Unsupervised multiplex graph representation learning via maximizing coding rate reduction

Unsupervised multiplex graph representation learning (UMGRL) has gained increasing attention for its effectiveness to extract discriminative and consistent representations without labels. However, previous methods ignore the diversity of extracted representations, leading to sub-optimal results. To address the aforementioned limitations, in this paper, we propose a unified framework to extract discriminative, diverse and consistent representations simultaneously for UMGRL. To do this, we first employ the Multi-Layer Perceptron encoder with the local preserve loss to extract high-quality representations, and then employ two constraints based on the coding rate to constrain representations' diversity, discrimination, and consistency. Comprehensive experiments are conducted to verify the effectiveness of the proposed model. The results show that our method outperforms fourteen existing methods on four public benchmark datasets for three different downstream tasks. The code is available at https://github.com/OllieWangx/D2CMG.