This website requires JavaScript.

Hard Sample Aware Network for Contrastive Deep Graph Clustering

Yue LiuXihong YangSihang Zhou ...+6 Cancan Chen
Dec 2022
Contrastive deep graph clustering, which aims to divide nodes into disjointgroups via contrastive mechanisms, is a challenging research spot. Among therecent works, hard sample mining-based algorithms have achieved great attentionfor their promising performance. However, we find that the existing hard samplemining methods have two problems as follows. 1) In the hardness measurement,the important structural information is overlooked for similarity calculation,degrading the representativeness of the selected hard negative samples. 2)Previous works merely focus on the hard negative sample pairs while neglectingthe hard positive sample pairs. Nevertheless, samples within the same clusterbut with low similarity should also be carefully learned. To solve theproblems, we propose a novel contrastive deep graph clustering method dubbedHard Sample Aware Network (HSAN) by introducing a comprehensive similaritymeasure criterion and a general dynamic sample weighing strategy. Concretely,in our algorithm, the similarities between samples are calculated byconsidering both the attribute embeddings and the structure embeddings, betterrevealing sample relationships and assisting hardness measurement. Moreover,under the guidance of the carefully collected high-confidence clusteringinformation, our proposed weight modulating function will first recognize thepositive and negative samples and then dynamically up-weight the hard samplepairs while down-weighting the easy ones. In this way, our method can mine notonly the hard negative samples but also the hard positive sample, thusimproving the discriminative capability of the samples further. Extensiveexperiments and analyses demonstrate the superiority and effectiveness of ourproposed method.