DeepWalk with Reinforcement Learning (DWRL) for node embedding

Rathinaraja Jeyaraj, Thirunavukarasu Balasubramaniam, Anandkumar Balasubramaniam, Anand Paul

Abstract: Highlights•DWRL is proposed to minimize the computation time during similarity calculation.•Subsampling is used to frequently select nodes with less number of edges.•For language modeling, the DWRL can be used to find similar words in latent space.•It can be used to find the distance between two nodes in a large graph.•Node classification and link prediction can be achieved with this method.