Question Answering using Vector Based Information Retrieval Paradigm with Word Sense Disambiguation

Vector Based Text Classification in the Question Answering field has long been explored. However, there has not been any attempt so far to take word senses into consideration in the development of the feature sets in the classifier. This paper aims to investigate the performance of a question answering text classifier built using not just the root form of words but also taking the senses of those words into thought. Having done a 10folded cross validation, the classification error rate using the tri-gram model actually shows that there is a significant improvement when the word sense of a word is actually known. A chi squared analysis performed to see the correlation between the use of word sense and the affect on the classification accuracy shows a 97.5% confidence level. This simply tells us that the usage of word sense in building the classifier has indeed a strong association with the classification accuracy.