AttrE2vec: Unsupervised attributed edge representation learning

Representation learning has overcome the often arduous and manual featurization of networks through (unsupervised) feature as it results in embeddings that can apply to a variety downstream tasks. The focus representation on graphs focused mainly shallow (node-centric) or deep (graph-based) approaches. While there have been approaches work homogeneous heterogeneous with multi-typed nodes edges, is gap edge representations. This paper proposes novel unsupervised inductive method called AttrE2Vec, which learns low-dimensional vector for edges attributed networks. It systematically captures topological proximity, attributes affinity, similarity edges. Contrary current advances embedding research, our proposal extends body methods providing representations capturing graph an manner. Experimental show that, compared contemporary approaches, builds more powerful representations, reflected by higher quality measures (AUC, accuracy) tasks classification clustering. also confirmed analyzing projections.