Personalized Red Envelope Marketing Using Data Mining and Logistic Regression Models

With the rapid growth of online consumers, companies are increasingly adopting fine‑grained operations based on user groups and scenarios to maximize resource utilization. The article introduces a personalized red‑envelope marketing approach that leverages data mining techniques—including data cleaning, feature selection, and regression modeling—to achieve cross‑project user acquisition.

Problems Identified

Lack of unified data standards across projects.

Inconsistent information flow and no linkage between push points.

Uniform red‑envelope amounts that ignore individual consumption preferences.

Need to increase new‑customer volume while controlling red‑envelope costs.

Requirement for a universal model that can be applied across multiple projects.

Solution Overview

The proposed solution consists of four modules: data, algorithm model, system, and operations. The system and operations rely on an existing membership marketing platform, while the article focuses on the data and algorithm components.

Data Preparation

Raw data are cleaned to remove dirty and fraudulent (e.g., scalper) records. Features are extracted to capture strong representational power, divided into common user attributes and project‑specific attributes, and then fused to create a unified feature set suitable for all cross‑project scenarios.

Algorithm Model

Two sub‑models are built: a consumption‑demand prediction model and a red‑envelope‑sensitivity model. Both models use Pearson correlation coefficients for feature extraction and employ logistic regression for training. The logistic regression formulation is shown in the accompanying equations.

Model Training and Validation

Historical data are split into training and validation sets. Different parameter configurations are evaluated on the validation set using accuracy and recall, and the best configuration is retrained on the full dataset to obtain the final predictive model.

Scoring and Cost Optimization

Using the two models, each user receives a score that is mapped to a personalized red‑envelope amount. The effectiveness metric combines cross‑project new‑customer conversion rate and red‑envelope cost, aiming for minimal cost and maximal conversion. A unified cost indicator is defined to guide optimization.

Iterative Optimization

Model parameters are updated daily based on performance monitoring.

User segmentation is refined by adjusting the score‑to‑amount mapping and introducing more red‑envelope types.

Results

AB‑tests demonstrate that the personalized model can increase conversion rates (e.g., 8% uplift for airline‑ticket users receiving hotel red‑envelopes) while reducing red‑envelope costs (e.g., 8% cost reduction for hotel‑ticket users receiving airline red‑envelopes), confirming the value of big‑data‑driven personalization.