In a paper presented at the 2019 Conference on Empirical Methods in Natural Language Processing (EMNLP), Master’s student Kriti Myer, along with Master’s students Pallavi Patil, Ronak Zala, and Arpit Singh, doctoral student Sheshera Mysore, Distinguished Professor Andrew McCallum, and industry mentors Adrian Benton and Amanda Stent from Bloomberg, described a new approach to predicting how a member of Congress will vote on a bill, using information about the member from news articles and Freebase, a manually curated knowledge base.
Being able to predict how a legislator will vote on a bill is useful for voters, constituents, and other stakeholders. Past work in this area has used existing data on how a legislator has voted on previous bills to predict how they will vote on future bills. However, this method is not possible when a legislator is newly elected or is a candidate for office -- arguably the time when this information would be most useful to voters. The methods proposed by the researchers overcomes this limitation by augmenting a prior model with news article text about the individual, and the knowledge base Freebase. Results showed that the proposed models outperformed the existing model by 33.4% point of accuracy for legislators with no historical voting records.
This paper was the result of a Center for Data Science Industry Mentorship Program project, sponsored by Bloomberg. The Industry Mentorship Program matches Master’s students in the College of Information and Computer Sciences with an industry-proposed project. The students work on the project over the course of a semester, supervised by a doctoral student and Professor McCallum, as well as mentors from the sponsor company.
Myer first heard about the Industry Mentorship Program from fellow CICS Master’s students, particularly those interested in NLP (natural language processing) technologies. The program is highly regarded among graduate students as an opportunity to work with industry partners on real-life problems. Myer and her fellow students benefited a great deal from the structure of the course, and from working directly with their Bloomberg mentors. “We had the opportunity to work on formulating the problem statement, which takes a lot of time and is a crucial step -- something that is generally missing in other coursework. Through interacting with the industry mentors we got to talk to people who have a lot of experience in the field, so we got a sense of how to go about the whole process.”
Traveling to Hong Kong to present at EMNLP was Myer’s first time presenting a poster at a conference, which was an eye-opening experience. “It was really interesting -- the ideas that you hear people presenting, talking about their projects. There was so much discussion about what is going on in the field. It gives you a whole perspective about how much research actually goes on in the world.”
Myer would recommend the Industry Mentorship program to other CICS Master’s students, particularly those excited by NLP. “If you are interested in working in the NLP space, plus in understanding how to go from a vague problem statement to a solution, this is one thing you really should do.”
She sees the program as a win-win for companies and students. “If a company has a problem they are interested in but doesn’t have the resources to devote to it, this is a great opportunity for them -- they get to make progress on work they may not have time for, and it benefits students as well.”
EMNLP, organized by the ACL (Association for Computational Linguistics) special interest group on linguistic data, is a leading conference in the area of Natural Language Processing.
Learn more about the CDS Industry Mentorship Program.