Hard Sample Aware Network for Contrastive Deep Graph Clustering

Contrastive deep graph clustering, which aims to divide nodes into disjoint groups via contrastive mechanisms, is a challenging research spot. Among the recent works, hard sample mining-based algorithms have achieved great attention for their promising performance. However, we find that existing mining methods two problems as follows. 1) In hardness measurement, important structural information overlooked similarity calculation, degrading representativeness of selected negative samples. 2) Previous works merely focus on pairs while neglecting positive pairs. Nevertheless, samples within same cluster but with low should also be carefully learned. To solve problems, propose novel clustering method dubbed Hard Sample Aware Network (HSAN) by introducing comprehensive measure criterion and general dynamic weighing strategy. Concretely, in our algorithm, similarities between are calculated considering both attribute embeddings structure embeddings, better revealing relationships assisting measurement. Moreover, under guidance collected high-confidence information, proposed weight modulating function will first recognize then dynamically up-weight down-weighting easy ones. this way, can mine not only sample, thus improving discriminative capability further. Extensive experiments analyses demonstrate superiority effectiveness method. The source code HSAN shared at https://github.com/yueliu1999/HSAN collection (papers, codes and, datasets) https://github.com/yueliu1999/Awesome-Deep-Graph-Clustering Github.