Semi-Supervised Heterogeneous Graph Learning with Multi-level Data Augmentation

In recent years, semi-supervised graph learning with data augmentation (DA) has been the most commonly used and best-performing method to improve model robustness in sparse scenarios few labeled samples. However, of existing DA methods are based on homogeneous while none specific for heterogeneous graph. Differing from graph, faces greater challenges: heterogeneity information requires strategies effectively handle relations, which considers contribution different types neighbors edges target nodes. Furthermore, over-squashing is caused by negative curvature that formed non-uniformity distribution strong clustering complex To address these challenges, this paper presents a novel named Semi-Supervised Heterogeneous Graph Learning Multi-level Data Augmentation (HG-MDA). For problem DA, node topology proposed characteristics And meta-relation-based attention applied as one indexes selecting augmented nodes edges. information, triangle edge adding removing designed alleviate bring gain topology. Finally, loss function consists cross-entropy consistency regularization unlabeled data. order fuse prediction results various strategies, sharpening used. Existing experiments public datasets, i.e., ACM, DBLP, OGB, industry dataset MB show HG-MDA outperforms current SOTA models. Additionly, user identification internet finance scenarios, helping business add 30% key users, increase loans balances 3.6%, 11.1%, 9.8%.