Graph Attention Networks with Positional Embeddings

Graph Neural Networks (GNNs) are deep learning methods which provide the current state of art performance in node classification tasks. GNNs often assume homophily – neighboring nodes having similar features and labels–, therefore may not be at their full potential when dealing with non-homophilic graphs. In this work, we focus on addressing limitation enable Attention (GAT), a commonly used variant GNNs, to explore structural information within each graph locality. Inspired by positional encoding Transformers, propose framework, termed Attentional Positional Embeddings (GAT-POS), enhance GATs embeddings capture graph. learned model predictive context, plugged into an enhanced GAT architecture, is able leverage both content node. The trained jointly optimize for task as well predicting context. Experimental results show that GAT-POS reaches remarkable improvement compared strong GNN baselines recent embedding