Edge Role Discovery via Higher-Order Structures

Previous work in network analysis has focused on modeling the roles of nodes in graphs. In this paper, we introduce edge role discovery and propose a framework for learning and extracting edge roles from large graphs. We also propose a general class of higher-order role models that leverage network motifs. This leads us to develop a novel edge feature learning approach for role discovery that begins with higher-order network motifs and automatically learns deeper edge features. All techniques are parallelized and shown to scale well. They are also efficient with a time complexity of $$\mathcal {O}(|E|)$$ . The experiments demonstrate the effectiveness of our model for a variety of ML tasks such as improving classification and dynamic network analysis.