On Generalizing Neural Node Embedding Methods to Multi-Network Problems

Representation learning has attracted significant interest in the community and has been shown to be successful in tasks involving one graph, such as link prediction and node classification. In this paper, we conduct an empirical study of two leading deep learning based node embedding methods, node2vec and SDNE, to examine their suitability for problems that involvemultiple graphs. Although they have been shown to preserve properties necessary for the success of canonical tasks on a single graph, we find that different runs of the same algorithm even on the same graph yield different embeddings. For node embedding methods to apply to multi-graph problems, we note that this finding motivates additional work in learning how to embed different graphs similarly.