文献类型：会议论文

英文题名：Heterogeneous Graph Contrastive Multi-view Learning

作者：Wang, Zehong[1];Li, Qi[2];Yu, Donghua[2];Han, Xiaolong[3];Gao, Xiao-Zhi[4];Shen, Shigen[2,5]

机构：[1]Univ Leeds, Sch Math, Leeds, England;[2]Shaoxing Univ, Dept Comp Sci & Engn, Shaoxing, Peoples R China;[3]Nanjing Univ Informat Sci & Technol, Sch Comp & Software, Nanjing, Peoples R China;[4]Univ Eastern Finland, Sch Comp, Kuopio, Finland;[5]Huzhou Univ, Sch Infonnat Engn, Huzhou, Peoples R China

会议论文集：SIAM International Conference on Data Mining (SDM)

会议日期：APR 27-29, 2023

会议地点：St. Paul Twin Cities, MN

语种：英文

外文关键词：Graph contrastive learning; heterogeneous information network; self-supervised learning; graph neural network; multi-view learning

外文摘要：Inspired by the success of Contrastive Learning (CL) in computer vision and natural language processing, Graph Contrastive Learning (GCL) has been developed to learn discriminative node representations on graph datasets. How-ever, the development of GCL on Heterogeneous Information Networks (HINs) is still in the infant stage. For ex-ample, it is unclear how to augment the HINs without substantially altering the underlying semantics, and how to de-sign the contrastive objective to fully capture the rich semantics. Moreover, early investigations demonstrate that CL suffers from sampling bias, whereas conventional debiasing techniques are empirically shown to be inadequate for GCL. How to mitigate the sampling bias for heterogeneous GCL is another important problem. To address the afore-mentioned challenges, we propose a novel Heterogeneous Graph Contrastive Multi-view Learning (HGCML) model. In particular, we use metapaths as the augmentation to generate multiple subgraphs as multi-views, and propose a contrastive objective to maximize the mutual information be-tween any pairs of metapath-induced views. To alleviate the sampling bias, we further propose a positive sampling strategy to explicitly select positives for each node via jointly considering semantic and structural information preserved on each metapath view. Extensive experiments demonstrate HGCML consistently outperforms state-of-the-art baselines on five real-world benchmark datasets. To enhance the reproducibility of our work, we make all the code publicly avail-able at https://github.com/Zehong-Wang/HGCML.