Creating missing classes automatically to improve question classification in question answering systems

Internet is one of the main sources for information which attracts million of users to find the answers for their questions. Finding the accurate answer of the question from giant databases or web pages is a very challenging task. Question answering systems answer the question from structured or unstructured databases. Question answering is different from the traditional ad-hoc document retrieval tasks in a way that in simple document search engine, the set of relevant documents are returned in response of the query, whereas, in the question answering systems the response of the query is the correct answer to what is asked. Finding the exact answer is more interesting and useful than getting a list of documents to look through and find the answer manually. Generally, the question classification is first phase in question answering systems. This phase reduces the answer space by pruning out the extra information that is irrelevant by finding out the expected answer type. This paper mainly focuses on Numeric type questions and also discusses briefly about the questions of type Entity and Location. Almost all the previous question classification algorithms evaluated their work by using the classes defined by Li and Roth [1]. The coarse grained class Numeric has fine grained class Other. In this paper, we target and present the mechanism to create new classes to replace the Other class in Numeric class. We present an automatic hierarchy creation method to add new class nodes using the knowledge resources and shallow language processing.