Topic: Choose No More: Combining Keyword Search and Supervised Learning
Speaker: Mr.Eugene Yang
Date: 6/17 (Wed.) 3:30-5:00PM
Location: Google Hangouts Meet
URL is https://meet.google.com/jmz-oxqu-euw
Bio:
Eugene Yang is a computer science Ph.D. candidate at Georgetown University under the advice of Ophir Frieder, Jeremy Fineman, and David D. Lewis. He received a bachelor's degree in quantitative finance from National Tsing Hua University. His research focuses on total recall retrieval and technology-assisted review, especially in the legal applications. His interest includes Bayesian supervised learning, sequential decision problems, and the explainability of machine learning models.
Abstract:
Traditionally, supervised classification and information retrieval are considered as distinct problems with differing input. While classification requires a set of annotated data points, retrieval models only demand a textual query to rank the documents. Classification models, in contrast, once trained, sustain greater accuracy and efficiency at separating the wheat from the chaff. The obvious question is: Given both forms of information — textual keywords and labeled documents -- can we utilize both? Ignoring either is information loss; combining them is believed complicated. In this talk, within the domain of legal information processing, we develop an integration framework that combines both information types into a single model. The resulting approach capitalizes on the advantages of each information type, achieving a resource-efficient and accurate system. Ethical issues of machine learning within legal applications are likewise addressed.
earning models.