transformers) are then used to model and predict the computing logic. For example, based on input data characteristics and business rules, corresponding computing logic expressions or algorithmic steps are predicted. Within the computing engine, the predicted computing logic is combined with actual business needs to dynamically generate customized computing logic, improving computational efficiency and accuracy. The computing logic is represented as genetic code and optimized using a genetic algorithm. A fitness function is defined to evaluate the computing logic based on the accuracy and efficiency of the computational results. Through genetic operations such as selection, crossover, and mutation, the computing logic is continuously evolved to adapt to complex business scenarios and data characteristics. If supplementary data is needed during the calculation process (for example, querying extended product information (brand, category, SPUK / SU code, etc.) from an HBase table using the key product ID (item_id) or product specification ID (sku_id)), Flink's asynchronous input / output (I / O) functionality enables efficient data query interaction with HBase, ensuring the integrity of the data required for calculation. Finally, the data is assembled and the calculated indicator values ​​are output according to a preset data format (such as JSON). After assembly, they are sent to a Kafka topic. A separate Flink task subscribes to this topic and writes the data to a ClickHouse table through aggregation operations. The ClickHouse table engine uses MergeTree and shards by user ID to improve query efficiency. A materialized view is also created to pre-aggregate real-time customer indicators, facilitating subsequent rapid data analysis and query utilization.

[0042] Specifically, in the aforementioned solution for the consumption metrics monitoring module, the payment and refund topics can be merged into a unified event stream using the Flink Union operator. An event header tagging mechanism (source_type = 1 / 2 indicates payment / refund) is introduced, enabling event type identification with O(1) complexity, avoiding the need for repetitive development of consumption logic. Business rule management breaks away from traditional hard-coding models, employing Groovy Domain-Specific Language (DSL) scripts to encapsulate complex filtering logic. This, combined with the distributed configuration capabilities of ZooKeeper (an open-source distributed coordination service), enables millisecond-level rule updates and supports simultaneous push across multiple clusters. For anomaly detection, an optimized isolation forest algorithm (using 50 decision trees with a maximum depth of 12) is embedded in the stream processing pipeline, constructing an 18-dimensional feature vector to analyze transaction anomalies. A specific HBase table structure (TTL = 90 days) is created to store detection results and manual feedback, enabling self-iterative model evolution. The data completion process makes a breakthrough by applying time series analysis. Based on the RPC performance indicators collected by Prometheus (an open source monitoring and alarm system) over the past seven days, an ARIMA(2,1,1) model (autoregressive integrated moving average model (2nd-order autoregression, 1st-order difference, 1st-order moving average)) is constructed. This achieves precise dynamic timeout control (timeout = predicted value + 3σ) and combines multi-level priority queues to implement an intelligent retry strategy: an exponential backoff algorithm is used for high-value data with a transaction amount greater than 1,000 yuan (initial interval 100ms, maximum 5 retries), while a fixed interval strategy (200ms, maximum 3 times) is used for low-priority data. In the design of the computing engine, through the "atomic operation-gene encoding" model, complex computing rules are broken down into 32 basic operations and encoded into structured gene fragments (such as ADD: field1, field2, result represents the addition of fields). A weighted and adjustable multi-objective fitness function fitness = 0.7*accuracy + 0.3*execution efficiency is constructed. Through the tournament selection mechanism, two-point crossover operation and 5% directional mutation rate, intelligent optimization of computing logic is achieved, providing a millisecond-level data foundation for real-time marketing decisions.

[0043] The above solution can realize the real-time update of the actual consumption indicators of all orders in the ClickHouse table, as well as the real-time update of the actual consumption indicators of each customer.

[0044] S120. Determine whether the target customer corresponding to the target customer identifier is a risky customer based on the target customer's actual consumption index, the target rights issuance rule, the first multi-dimensional user feature data corresponding to the target customer identifier, and a preset AI anomaly monitoring model. The target rights issuance rule is the rights issuance rule associated with the issued rights corresponding to the target rights order.

[0045] Among them, the first multi-dimensional user feature data includes transaction behavior features, transaction timing features, user portrait features and associated network features.

[0046] In this embodiment, when a refund operation is detected for an order in the equity order pool, it is necessary to perform risk identification processing on the customer corresponding to the order to determine whether the customer is a risky customer to avoid the abuse of equity.

[0047] In some embodiments, see Figure 3 , step S120 includes:

[0048] S1201: Determine whether the target customer meets the target benefit issuance rules based on the target customer's actual consumption index;

[0049] S1202: If the target customer does not meet the target benefit issuance rule, obtain the first multi-dimensional user feature data;

[0050] S1203: Perform a risk score on the target customer based on the AI ​​anomaly monitoring model and the first multi-dimensional user feature data to obtain a target risk score;

[0051] S1204: If the target risk score is greater than a preset risk score threshold, determine that the target customer is the risk customer;

[0052] S1205: If the target customer meets the target rights issuance rule, or the target risk score is less than or equal to the risk score threshold, the target customer is determined to be a common abnormal customer.

[0053] Specifically, the risk intelligent identification module provided by this implementation constructs a complete process of the dual-channel management and control mechanism of rules and AI models. When the real-time indicator data of the CDP fact model falls into ClickHouse, the consumption indicator monitoring module sends an NSQ (an open source distributed real-time message queue system) message that the indicator calculation is completed (the consumption indicator monitoring module sends consumption indicator update information to the risk intelligent identification module). The risk intelligent identification module responds to the consumption indicator update information, determines the target rights issuance rule corresponding to the target customer identifier, and performs semantic analysis on the target rights issuance rule to generate a target structured query language (Structured Query Language, SQL) statement; obtains the target customer's actual consumption indicator and target customer-related indicators from the ClickHouse table according to the target SQL statement; and judges whether the target customer meets the target rights issuance rule based on the target customer's actual consumption indicator and target customer-related indicators.

[0054] Among them, the above-mentioned target customer association indicator refers to the indicator indicated in the target benefit issuance rule for benefit issuance judgment. For example, the target customer association indicator is the specific order actual consumption indicator of the benefit order of a specific category associated with the target customer. At this time, the target benefit issuance rule indicates that when the specific order actual consumption indicator is greater than the first preset amount, or the actual consumption amount in the target customer's actual consumption indicator is greater than the second preset amount, it is determined that the target customer meets the target benefit issuance rule, otherwise it does not meet the requirement.

[0055] Specifically, the intelligent risk identification module dynamically assembles logical instructions from message elements in consumption indicator updates with pre-set target benefit issuance rules (e.g., "Payment amount minus refund ≥ X yuan"). The system then uses an NLP model to automatically parse these rules into SQL statements executable by the CDP, which then queries the customer's behavior before and after the refund in real time. If the customer is determined not to meet the activity rules after the refund (e.g., the payment amount falls below a threshold), the intelligent risk identification module invokes an AI anomaly detection model to perform a risk-weighted assessment of the behavior. This AI anomaly detection model, built using the eXtreme Gradient Boosting (XGBoost) algorithm, collects multi-dimensional user feature data, including transaction behavior (refund frequency, refund amount ratio), transaction timing (refund pattern, transaction-refund interval), user profile features (historical purchase volume, membership level), and associated network features (device ID-linked behavior). The model calculates SHAP values ​​through hierarchical feature importance analysis to identify key risk signals. It also employs imbalanced learning strategies (Synthetic Minority Over-sampling Technique (SMOTE) oversampling and Focal Loss) to enhance anomaly detection. Risk scoring utilizes a weighted scorecard approach, generating a comprehensive score ranging from 0 to 100, which is compared against a dynamic threshold (the risk score threshold). This threshold is dynamically adjusted by a reinforcement learning algorithm based on historical attack data and the cost of false positives to ensure system accuracy. When the risk score exceeds the dynamic threshold, the system triggers dual control measures: automatically recovering issued but unused benefits (coupons, points, etc.) and blocking unsent marketing messages. The system also generates an interpretable rule-matching report (e.g., "Post-refund payment amount is less than the rule threshold of XX yuan") and a detailed list of risk factors (e.g., four similar refunds within 15 days) for merchant activity analysis. The model is continuously updated through incremental learning, ingesting new labeled data every day, identifying new risk patterns, and converting risk scores into understandable business language through the model interpreter, helping operators understand the system's decision-making logic and achieve efficient and accurate marketing activity management.

[0056] Furthermore, this embodiment improves interception rates and reduces false positives by building a parallel "rule-AI" dual-channel decision-making architecture. The rule channel utilizes a customized Bidirectional Encoder Representations from Transformers (BERT) model (12 transformer layers, 768 hidden layers, fine-tuned on over 10 million marketing data) as its core semantic understanding framework. This model parses natural language rules defined by business personnel (e.g., "Payment amount excluding refunds ≥ X yuan") into a structured semantic tree in real time. A three-layer parsing architecture is designed: First, the Named Entity Recognition (NER) layer accurately extracts elements such as the indicator name (payment amount), operator (≥), and threshold (X yuan). Next, the Relationship Extraction layer analyzes the logical relationships between these elements, specifically performing dependency tree analysis on complex conditions such as "and," "or," and "not," transforming complex rules into computable conditional combinations. Finally, the Semantic Matching layer intelligently selects the most matching template from a predefined SQL template library (containing 42 query patterns) and generates the execution SQL through dynamic variable replacement. The system also integrates the Apache Calcite query optimization engine (a framework for query optimization of heterogeneous data sources) to perform syntax validation, permission checks, and execution plan optimization on generated SQL. Specifically targeting ClickHouse's distributed query characteristics, it automatically applies partition pruning and column-level optimization. The AI ​​risk assessment pipeline utilizes an optimized XGBoost algorithm (maximum depth 8, learning rate 0.05, 300 trees, and L1 regularization parameter 0.3). The first multi-dimensional user feature dataset is composed of 237 detailed features across four categories: 1) Transaction behavior features include 60 indicators such as refund frequency, amount share, and category distribution within 7 / 15 / 30 days; 2) Transaction time series features use the Dynamic Time Warping (DTW) algorithm to capture refund time series patterns and combine them with the Fast Fourier Transform (FFT) to extract periodic features and identify anomalous time intervals; 3) User profile features integrate the RFM model and lifecycle metrics to construct spending power and loyalty scores; and 4) Correlation network features use the GraphSAGE algorithm to analyze risk propagation patterns among users linked to device IDs and IP addresses. SMOTE oversampling (expanding the minority class by 300%) combined with Focal Loss (γ=2.0) was used to address sample imbalance. Automatic feature crossover was achieved through a feature concatenation DNN. The Kernel Explainer (SHAP tool) was used to calculate SHAP values, quantifying each feature's contribution to the risk score. The raw scores were then mapped to a 0-100 risk score using a piecewise function.Based on the dynamic threshold optimization mechanism of dual Q-learning, risk threshold setting is transformed into a reinforcement learning problem in a continuous state space. A cost-sensitive reward function, R(s, a) = α × (anti-fraud benefit) - β × (misjudgment cost) + γ × (adjustment penalty), guides policy learning and enables adaptive threshold adjustment. R(s, a) is the cost-sensitive reward function, and α, β, and γ are the corresponding weights. The scoring results trigger millisecond-level dual control via a Redis distributed lock, and a visual report containing the top 5 risk factors is generated.

[0057] S130: If the target customer is a risky customer, recover the issued rights and interests of the target customer.

[0058] In this embodiment, when it is identified that the target customer is a risky customer, the issued rights and interests associated with the target rights and interests order of the customer are recovered, and marketing messages that have not yet been sent to the target customer are intercepted.

[0059] It can be seen that the risk intelligent identification module provided by this application can accurately compare the customer's real-time consumption indicators with the corresponding rights and interests issuance rules, and at the same time introduce an AI anomaly monitoring model to assist in judgment, automatically triggering rights and interests recovery and marketing message blocking when an anomaly is detected, to achieve precise management and control.

[0060] S140: Determine a target customer level corresponding to the target customer based on the second multi-dimensional user feature data corresponding to the target customer identifier and a preset customer stratification algorithm.

[0061] The second multi-dimensional user feature data includes value dimension features, behavior dimension features, loyalty dimension features, and preference dimension features.

[0062] In some embodiments, see Figure 4 , step S140 includes:

[0063] S1401. Acquire the second multi-dimensional user feature data through the feature input interface of the customer intelligent retention module;

[0064] S1402: Determine the target customer group identifier corresponding to the second multi-dimensional user feature data by using the K-means++ clustering algorithm in the customer intelligent retention module;

[0065] S1403: Input the target customer group identifier and the second multi-dimensional user feature data into the LightGBM classification model in the customer intelligent retention module to perform customer stratification processing to obtain the target customer level.

[0066] This embodiment performs customer stratification through a dual-model collaborative grading architecture including a K-means++ clustering algorithm (an improved version of the K-means clustering algorithm) and a Light Gradient Boosting Machine (LightGBM) classification model, which can improve the accuracy of customer stratification.

[0067] S150: Determine a target compensation strategy corresponding to the target customer level according to a preset differentiated compensation strategy.

[0068] In some embodiments, specifically, if the target customer level is the high-value level, the preset first compensation strategy is determined as the target compensation strategy, and the first compensation strategy includes an automatic equity recovery strategy and an exclusive customer service strategy; if the target customer level is the potential level, the preset second compensation strategy is determined as the target compensation strategy, and the second compensation strategy includes a step-by-step compensation strategy; if the target customer level is the ordinary level, the preset third compensation strategy is determined as the target compensation strategy, and the third compensation strategy includes a general points issuance strategy or a general category voucher issuance strategy; if the target customer level is the risk level, the preset fourth compensation strategy is determined as the target compensation strategy, and the fourth compensation strategy includes a rule description strategy and a risk control list addition strategy.

[0069] S160: Perform equity compensation processing on the target customer according to the target compensation strategy.

[0070] In this embodiment, after the target compensation strategy is determined, the target customer will be compensated for the equity through the target compensation strategy. It should be noted that for risk-layer customers, after the equity is recovered, only the rules will be explained to the customer, for example, why the customer's equity should be recovered, and the customer will be added to the risk control list, and the customer's customer identification will be sent to the risk intelligent identification module to facilitate the risk intelligent identification module to conduct subsequent risk monitoring of the customer.

[0071] Furthermore, after executing the equity compensation processing for the target customer according to the target compensation strategy, the method also includes: obtaining the behavioral feedback indicators of the target customer through preset embedding points, the behavioral feedback indicators including the compensation utilization rate, the length of the second consumption interval and the change in the average order value; adjusting the customer stratification algorithm and the differentiated compensation strategy according to the behavioral feedback indicators.

[0072] It can be seen that the embodiments of the present application can realize the marketing activity management system and the automatic upgrade and optimization of the customer intelligent retention module.

[0073] The following is a detailed description of the customer intelligent retention module provided by this application. The customer intelligent retention module provided in this embodiment constructs a closed-loop system for customer value grading and dynamic compensation. It integrates static attributes and dynamic behaviors through multi-dimensional real-time data analysis of customer characteristics, and adopts an advanced dual-model collaborative architecture to achieve accurate customer grading. This dual-model collaborative architecture combines the advantages of K-means++ clustering and LightGBM classification models to form a method that combines unsupervised and supervised learning. In the K-means++ clustering stage, the system optimizes the initial center point selection strategy of the traditional K-means algorithm, adopts a probability weighted method to select the initial cluster center, and significantly improves the clustering stability.

[0074] Among them, the Silhouette coefficient is introduced in the clustering process to automatically evaluate the optimal number of clusters, and the Principal Component Analysis (PCA) dimensionality reduction technology is used to project high-dimensional features into two-dimensional space for visualization. The clustering model mainly processes the characteristics of customer behavior patterns, discovers potential customer similarities, and provides structured feature enhancement for the classification model. In the LightGBM classification stage, historical labeled data is used to train the supervised learning model. LightGBM uses a histogram-based decision tree learning algorithm, which has the advantages of fast training speed and low memory usage. The model input includes static attributes, dynamic behavior, and customer group identifiers generated by clustering, and outputs customer value scores and hierarchical probability distributions.

[0075] The two models achieve synergy through feature fusion, result validation, and feedback optimization. A time decay function is introduced to dynamically adjust the weight of historical behaviors, ensuring focus on recent customer behavior. The system also features a flexible interpretability layer to translate model decisions into a language understandable to the business.

[0076] Through the above-mentioned architecture of the customer intelligent retention module, the system realizes the precise four-tier classification of customers (high-value layer, potential layer, ordinary layer, and risk layer), and provides rights recovery and exclusive services for high-value customers based on hierarchical intelligent matching. Potential customers are issued tiered incentive coupons, ordinary customers are given universal points, and risky customers are only provided with rule descriptions and risk control tags.

[0077] In addition, the intelligent customer retention module also uses full-link tracking technology to capture customer feedback signals (compensation usage rate, second consumption interval length, change in average order value, etc.) in real time, and combines it with the Deep Q Network (DQN) to build an adaptive optimization engine. It uses customer retention rate and consumption growth as the core reward indicators of the DQN network, and continuously adjusts the grading weights and compensation strategies. At the same time, it introduces an online learning mechanism to prioritize high-value customer misjudgment samples, forming a self-evolutionary closed loop of "grading-compensation-feedback-tuning", achieving the optimal customer compensation strategy and model self-iteration, and ultimately achieving the dual goals of improving high-value customer retention and reducing risk control costs.

[0078] The following is a further detailed description of the customer intelligent retention module provided by this application:

[0079] First, a multi-dimensional stratified indicator system was constructed, encompassing 57 core features across four categories: value features (total spending over the past 90 / 180 / 365 days, average order value, purchase frequency, etc.), behavioral features (last visit time, page dwell time, interaction frequency, etc.), loyalty features (membership level, account age, repurchase rate, etc.), and preference features (category preference, marketing campaign response rate, etc.). All features were normalized using Min-Max normalization, and feature importance weighting based on the Pearson correlation coefficient was applied to ensure stratification accuracy. For the stratification calculation, a dual-model collaborative architecture was designed: an optimized K-means++ algorithm was used to perform preliminary clustering using an improved initial center point selection strategy. The system then applied the silhouette coefficient to automatically determine the optimal number of clusters, 12-15 segments. Then, based on these preliminary clustering results and labeled samples, a LightGBM classification model (with 57 features, a tree depth of 8, a learning rate of 0.03, and 32 leaves) was trained to generate a continuous value score of 0-100 for each customer. A formula for defining customer value based on time decay was established: , each coefficient is determined by Bayesian optimization, and the weight of recent behavior decays exponentially over time (λ=0.05 / day).

[0080] Based on the calculation results, the customer intelligent retention module can accurately divide customers into four levels: 1) high-value level (value score 85-100, accounting for approximately 2-5%); 2) potential level (score 65-84, accounting for approximately 10-15%); 3) ordinary level (score 30-64, accounting for approximately 60-70%); 4) risk level (score 0-29, accounting for approximately 15-20%). The Intelligent Customer Retention Module implements a differentiated compensation strategy engine for different customer tiers. High-value customers receive a VIP compensation plan, including automatic benefit restoration (original benefit limit + 10% additional compensation), proactive communication with a dedicated customer service representative within one hour (via an NLP-driven intelligent speech system), and customized exclusive benefit packages. Potential customers trigger a tiered incentive program, awarding three levels of progressive compensation coupons (base limit × {1.0, 1.3, 1.5}) based on their spending potential, with a 24-hour expiration date to encourage repeat purchases. Standard compensation is provided to ordinary customers, including basic points and general category coupons. Risky customers receive only a guide to the rules and are flagged for risk management to prevent abuse of benefits. Furthermore, the Intelligent Customer Retention Module includes an intelligent compensation decision engine based on a Deep Q-Network (DQN). This engine constructs a three-dimensional state space (customer tier, historical compensation response rate, and recent activity), designs seven compensation actions (adjusting the limit ± {0%, 10%, 20%} and changing the benefit type), and uses the customer's seven-day consumption growth rate as the primary reward signal (r = α × consumption growth - β × compensation cost). Through a dual-network architecture (target network + evaluation network) and an experience replay buffer (capacity of 100,000 records, batch size of 64), the customer intelligent retention module continuously optimizes compensation strategies. Compensation execution is triggered in milliseconds through a distributed workflow engine, and full-link tracking technology tracks customer behavior after compensation, forming a self-evolving closed loop of "grading-compensation-feedback-optimization."

[0081] To sum up, on the one hand, when this solution monitors the existence of refund information in equity orders, it will conduct risk detection on the corresponding customers. When the customer is detected as a risky customer, risk recovery will be carried out, reducing the interval time between the occurrence of abnormal behavior and the actual equity recovery, and avoiding the abuse of equity; on the other hand, this solution intelligently stratifies risky customers and sets differentiated compensation strategies for customers at different levels. Different compensation strategies are implemented for customers at different levels. Compared with the one-size-fits-all compensation measures, this solution can improve customer retention rate.

[0082] Figure 5 Schematic block diagram of a marketing activity management device 500 provided in an embodiment of the present application. Figure 5As shown, corresponding to the above marketing activity management method, the present application also provides a marketing activity management device 500. The marketing activity management device 500 includes a unit for executing the above marketing activity management method, and the marketing activity management device 500 can be configured in a terminal or a server. Specifically, the marketing activity management device 500 is deployed in a marketing activity management system, and the marketing activity management system maintains a stake order pool, which includes stake orders associated with multiple customer identifiers, and the stake orders are associated with issued stakes. Figure 5 The marketing activity management device 500 includes a transceiver unit 501 and a processing unit 502, wherein:

[0083] The transceiver unit 501 is configured to obtain refund information of a target equity order in the equity order pool;

[0084] Processing unit 502 is used to update the target order actual consumption index of the target equity order according to the refund information when obtaining the refund information of the target equity order in the equity order pool, and update the target customer actual consumption index corresponding to the target customer identifier according to the updated target order actual consumption index, where the target customer identifier is the customer identifier corresponding to the target equity order; determine whether the target customer corresponding to the target customer identifier is a risk customer based on the target customer actual consumption index, the target equity issuance rule, the first multi-dimensional user feature data corresponding to the target customer identifier, and a preset AI anomaly monitoring model, where the target equity issuance rule is the equity issuance rule associated with the issued equity corresponding to the target equity order; if the target customer is a risk customer, recover the issued equity of the target customer; determine the target customer level corresponding to the target customer based on the second multi-dimensional user feature data corresponding to the target customer identifier and a preset customer stratification algorithm; determine the target compensation strategy corresponding to the target customer level based on the preset differentiated compensation strategy; and perform equity compensation processing on the target customer based on the target compensation strategy.

[0085] In some embodiments, when executing the step of determining whether the target customer corresponding to the target customer identifier is a risky customer based on the target customer's actual consumption indicator, the target benefit issuance rule, the first multi-dimensional user feature data corresponding to the target customer identifier, and the preset AI anomaly monitoring model, the processing unit 502 is specifically configured to:

[0086] Determine whether the target customer meets the target benefit issuance rules based on the target customer's actual consumption indicators; if the target customer does not meet the target benefit issuance rules, obtain the first multi-dimensional user feature data; perform a risk score on the target customer based on the AI ​​anomaly monitoring model and the first multi-dimensional user feature data to obtain a target risk score; if the target risk score is greater than a preset risk score threshold, determine that the target customer is the risky customer; if the target customer meets the target benefit issuance rules, or the target risk score is less than or equal to the risk score threshold, determine that the target customer is an ordinary abnormal customer.

[0087] In some embodiments, the marketing activity management system includes a consumption index monitoring module, a risk intelligent identification module, and a customer intelligent retention module; after executing the step of updating the target customer actual consumption index corresponding to the target customer identifier based on the updated target order actual consumption index, the processing unit 502 is further configured to:

[0088] The actual consumption index of the target order and the actual consumption index of the target customer are written into the ClickHouse table of the equity order pool through the consumption index monitoring module, and the consumption index update information is sent to the risk intelligent identification model. The consumption index update information carries the target customer identifier. The actual consumption index of each equity order, the order attribute information and the actual consumption index of each customer identifier are stored in the ClickHouse table;

[0089] At this time, when the processing unit 502 executes the step of determining whether the target customer meets the target benefit issuance rule based on the target customer's actual consumption index, it is specifically configured to:

[0090] The risk intelligent identification module responds to the consumption indicator update information, determines the target rights issuance rule corresponding to the target customer identifier, performs semantic analysis on the target rights issuance rule, and generates a target SQL statement; utilizes the transceiver unit 501, obtains the target customer's actual consumption indicator and the target customer-related indicator from the ClickHouse table according to the target SQL statement; judges whether the target customer meets the target rights issuance rule based on the target customer's actual consumption indicator and the target customer-related indicator.

[0091] In some embodiments, when executing the step of determining the target customer level corresponding to the target customer based on the second multi-dimensional user feature data corresponding to the target customer identifier and a preset customer stratification algorithm, the processing unit 502 is specifically configured to:

[0092] Utilizing the transceiver unit 501, the second multi-dimensional user feature data is obtained through the feature input interface of the customer intelligent retention module; the target customer group identifier corresponding to the second multi-dimensional user feature data is determined through the K-means++ clustering algorithm in the customer intelligent retention module; the target customer group identifier and the second multi-dimensional user feature data are input into the LightGBM classification model in the customer intelligent retention module for customer stratification processing to obtain the target customer level.

[0093] In some embodiments, the target customer level is a high-value level, a potential level customer, a general level customer, or a risk level customer; when the processing unit 502 performs the step of determining the target compensation strategy corresponding to the target customer level according to the preset differentiated compensation strategy, it is specifically configured to:

[0094] If the target customer level is the high-value level, the preset first compensation strategy is determined as the target compensation strategy, and the first compensation strategy includes an automatic equity recovery strategy and an exclusive customer service strategy;

[0095] If the target customer level is the potential level, the preset second compensation strategy is determined as the target compensation strategy, and the second compensation strategy includes a stepped compensation strategy;

[0096] If the target customer level is the general level, the preset third compensation strategy is determined as the target compensation strategy, and the third compensation strategy includes a general points issuance strategy or a general category voucher issuance strategy;

[0097] If the target customer level is the risk level, the preset fourth compensation strategy is determined as the target compensation strategy, and the fourth compensation strategy includes a rule description strategy and a risk control list addition strategy.

[0098] In some embodiments, after executing the step of performing equity compensation processing on the target customer according to the target compensation strategy, the processing unit 502 is further configured to:

[0099] The target customer's behavioral feedback indicators are obtained through preset tracking points. The behavioral feedback indicators include compensation usage rate, second consumption interval length, and customer unit price change. The customer stratification algorithm and the differentiated compensation strategy are adjusted according to the behavioral feedback indicators and the DQN network.

[0100] In some embodiments, the first multi-dimensional user feature data includes transaction behavior features, transaction timing features, user portrait features, and associated network features; the second multi-dimensional user feature data includes value dimension features, behavior dimension features, loyalty dimension features, and preference dimension features.

[0101] To sum up, on the one hand, the marketing activity management device 500 provided by the present solution performs risk detection on the corresponding customer when it monitors the existence of refund information in the equity order. When the customer is detected as a risky customer, risk recovery is carried out, thereby reducing the interval time between the occurrence of abnormal behavior and the actual equity recovery, and avoiding the abuse of equity. On the other hand, the present solution provides a marketing activity management device 500 to intelligently stratify risky customers, and set differentiated compensation strategies for different levels of customers, and implement different compensation strategies for different levels of customers. Compared with the one-size-fits-all compensation measures, this solution can improve customer retention rate.

[0102] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned marketing activity management device and each unit can refer to the corresponding description in the aforementioned method embodiment. For the convenience and brevity of description, it will not be repeated here.

[0103] The marketing activity management device can be implemented in the form of a computer program. Figure 6 Runs on the computer equipment shown.

[0104] See also Figure 6 , Figure 6 This is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 600 can be a terminal or a server, and the marketing activity management system provided in the present application is composed of the computer device. The marketing activity management system maintains a pool of equity orders, which includes equity orders associated with multiple customer identifiers, each of which is associated with issued equity.

[0105] See Figure 6 The computer device 600 includes a processor 602 , a memory, and a network interface 605 connected via a system bus 601 , wherein the memory may include a non-volatile storage medium 603 and an internal memory 604 .

[0106] The non-volatile storage medium 603 can store an operating system 6031 and a computer program 6032. The computer program 6032 includes program instructions, which, when executed, can cause the processor 602 to execute a marketing campaign management method.

[0107] The processor 602 is used to provide computing and control capabilities to support the operation of the entire computer device 600.

[0108] The internal memory 604 provides an environment for the operation of the computer program 6032 in the non-volatile storage medium 603 . When the computer program 6032 is executed by the processor 602 , the processor 602 can execute a marketing activity management method.

[0109] The network interface 605 is used to communicate with other devices over the network. Figure 6 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 600 to which the solution of the present application is applied. The specific computer device 600 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0110] The processor 602 is configured to execute a computer program 6032 stored in the memory to implement the following steps:

[0111] When refund information of a target equity order in the equity order pool is obtained, the target order actual consumption index of the target equity order is updated according to the refund information, and the target customer actual consumption index corresponding to the target customer identifier is updated according to the updated target order actual consumption index, where the target customer identifier is the customer identifier corresponding to the target equity order;

[0112] Determining whether the target customer corresponding to the target customer identifier is a risky customer based on the target customer's actual consumption index, the target equity issuance rule, the first multi-dimensional user feature data corresponding to the target customer identifier, and a preset AI anomaly monitoring model, wherein the target equity issuance rule is the equity issuance rule associated with the issued equity corresponding to the target equity order;

[0113] If the target customer is a risky customer, the issued rights and interests of the target customer will be recovered;

[0114] Determining a target customer level corresponding to the target customer based on the second multi-dimensional user feature data corresponding to the target customer identifier and a preset customer stratification algorithm;

[0115] Determining a target compensation strategy corresponding to the target customer level according to a preset differentiated compensation strategy;

[0116] Execute equity compensation processing for the target customer according to the target compensation strategy.

[0117] It should be understood that in the embodiment of the present application, the processor 602 may be a central processing unit (CPU), and the processor 602 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0118] Those skilled in the art will appreciate that all or part of the steps in the method of the above-described embodiment can be implemented by instructing the relevant hardware through a computer program. The computer program includes program instructions, which can be stored in a storage medium that is computer-readable. The program instructions are executed by at least one processor in the computer system to implement the steps in the method of the above-described embodiment.

[0119] Therefore, the present application also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein the computer program includes program instructions. When the program instructions are executed by a processor, the processor performs the following steps:

[0120] When refund information of a target equity order in the equity order pool is obtained, the target order actual consumption index of the target equity order is updated according to the refund information, and the target customer actual consumption index corresponding to the target customer identifier is updated according to the updated target order actual consumption index, where the target customer identifier is the customer identifier corresponding to the target equity order;

[0121] Determining whether the target customer corresponding to the target customer identifier is a risky customer based on the target customer's actual consumption index, the target equity issuance rule, the first multi-dimensional user feature data corresponding to the target customer identifier, and a preset AI anomaly monitoring model, wherein the target equity issuance rule is the equity issuance rule associated with the issued equity corresponding to the target equity order;

[0122] If the target customer is a risky customer, the issued rights and interests of the target customer will be recovered;

[0123] Determining a target customer level corresponding to the target customer based on the second multi-dimensional user feature data corresponding to the target customer identifier and a preset customer stratification algorithm;

[0124] Determining a target compensation strategy corresponding to the target customer level according to a preset differentiated compensation strategy;

[0125] Execute equity compensation processing for the target customer according to the target compensation strategy.

[0126] The storage medium may be any computer-readable storage medium that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk.

[0127] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0128] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and other division methods may be used in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not implemented.

[0129] The steps in the method of the embodiment of the present application can be adjusted in order, combined, and deleted according to actual needs. The units in the device of the embodiment of the present application can be combined, divided, and deleted according to actual needs. In addition, the functional units in the various embodiments of the present application can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into a single unit.

[0130] If this integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of this application, or the part that contributes to the existing technology, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, terminal, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of this application.

[0131] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present application, and such modifications or substitutions should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A marketing activity management method, characterized in that: The method is applied to a marketing activity management system, wherein a rights order pool is maintained in the marketing activity management system, wherein the rights order pool includes rights orders respectively associated with a plurality of customer identifiers, wherein the rights orders are associated with issued rights, and the method comprises: When refund information of a target equity order in the equity order pool is obtained, the target order actual consumption index of the target equity order is updated according to the refund information, and the target customer actual consumption index corresponding to the target customer identifier is updated according to the updated target order actual consumption index, where the target customer identifier is the customer identifier corresponding to the target equity order; Determining whether a target customer corresponding to the target customer identifier is a risky customer based on the target customer's actual consumption index, the target equity issuance rule, the first multi-dimensional user feature data corresponding to the target customer identifier, and a preset AI anomaly monitoring model, wherein the target equity issuance rule is the equity issuance rule associated with the issued equity corresponding to the target equity order, and the AI ​​anomaly monitoring model is constructed based on the XGBoost algorithm; If the target customer is a risky customer, the issued rights and interests of the target customer will be recovered; Determining a target customer level corresponding to the target customer based on the second multi-dimensional user feature data corresponding to the target customer identifier and a preset customer stratification algorithm; Determining a target compensation strategy corresponding to the target customer level according to a preset differentiated compensation strategy; Performing equity compensation processing on the target customer according to the target compensation strategy; The determining whether the target customer corresponding to the target customer identifier is a risky customer based on the target customer's actual consumption index, the target benefit issuance rule, the first multi-dimensional user feature data corresponding to the target customer identifier, and the preset AI anomaly monitoring model includes: Determine whether the target customer meets the target benefit issuance rules based on the target customer's actual consumption indicators; If the target customer does not meet the target benefit issuance rules, obtaining the first multi-dimensional user feature data; Performing a risk score on the target customer based on the AI ​​anomaly monitoring model and the first multi-dimensional user feature data to obtain a target risk score; If the target risk score is greater than a preset risk score threshold, the target customer is determined to be the risky customer; If the target customer meets the target benefit issuance rule, or the target risk score is less than or equal to the risk score threshold, the target customer is determined to be a common abnormal customer.

2. The method according to claim 1, characterized in that The marketing activity management system includes a consumption index monitoring module, a risk intelligent identification module, and a customer intelligent retention module; after updating the target customer actual consumption index corresponding to the target customer identifier according to the updated target order actual consumption index, the method further includes: The actual consumption index of the target order and the actual consumption index of the target customer are written into the ClickHouse table of the equity order pool through the consumption index monitoring module, and the consumption index update information is sent to the risk intelligent identification model. The consumption index update information carries the target customer identifier. The actual consumption index of each equity order, the order attribute information and the actual consumption index of each customer identifier are stored in the ClickHouse table; The determining whether the target customer meets the target benefit issuance rules based on the target customer's actual consumption index includes: In response to the consumption indicator update information, the intelligent risk identification module determines the target benefit issuance rule corresponding to the target customer identifier, performs semantic analysis on the target benefit issuance rule, and generates a target SQL statement; Obtain the target customer actual consumption indicator and the target customer association indicator from the ClickHouse table according to the target SQL statement; It is determined whether the target customer meets the target benefit issuance rules according to the target customer's actual consumption index and the target customer association index.

3. The method according to claim 2, characterized in that The determining the target customer level corresponding to the target customer based on the second multi-dimensional user feature data corresponding to the target customer identifier and a preset customer stratification algorithm includes: Acquiring the second multi-dimensional user feature data through the feature input interface of the customer intelligent retention module; Determining the target customer group identifier corresponding to the second multi-dimensional user feature data by using the K-means++ clustering algorithm in the customer intelligent retention module; The target customer group identifier and the second multi-dimensional user feature data are input into the LightGBM classification model in the customer intelligent retention module to perform customer stratification processing to obtain the target customer level.

4. The method according to claim 1, wherein The target customer level is a high-value customer, a potential customer, an ordinary customer, or a risk customer; and determining the target compensation strategy corresponding to the target customer level according to the preset differentiated compensation strategy includes: If the target customer level is the high-value level, the preset first compensation strategy is determined as the target compensation strategy, and the first compensation strategy includes an automatic equity recovery strategy and an exclusive customer service strategy; If the target customer level is the potential level, the preset second compensation strategy is determined as the target compensation strategy, and the second compensation strategy includes a stepped compensation strategy; If the target customer level is the general level, the preset third compensation strategy is determined as the target compensation strategy, and the third compensation strategy includes a general points issuance strategy or a general category voucher issuance strategy; If the target customer level is the risk level, the preset fourth compensation strategy is determined as the target compensation strategy, and the fourth compensation strategy includes a rule description strategy and a risk control list addition strategy.

5. The method according to claim 1, wherein After performing equity compensation processing on the target customer according to the target compensation strategy, the method further includes: Obtaining behavioral feedback indicators of the target customers through preset tracking points, wherein the behavioral feedback indicators include compensation usage rate, second consumption interval length, and change in average order value; The customer stratification algorithm and the differentiated compensation strategy are adjusted according to the behavioral feedback indicator and the DQN network.

6. The method according to any one of claims 1 to 5, characterized in that The first multi-dimensional user feature data includes transaction behavior features, transaction timing features, user portrait features, and associated network features; the second multi-dimensional user feature data includes value dimension features, behavior dimension features, loyalty dimension features, and preference dimension features.

7. A marketing activity management device, characterized in that: The marketing activity management device is deployed in a marketing activity management system. The marketing activity management system maintains a rights order pool. The rights order pool includes rights orders associated with multiple customer identifiers. The rights orders are associated with issued rights. The marketing activity management device includes: A transceiver unit, configured to obtain refund information of a target equity order in the equity order pool; a processing unit configured to, when obtaining refund information of a target equity order in the equity order pool, update a target order actual consumption index of the target equity order according to the refund information, and update a target customer actual consumption index corresponding to a target customer identifier according to the updated target order actual consumption index, wherein the target customer identifier is a customer identifier corresponding to the target equity order; determine whether the target customer corresponding to the target customer identifier is a risk customer according to the target customer actual consumption index, a target equity issuance rule, first multi-dimensional user feature data corresponding to the target customer identifier, and a preset AI anomaly monitoring model, wherein the target equity issuance rule is an equity issuance rule associated with the issued equity corresponding to the target equity order, and the AI ​​anomaly monitoring model is constructed based on an XGBoost algorithm; if the target customer is a risk customer, reclaim the issued equity of the target customer; determine a target customer level corresponding to the target customer according to the second multi-dimensional user feature data corresponding to the target customer identifier and a preset customer stratification algorithm; determine a target compensation strategy corresponding to the target customer level according to a preset differentiated compensation strategy; and perform equity compensation processing on the target customer according to the target compensation strategy; When executing the step of determining whether the target customer corresponding to the target customer identifier is a risky customer based on the target customer's actual consumption index, the target benefit issuance rule, the first multi-dimensional user feature data corresponding to the target customer identifier, and the preset AI anomaly monitoring model, the processing unit is specifically configured to: Determine whether the target customer meets the target benefit issuance rules based on the target customer's actual consumption indicators; if the target customer does not meet the target benefit issuance rules, obtain the first multi-dimensional user feature data; perform a risk score on the target customer based on the AI ​​anomaly monitoring model and the first multi-dimensional user feature data to obtain a target risk score; if the target risk score is greater than a preset risk score threshold, determine that the target customer is the risky customer; if the target customer meets the target benefit issuance rules, or the target risk score is less than or equal to the risk score threshold, determine that the target customer is an ordinary abnormal customer.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the marketing activity management method according to any one of claims 1 to 6 is implemented.

9. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor executes the marketing activity management method according to any one of claims 1 to 6.

Images