Robust Node Embedding Based on Subspace Attribute Complement Topology

Graph Neural Networks (GNNs) have been proven to be powerful tools for graph analysis. However, despite their success, existing GNNs are often sensitive to the quality of the input graph. Real-world graphs are typically noisy, often represented topologically as containing some edges irrelevant to the task or missing some edges, which can lead to sub-optimal generalization performance of the learned GNNs models. To address this issue, we introduce a robust node representation algorithm for attribute networks based on the concept of subspace (STSAI) algorithm. Adequate experiments were also done on the real-world dataset to verify the validity of this work.