LAUSR

This study introduces a pipeline to categorize a rich set financial clients (or investors) into multiple financial communities (we call them circles), by leveraging a regularized topic model to describe each client’s financial behavior. Such discovered financial circles can be applied in many tasks. For example, different financial circles can reveal the different risk-tolerant levels or investment tendency. We can recommend different investment plans to different financial circles. We propose a geometry-based feature selector to extract high-quality features that represent the transactional nature of each record, followed by a probabilistic model that maps transaction attributes as distributions within a latent space. A graph-based approach is then employed to capture transactional similarities, using an advanced algorithm to identify densely connected subgraphs and cluster transactions into circles based on shared behaviors. Experimental results on diverse financial datasets demonstrate the robustness of this method in accurately distinguishing and characterizing distinct financial behavior types, providing a valuable tool for enhancing the analysis of transactional strategies and patterns.