MIT EECS Undergraduate Research and Innovation Scholar
Increasing the robustness of the START question-answering system
START is a natural language question-answering system. Instead of providing a list of results, START gives succinct information answering the question. A question can have many syntactic or lexical variations, which differ significantly. STARTs current design favors precision over recall. The goal of this project is to increase STARTs recall by providing a more flexible, yet still precise, matching system. The idea is to integrate the symbolic representation of START in a learning system, in order to achieve the desired combination of high precision and vast recall. The challenge of the project lies in bridging the gap between natural language annotation and machine learning techniques.
I previously worked on START, developing features for Omnibase, a system that provides uniform access to data on the web for question answering. I updated interfaces to sites including IMDB.com and USNews.com. I interned with Yahoo on the Social Signals team, developing a distributed system for processing content and signals from social networks like Facebook.