Representation Learning for Graph-Structured Data

Graph-structured data is ubiquitous in science, engineering and has been successfully used in various real-life applications for social networks, molecular graphs, and biological networks. Hence, it is worth exploring prospective mechanisms to deal with the unprecedented growth in volumes and problem complexity of graph-structured data. Graph representation learning has recently emerged as a new promising paradigm, which learns a parametric mapping function that embeds nodes, subgraphs, or the entire graph into low-dimensional continuous vectors. In this thesis, we focus on developing novel and advanced graph embedding models for the two most popular types of graphs: undirected graph and knowledge graph.