New Graph Clustering Method Boosts Scalability and Cohesion.

Unsupervised graph clustering is vital for uncovering hidden patterns in large-scale networks, but current Graph Contrastive Learning methods often suffer from "structural isolation" during mini-batch training. This issue makes it difficult to capture cohesive community structures that reflect the global topological distribution of the network. To address this, researchers propose SCISE (Scalable unsupervised graph Clustering framework that preserves structural Integrity). SCISE integrates a Structural Entropy Community Constraint (SECC) operator, which optimizes structural information within a constrained solution space to prevent community fragmentation and improve partition cohesion. Furthermore, SCISE employs a Community-Aware Sampling Expansion (CSampE) mechanism. This mechanism incorporates the community context of target nodes into sampling batches, effectively breaking structural barriers and maintaining topological integrity during batch training. Finally, a Structural Contrastive Learning (StructCL) module refines edge weights based on intra-batch structural similarity, guiding the encoder to learn higher-order structural representations. Extensive experiments on six mainstream benchmark datasets demonstrate that SCISE significantly outperforms state-of-the-art algorithms.