GFS-Edge: Graph Fuzzy System for Edge Prediction

Edge prediction (EP) is a long-standing and fundamental task in graph-structured data analysis, and has served as a key driver in the advancement of graph learning. Traditional EP methods can be broadly classified into three major categories: heuristic approaches, embedding-based methods, and graph neural network (GNN)-based techniques. While each category has achieved notable performance improvements, these methods often fall short in handling uncertainty and generally lack robust interpretability. These limitations highlight an urgent need for novel approaches that can effectively address uncertain EP tasks while offering strong interpretability and rich feature representations. To tackle these challenges, we propose a novel graph fuzzy system (GFS) for edge prediction, termed GFS-edge, and systematically introduces its core concepts, model architecture, and construction methodology. Specifically, we first define key concepts of GFS-edge, including rule bases, fuzzy sets, and edge consequent processing module (ECPM). We then present a general modeling framework for GFS-edge, detailing the formulation of rule antecedents and consequents. Subsequently, a learning framework is introduced, in which edge-based clustering is employed for antecedent generation, while a linear variational graph auto-encoder (LVGAE) is adopted as the ECPM to facilitate consequent learning. Extensive experiments are conducted on 12 benchmark datasets to evaluate the performance of GFS-edge. The results consistently demonstrate that GFS-edge outperforms existing methods from all three traditional EP categories, showcasing superior performance while effectively addressing uncertainty modeling and interpretability challenges.