Dual-level noise augmentation for graph clustering with triplet-wise contrastive learning

Open Webpage

Tianxiang Zhao, Youqing Wang, Shilong Xu, Tianchuan Yang, Junbin Gao, Jipeng Guo

Published: 2026, Last Modified: 23 Jan 2026Pattern Recognit. 2026EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•DNA-CGC strengthens noise benefits in contrastive graph clustering from two levels.•Noisy high-frequency information and random noise improve embedding discriminability.•Triplet contrastive learning with noise as exclusive negatives alleviate contrastive bias.•Extensive experimental results verify its clear superiority and effectiveness.