Improving Gaussian Embedding for Extracting Local Semantic Connectivity in Networks

Gaussian embedding in unsupervised graph representation learning aims to embed vertices into Gaussian distributions. Downstream tasks such as link prediction and node classification can be efficiently computed on Gaussian distributions. Existing Gaussian embedding methods depending on adjacent neighbors of vertexes leave out of consideration of indirect connectivity information carried by remote vertice, which is still a part of local semantic connectivity in a network. In this paper, we propose an unsupervised graph representation model GLP2Gauss to improve Gaussian embedding by coarsening graph as indirect local connectivity. First, we decompose the original graph into paths and subgraphs. Secondly, we present a path-based embedding strategy combined with the Gaussian embedding method to maintain better local relevance of embedded nodes. Experiments evaluated on benchmark datasets show that GLP2Gauss has competitive performance on node classification and link prediction for networks compared with off-the-shelf network representation models.