Question Classification using Maximum Entropy Models
Krystle Kocik · 2004
User demand for a more powerful search has introduced research in Question Answering (QA) systems. QA systems return the exact answer to a users question. This differs to current search engines such as Google, which return a list of documents that may be relevant. Components within a QA system require deep linguistic, semantic and syntactic analysis of questions posed by users and potential answers within documents. One such component is Question Classification (QC). This involves determining what a question is asking and classifying it into its corresponding answer type. Current QC systems use Machine Learning methods to perform this task. Maximum Entropy Modelling, a statistical technique, has been successfully used in many areas of Natural Language Processing however has not been applied to QC. This project involves identifying a rich feature set for QC using Maximum Entropy Models. Results will show that this technique achieves a state-of-the-art accuracy. iii Acknowledgements Thank you to my family, for your continuous love and support particularly during this hectic year.