Many research disciplines study data with an underlying graph structure, including social networks, knowledge graphs, information networks, protein networks, and sensor networks. Due to the elaborate and complex information embedded in the graph-structured data, there is an increasing need for new optimization methods and neural network architectures that can accommodate such relational structures and the embedded information. Recent years have seen a surge in approaches addressing this need. These network representation learning (NRL) approaches have led to promising results in numerous tasks such as node/graph classification, link prediction, personalized recommendation, and document representation.
In this talk, I will start with a brief tour of this research and discuss its rapid growth over the past couple of years. Next, I will review some of the more recent advancements in random-walk and deep learning based approaches. In the latter half, I will present real-world applications where the graph structure enables us to learn enhanced representations of the data. Finally, I will demonstrate my research agenda and propose future research directions.