面向电商复杂用户行为序列的多视角协同学习框架研究

Research on Multi-Perspective Collaborative Learning Framework for Complex E-Commerce User Behavior Sequences

摘要: 针对电商用户行为序列建模中特征提取不全、时序与频率特征融合不足的问题，本文提出多视角协同学习框架TPA-Net。该框架通过多尺度特征提取、时空频三路径并行编码与跨路径注意力融合，实现复杂行为序列的高维表征。在阿里巴巴用户行为数据集的6分类任务中，TPA-Net准确率达92.17%，Macro-F1达91.12%，较双路径CNN-LSTM显著提升。模型精度提升的同时计算成本有所增加，可为电商用户价值细分、流失预警与精准营销提供技术支撑，也为复杂时序预测提供参考。

Abstract: To address the problems of incomplete feature extraction and insufficient fusion of temporal and frequency features in modeling e-commerce user behavior sequences, this paper proposes a multi-perspective collaborative learning framework, TPA-Net. The framework realizes high-dimensional representation of complex behavior sequences through multi-scale feature extraction, temporal-spatial-frequency three-path parallel encoding, and cross-path attention fusion. In the 6-class classification task on the Alibaba user behavior dataset, TPA-Net achieves 92.17% accuracy and 91.12% Macro-F1, with significant improvements over the dual-path CNN-LSTM. While improving accuracy, the model incurs a higher computational cost. It can provide technical support for user value segmentation, churn warning, and precision marketing in e-commerce, and also serve as a reference for complex time-series prediction.