Understanding customer behaviour is crucial for business success. For achieving this goal, the Recency-Frequency-Monetary (RFM) model has been commonly recognised as an effective approach to analyse customer behaviour. However, the traditional RFM approach is a coarse method for quantifying customer loyalty and contribution that can only provide a single lump-sum value of the recency (R), frequency (F), and monetary value (M); hence, it discards information regarding customers' product preferences. Typically, different customers make different purchases. Subsequently, purchases are likely to be different across customers. This creates data sparsity, which affects the performance of conventional clustering methods. In this study, we integrated the group RFM analysis and probabilistic latent semantic analysis models to perform customer segmentation and customer analysis. The results indicated that the developed approach takes into account the product preference and provides insight into and captures a wide ABOUT THE AUTHORS Arthit Apichottanakul is currently working as lecturer in the Faculty of Technology, Khon Kaen University, Thailand. He completed his PhD in Industrial Engineering from Khon Kaen University, Thailand. His current research interests include intelligent applications, optimization and data science in logistics and supply chain management.