Multi-graph Fusion and Virtual Node Enhanced Graph Neural Networks

Graph Neural Networks (GNNs) have emerged as a dominant tool for effectively learning from graph data, leveraging their remarkable learning capabilities. However, many GNN-based techniques assume complete and accurate graph relations. Unfortunately, this assumption often diverges from reality, as real-world scenarios frequently exhibit missing and erroneous edges within graphs. Consequently, GNNs that rely solely on the original graph structure inevitably lead to suboptimal results. To address this challenge, we propose a novel approach known as Multi-graph fusion and Virtual node enhanced Graph Neural Networks (MVGNN). Initially, we introduce an adaptive graph that complements the original and feature graphs. This adaptive graph serves to bridge gaps in the original and feature graphs, capturing missing edges and refining the graph's structure. Subsequently, we merge the original, feature, and adaptive graphs by applying attention mechanisms. In addition, MVGNN strategically designs virtual nodes, which act as auxiliary elements, changing the propagation mode between low-weighted edges and further enhancing the robustness of the model. The proposed MVGNN is evaluated on six benchmark datasets, demonstrating its superiority over existing state-of-the-art classification methodologies.