CMU OAQA at TREC 2015 LiveQA: Discovering the Right Answer with Clues
Carnegie Mellon University Pittsburgh United States
Pagination or Media Count:
In this paper, we present CMUs automatic, web-based, real-time question answering QA system that was evaluated in the TREC 2015 LiveQA Challenge.This system answers real-user questions freshly submitted to the Yahoo Answers website that have not been previously answered by humans. Given the title and body of the question, we generated multiple sets of keyword queries and retrieved a collection of web pages based on those queries. Then we extracted answer candidates from web pages in the form of answer passages and their associated clue. Finally, we combined both IR- and NLP-based relevance models to rank and select answer candidates. In the TREC 2015 LiveQA evaluations, human assessors gave our system an average score of 1.081 on a three-point scale, the highest average score achieved by a system in the competition the second-best score was .677, and the average score was .465 for the 21 systems evaluated.