Behavioral Segmentation in Digital Payment Platforms Using Temporal-Adaptive Attentive Fuzzy Clustering

Digital payment platforms record a wealth of transaction data, capturing intricate and dynamic patterns of user behavior. Clustering users according to their behavioral patterns presents unique challenges due to overlapping behavior sets, dynamicity, and unlabeled data. In this paper, we propose a temporal-adaptive attentive fuzzy clustering method for behavioral user segmentation in digital payment platforms. We construct a behavioral representation for each user based on transaction records using a temporal decay function to emphasize recent behaviors. An attention mechanism is incorporated into the fuzzy clustering objective to adaptively assess the importance of each behavior feature on the clustering results. We employ a soft assignment approach that allows each user to have varying degrees of membership across multiple clusters to capture the gradual transition of payment behaviors. We evaluate our approach on a publicly available digital wallet transaction dataset. Our results show that the proposed clustering method achieves better clustering quality and temporal consistency compared to traditional clustering algorithms. The experimental results show that incorporating temporal adaptivity and attention provides a promising and effective basis for behavioral clustering in digital payment platforms.