Complex network structural analysis based on information supplementation graph contrastive learning

Contrastive learning has garnered significant interest in analyzing complex network architectures, including node and network classifications. In this context, current contrastive learning methods often create different views by removing nodes or features through data augmentation. These methods apply contrastive learning between these views to derive representations of nodes or networks for downstream tasks. However, these methods may not effectively use the information contained in the discarded nodes or features. This paper proposes a contrastive learning framework based on information supplementation for graph structure analysis that aligns with information theory. The study introduces two enhanced graph contrastive learning methods: the first for information compensation and the second for information completion. Then, we applied graph and node classification methods in network structure analysis. Experimental results demonstrate that the contrastive learning method based on the information supplementation framework outperforms existing methods in subsequent tasks. These results validate the effectiveness of contrastive learning with information supplementation.