From Node Embedding To Community Embedding

In this paper, we introduce a new setting for graph embedding, which considers embedding communities instead of individual nodes. Community embedding is useful as a natural community representation for applications, and it provides an exciting opportunity to improve community detection. Specifically, we see the interaction between community embedding and detection as a closed loop, through node embedding. On the one hand, we rely on node embedding to generate good communities and thus meaningful community embedding. On the other hand, we apply community embedding to improve node embedding through a novel community-aware higher-order proximity. This closed loop enables us to improve community embedding, community detection and node embedding at the same time. Guided by this insight, we propose ComE, the first community embedding method so far as we know. We evaluate ComE on multiple real-world data sets, and show ComE outperforms the state-of-theart baselines in both tasks of community prediction and node classification. Our code is available at https://github.com/andompesta/ nodeembedding-to-communityembedding.