Lunch at 12:30pm, talk at 1pm, in 148 Fitzpatrick
Title: Incorporating Knowledge Graphs to Enhance Open-domain Question Answering
Abstract: Open-domain question answering (QA) is a knowledge intensive NLP task, where in order to produce a correct answer, the learning modules usually need to fetch helpful resources from external knowledge bases. A common thread of open-domain QA models employs a retriever-reader pipeline that first retrieves a handful of relevant passages from Wikipedia and then peruses the passages to produce an answer. Besides relevant passages, to improve the QA performance, many state-of-the-art open-domain QA frameworks also explore knowledge graphs (KGs) that describe abstract semantic relationships between millions of entities. In this talk, three exemplary works that incorporate KGs into open-domain QA frameworks will be discussed, and each of these works spans one of three high-level philosophies, including unifying KG triplets as raw texts, enhancing the retriever by KGs, and enhancing the reader by KGs. Some very recent works on open-domain QA that go beyond the pipeline of retrieve-and-generate will also be discussed.
Bio: Mingxuan (Clark) Ju is a third-year Ph.D. student at the University of Notre Dame, fortunately advised by Dr. Fanny Ye. His research focuses on graph machine learning and natural language processing.