Abstract
Graph Neural Networks (GNNs) have become a prominent approach for learningfrom graph-structured data. However, their effectiveness can be significantlycompromised when the graph structure is suboptimal. To address this issue,Graph Structure Learning (GSL) has emerged as a promising technique thatrefines node connections adaptively. Nevertheless, we identify two keylimitations in existing GSL methods: 1) Most methods primarily focus on nodesimilarity to construct relationships, while overlooking the quality of nodeinformation. Blindly connecting low-quality nodes and aggregating theirambiguous information can degrade the performance of other nodes. 2) Theconstructed graph structures are often constrained to be symmetric, which maylimit the model's flexibility and effectiveness. To overcome these limitations,we propose an Uncertainty-aware Graph Structure Learning (UnGSL) strategy.UnGSL estimates the uncertainty of node information and utilizes it to adjustthe strength of directional connections, where the influence of nodes with highuncertainty is adaptively reduced. Importantly, UnGSL serves as a plug-inmodule that can be seamlessly integrated into existing GSL methods with minimaladditional computational cost. In our experiments, we implement UnGSL into sixrepresentative GSL methods, demonstrating consistent performance improvements.