Statistical QA - Classifier vs. Re-ranker: What's the difference?

摘要

In this paper, we show that we can obtaina good baseline performance forQuestion Answering (QA) by usingonly 4 simple features. Using these features,we contrast two approaches usedfor a Maximum Entropy based QA system.We view the QA problem as aclassification problem and as a rerankingproblem. Our results indicatethat the QA system viewed as a rerankerclearly outperforms the QA systemused as a classifier. Both systemsare trained using the same data.