GL2vec: Graph Embedding Enriched by Line Graphs with Edge Features

Recently, several techniques to learn the embedding for a given graph dataset have been proposed. Among them, Graph2vec is significant in that it unsupervisedly learns the embedding of entire graphs which is useful for graph classification. This paper develops an algorithm which improves Graph2vec. First, we point out two limitations of Graph2vec: (1) Edge labels cannot be handled and (2) Graph2vec does not always preserve structural information enough to evaluate the structural similarity, because it bundles the node label information and the structural information in extracting subgraphs. Our algorithm overcomes these limitations by exploiting the line graphs (edge-to-vertex dual graphs) of given graphs. Specifically, it complements either the edge label information or the structural information which Graph2vec misses with the embeddings of the line graphs. Our method is named as GL2vec (Graph and Line graph to vector) because it concatenates the embedding of an original graph to that of the corresponding line graph. Experimentally, GL2vec achieves significant improvements in graph classification task over Graph2vec for many benchmark datasets.