Learning customer preference dynamics using rank-aware matrix factorization and enhanced collaborative filtering model

Abstract

Understanding how customer preferences evolve over time is a critical challenge for modern recommender systems operating in large-scale, implicit-feedback–driven e-commerce environments. The primary objective of this study is to develop a unified and interpretable framework that simultaneously models ranking-based preferences, collaborative similarity structures, and temporal behavioral evolution of customers. To achieve this, the study proposes a novel hybrid framework that integrates rank-aware matrix factorization (RA-MF), enhanced collaborative filtering (CF), K-means clustering, and temporal cluster migration matrices (TCMM) for learning customer preference dynamics. The ranking factorization model effectively captures implicit signals such as purchase frequency and recency decay, while CF provides complementary similarity-based insights. K-means segmentation reveals diverse customer personas, including high-value loyal buyers and exploratory shoppers, with significant differences in spending and purchasing behavior. Quantitative evaluations demonstrate strong performance improvements, with 11–18% gains in NDCG@10, 10–15% increases in Precision@10, and notable reductions in root mean square error (RMSE) and mean absolute error (MAE). The results highlight the framework’s ability to deliver both accurate recommendations and interpretable behavioral insights, offering valuable contributions to personalized marketing, customer retention, and data-driven e-commerce strategy.