Question To Answer To The Classification Of The System

More and more information sources are now available in machine-readable form due to the rapid development of Internet. As more and more information is accessible to users, people expect quickly, and accurately to obtain the information needed. This brings new challenges to the area of information retrieval (IR) in both query and answer processing. As a very active branch of natural language processing, Open-Domain Question Answering (QA) investigates how to understand user's questions and give the correct answer to user.Question classification, an active research field, is an important module of QA system. Its task is to understand the demand of users. It is very helpful for QA system to find the correct answer.This research provides some new insights into Question Classification in QA:Firstly new feature sets such as Dependency Structure and WordNet synset, etc. are used in Vector Space Model to represent question. In addition, feature selection in question representation was also investigated. These improve the performance by 6%.Secondly different classifiers and ensemble classifiers with different feature sets are compared. With synset from WordNet and Dependency Structure from Minipar as question representations, and using ensemble classifiers based on TBL, a 1.6% improvement over the best known accuracy is achieved.In addition, Question taxonomy is also studied. It is argued that a good taxonomy should be evaluated, on not only itself but also its impact on QA.This question classification system has been used in FDUQA, a QA system of Fudan University, as a module and the FDUQA performed well in TREC13.