Joint Community and Structural Hole Spanner Detection via Graph Contrastive Learning

Structural hole spanners are nodes in a network that connect different communities, which are located on key information paths and control information flow between different communities, and therefore have an important status from the perspective of network analysis. Due to its definition, the detection of structural hole spanners relies on the partitioning of subgraph or community structures in network, but most existing methods for detecting structural hole spanner rely on known community labels or complex global search, which are difficult to apply to large-scale real-world networks without labels. To address the aforementioned challenges, inspired by success of graph contrastive learning, we propose a self-supervised method for jointly detecting community and structural hole spanner, i.e., a Augmentation-Free contrastive learning framework for jointly detecting Community and structural hole spanner, named AF-Comm. Experimental results on multiple real-world networks demonstrate the superiority of our algorithm on both community detection and structural hole spanner detection tasks.