Abstract:Heterogeneous graphs are especially important in our daily life, which describe objects and their connections through nodes and edges. For this complex network structure, many heterogeneous graph neural networks have been designed, but the traditional heterogeneous graph neural network has several obvious shortcomings: (1) Models using meta-paths require selection of meta-paths, failing to learn all meta-paths in the dataset, and cannot capture structural information beyond meta-paths. (2) Models that do not apply meta-paths will be limited by the relationships of the original graph. Due to the absence or omission of some very important but costly relationships in the dataset, the ability of the model to learn node features is limited. (3) The existence of noise will affect the optimization of the parameters of the heterogeneous graph neural network models. In response to these problems, we propose our model: Heterogeneous Graph Neural Network with Graph-data Augmentation and Adaptive Denoising (GAAD). We hope that our model is not affected by differences in graph structure and noise nodes in the graph structure. In summary, we hope that our model can adapt to most graph structures. So we propose a method based on graph neural network to mine the hidden relationship between heterogeneous nodes, calculate the possibility of the existence of edges between heterogeneous nodes, strengthen the graph structure by adding highly possible edges, enrich the information transmission between nodes. Then we apply an adaptive noise reduction algorithm to prevent the possible noise diffusion caused by data enhancement in the graph structure. Extensive experiments on three real-world network datasets demonstrate the superiority of GAAD over the state-of-the-art methods. The code is available on https://github.com/xiaojunlou0831/GAAD.

Heterogeneous graph neural network with graph-data augmentation and adaptive denoising

Beyond Homophily and Homogeneity Assumption: Relation-Based Frequency Adaptive Graph Neural Networks

Heterogeneous Graph Contrastive Learning With Augmentation Graph

HGAMLP: Heterogeneous Graph Attention MLP with De-Redundancy Mechanism

Efficient Topology-aware Data Augmentation for High-Degree Graph Neural Networks

Addressing Heterogeneity and Heterophily in Graphs: A Heterogeneous Heterophilic Spectral Graph Neural Network

SHGNN: Structure-Aware Heterogeneous Graph Neural Network

OSGNN: Original graph and Subgraph aggregated Graph Neural Network

Scalable Graph Neural Networks for Heterogeneous Graphs

DAHGN: Degree-Aware Heterogeneous Graph Neural Network

Gahne: Graph-Aggregated Heterogeneous Network Embedding

HAGNN: Hybrid Aggregation for Heterogeneous Graph Neural Networks

A Graph Neural Network with Type-Feature Attention for Node Classification on Heterogeneous Graphs

Node classification oriented Adaptive Multichannel Heterogeneous Graph Neural Network

Heterogeneous graph neural network for attribute completion

HGNN-QSSA: Heterogeneous Graph Neural Networks With Quantitative Sampling and Structure-Aware Attention

Advancing Graph Neural Networks with HL-HGAT: A Hodge-Laplacian and Attention Mechanism Approach for Heterogeneous Graph-Structured Data

Meta-path-based heterogeneous graph neural networks in academic network

Megnn: Meta-path extracted graph neural network for heterogeneous graph representation learning

Graph Data Augmentation for Node Classification

HGBER: Heterogeneous Graph Neural Network With Bidirectional Encoding Representation