Tag
This paper proposes SCISE, a scalable unsupervised graph clustering framework that uses community-aware sampling and structural entropy to overcome structural isolation in mini-batch training, achieving state-of-the-art results on benchmark datasets.