- Abstract: Graph convolutional networks and its variants are the state-of-the-art methods for learning node embeddings in a graph-structured data. However, these existing approaches fail to capture the neighborhood information efficiently beyond a certain depth from the node. In this work, we propose a novel hierarchical feature aggregation approach which explictly aggregates the feature information from different depths of a node's neighborhood using an LSTM model. Proposed model gives promising results on four real-world datasets as compared to state-of-the-art methods.
- Keywords: Graph Learning, Networks, Representation Learning, Deep Learning