Cyclic label propagation for graph semi-supervised learning

Graph neural networks (GNNs) have emerged as effective approaches for graph analysis, especially in the scenario of semi-supervised learning. Despite its success, GNN often suffers from over-smoothing and over-fitting problems, which affects performance on node classification tasks. We analyze that an alternative method, label propagation algorithm (LPA), avoids aforementioned problems thus it is a promising choice Nevertheless, intrinsic limitations LPA feature exploitation relation modeling make propagating labels become less effective. To overcome these limitations, we introduce novel framework learning termed Cyclic Label Propagation (CycProp abbreviation), integrates GNNs into process cyclic mutually reinforcing manner to exploit advantages both LPA. In particular, our proposed CycProp updates embeddings learned by module with augmented information propagation, while fine-tunes weighted help embedding turn. After model converges, reliably predicted informative are obtained modules respectively. Extensive experiments various real-world datasets conducted, experimental results empirically demonstrate can achieve relatively significant gains over state-of-the-art methods.