MHNF: Multi-Hop Heterogeneous Neighborhood Information Fusion Graph Representation Learning

The attention mechanism enables graph neural networks (GNNs) to learn the weights between target node and its one-hop neighbors, thereby improving performance further. However, most existing GNNs are oriented toward homogeneous graphs, in which each layer can only aggregate information of neighbors. Stacking multilayer introduces considerable noise easily leads over smoothing. We propose here a multihop heterogeneous neighborhood fusion representation learning method (MHNF). Specifically, we hybrid metapath autonomous extraction model efficiently extract Then, formulate hop-level aggregation model, selectively aggregates different-hop within same metapath. Finally, hierarchical semantic (HSAF) is constructed, integrate different-path information. In this fashion, paper solves problem aggregating metapaths for tasks. This mitigates limitation manually specifying metapaths. addition, HSAF internal better present at different levels. Experimental results on real datasets show that MHNF achieves best or competitive against state-of-the-art baselines with fraction 1/10 $\sim$ 1/100 parameters computational budgets. Our code publicly available https://github.com/PHD-lanyu/MHNF