F-CAG: Fuzzy Clustering for Attributed Graphs

Abstract

Fuzzy clustering and perception are two fundamental concepts in the field of ML that complement each other to enhance the understanding and processing of complex data. In this paper, we focus on attributed networks and introduce a novel framework for clustering networked data using joint embedding and clustering. Our approach, based on entropy-based regularization in the fuzzy clustering criterion, employs low-rank subspaces and fuzzy clusters to better capture complex relationships between content and structure information, enhancing clustering robustness. Experiments on various benchmark datasets for document clustering, using both bag-of-words and large language model (LLM) representations, demonstrate that our algorithm surpasses state-of-the-art clustering methods, including task-specific deep learning approaches.