A Comparative Study of E-commerce Purchase Intention Prediction Models Based on User Behavior Sequences 
Ye Chengkai

College of Computer and Data Science, Fuzhou University, Fuzhou Fujian, 350108;

Abstract: Accurate prediction of purchase intention from user behavior sequences is central to personalized recommendation and targeted marketing in e-commerce. This study presents a comparative evaluation of representative models on clickstream data to identify the most effective approach for capturing sequential purchase signals. Compare traditional classifiers, specifically Logistic Regression and Random Forest, with deep sequence models, namely LSTM, GRU, and a Transformer-based architecture. To ensure a fair comparison regarding product awareness, traditional models are augmented with aggregated item ID embeddings, while temporal order is still ignored. Each action is represented by the concatenation of item and category embeddings, behavior type encoding, and log-transformed time intervals; the Transformer further adopts a narrow configuration to match the GRU's parameter count. Experiments on a large-scale real-world dataset show that sequence-aware deep models consistently outperform conventional methods. The Transformer achieves the highest overall performance, demonstrating superior ability to model long-range dependencies, while the GRU offers a strong balance between accuracy and computational cost. The paper also discusses the effect of behavior sequence length and the trade-offs in model complexity, providing practical guidance for industrial deployment.

Keywords:Purchase intention prediction; User behavior sequences; Deep learning; Model comparison; E-commerce clickstream

References

[1]Tsou T H ,Chen S J ,Hsu Y H , et al.OTT recommender system features shape consumers’ continued usage intention: An integrated perspective based on the SOR model and theory of consumption values[J].Service Business,2026,20(2):13-13.DOI:10.1007/S11628-026-00613-W.

[2]Hu Y ,Niu Y . Application of clustering search combined with federated learning in secure advertising delivery [J]. Engineering Research Express, 2026, 8 (9): 095223-095223. DOI:10.1088/2631-8695/AE6234.

[3]Xie Y ,Chen Y ,Zeng W , et al. Fusing multi-head gating and focused-view contrastive learning for knowledge-aware recommendation [J]. Knowledge-Based Systems, 2026, 341 115673-115673. DOI:10.1016/J.KNOSYS.2026.115673.

[4]Huang Y ,Rojniruttikul N . Structural equation model of E-commerce live broadcasting influencing customers purchase intention prediction using machine learning [J]. International Journal of Information Technology, 2026, (prepublish): 1-12. DOI:10.1007/S41870-025-02851-Z.

[5]Xiaoye M ,Yanyan L ,Muhammad A . E-Commerce Review Sentiment Analysis and Purchase Intention Prediction Based on Deep Learning Technology [J]. Journal of Organizational and End User Computing (JOEUC), 2023, 36 (1): 1-29. DOI:10.4018/JOEUC.335122.