My general interest lies in the broad areas of data mining, machine learning and artificial intelligence. More specifically, I am working on the following research topics: (1) Learning on graphs and information networks; (2) Web and social media analytics; (3) Recommendation systems; (4) Application of knowledge graphs on data mining problems. Some of my research projects are summarized below, and you can read my research statement and the full list of publications.
Meta-learning on graphs
Meta-learning, also known as learning-to-learn, has shown great promise in many domains such as computer vision and natural language processing. It is particularly useful to the scenario where there is a need to adapt to new input or task (e.g., few-shot classification on novel classes), such that the adaptation process can be learned instead of hand designed. Specific techniques vary, such as the optimization-based MAML and metric-based ProtoNet, and in a broader sense hypernetworks. In this project, we investigate the emerging trend of meta-learning on graphs, which have found important use cases for graph-based learning and mining, such as few-shot or semi-supervised node classification [AAAI21a, SIGIR21], tail node embedding [CIKM20, KDD21], graph neural networks [AAAI21b, IJCAI21], and cold-start recommendation on graphs [KDD20].
Metagraph: Semantics-guided graph representations
Heterogeneous Information Networks (HIN) often model complex, multi-typed relationships between different types of objects. Metagraphs [ICDE16, TKDE19], which model recurring subgraph patterns on HINs, emerge as a powerful tool to capture the rich semantics on a HIN. Thus, metagraphs can be used to derive powerful graph representations [TKDE20] or pre-train graph neural network models [CIKM21] on HINs. The learned representations enable a wide variety of downstream applications, such as disease gene prediction [Method17].