[4E3-GS-2] Machine learning: graph structure

[4E3-GS-2-02] Link Embedding Learning Method

〇Shu Liu¹, Fujio Toriumi¹ (1. The University of Tokyo)

Keywords:Complex Networks, Link Embedding Learning, Hypergraph, Line Graph

A complex network is one of the data structures showing the elements (nodes) and the relationships (links) between the elements. It helps investigate the properties of the entire complex system regardless of individual elements. Network embedding is the task of embedding nodes, links, or the network itself in higher-order latent space to learn vector representations. This research proposes a method for learning embedded expressions of links. First, the link is switched to a node by converting the network into a hyper network-type edge-dual network. Next, we create a multi-resolution context for the node through a scalable random walk on the hyper network. Finally, the embedded representation of the node is learned from the context by the method of natural language processing. As evaluation experiments, the degree of discrimination of the structural features of the link is confirmed from the embedded representation using the toy network and the generation network. Furthermore, in the link prediction task, the accuracy is confirmed with the conventional method.