PyDGC
Benyu Wu†, Yue Liu† CODE

• PyDGC is a unified, flexible, and extensible framework for deep graph clustering.
• It provides a variety of graph clustering methods and datasets, making it easy to reproduce, implement, and evaluate algorithms.
• It also includes a comprehensive set of evaluation metrics for graph clustering tasks.

DGCBench: A Deep Graph Clustering Benchmark
Benyu Wu†, Yue Liu†, Qiaoyu Tan, Xinwang Liu, Wei Du, Jun Wang, Guoxian Yu*

CODE

• DGCBench is the first systematic DGC benchmark, including 12 diverse datasets, 12 SOTA methods and a standardized pipeline for fair, reproducible evaluations.
• It provides holistic multi-faceted analysis of DGC methods, covering effectiveness, efficiency, robustness, stability, scalability and discriminability beyond basic metrics to reveal weaknesses.
• It provides PyDGC, an open-source Python toolkit compatible with PyG and OGB, supports flexible integration of new models and datasets for DGC development.

Imputation-free Incomplete Multi-view Clustering via Knowledge Distillation
Benyu Wu, Wei Du*, Jun Wang, Guoxian Yu*

CODE

• I2MVC is an imputation-free framework for incomplete multi-view clustering, avoiding error accumulation and reducing processing complexity.
• It uses divide-and-conquer with pseudo-supervised knowledge distillation, enabling single-view-based incomplete data clustering without all views.

Synergistic Deep Graph Clustering Network
Shifei Ding, Benyu Wu*, Xiao Xu*, Ling Ding and Xindong Wu

CODE

• TIGAE uses simple linear transformation to introduce explicit graph Laplacian information to alleviate representation collapse.
• A synergistic framework of representation learning and structure augmentation is proposed to exploit the reciprocal relationship between them to jointly improve the embedding quality.
• Structure fine-tuning strategy improve the generalization ability of the model.

Towards Faster Deep Graph Clustering via Efficient Graph Auto-Encoder
Shifei Ding, Benyu Wu, Ling Ding, Xiao Xu, Lili Guo, Hongmei Liao* and Xindong Wu

CODE

• EGAE optimizes the GAE from the perspectives of data dimension and graph convolution efficiency.
• Dynamic graph weight updating strategy adjust the structure during the training process.

Graph Clustering Network with Structure Embedding Enhanced
Shifei Ding, Benyu Wu, Xiao Xu, Lili Guo and Ling Ding*

CODE

• GC-SEE enhances the structural information in embeddings for clustering by utilizing different types of structural information.