GraphDHV: Graph Neural Network with Dual Hybrid View on Imbalanced Node Classification

Longqing Du, Guangquan Lu, Yadan Han, Zhiping Luo, Guoqiu Wen, Liang Zhang, Wanxin Chen, Shichao Zhang

Abstract: It is crucial to address category imbalance in real-world data on node classification in a graph. This paper introduces the GraphDHV model, a dual encoder that integrates node attributes and topological information. GraphDHV improves inter-class separability and intra-class compactness by employing degree logical edge removal and synthesizing enhanced attribute features with original topology. We compare GraphDHV with ten state-of-the-art baseline methods on six public benchmark datasets. Our experimental results demonstrate that GraphDHV significantly outperforms the baseline methods, with an improvement of the F1 score.

External IDs:dblp:conf/cocoon/DuLHLWZCZ24