Improved Question Answering System by semantic refomulation

Abstract

A unbearable amount of textual information accessible in electronic form and need to deliver correct answer to user question is important task in Question Answering System (QAS). Question answer system is the form of information retrieval system which aims to deliver the exact answer to the user question rather than whole document. To answer this user need semantic based reformulation techniques can be used to retrieve the accurate answer from enormous number of document retrieved from the search engine. The goal is to generate the pattern from the web based on lexical semantic and syntactic constrain. These constrain should be defined in the question answering system to evaluate and rank the candidate answer. Here we used TREC-8, TREC-9, and TREC-10 collection as training set. Different types of question and corresponding answer can use from TREC collection. The proposed system retrieves the answer automatically from TREC collection. Word net can be used to help the semantic relation and syntactic tag between the questions and answer pair. Finally weight can be given to each candidate answer according to their length, the level of semantic similarity between question and answer pair and distance between the key word. The proposed QAS be different from other reformulation based Question answering system. The experiments on the TREC data set will show the better result which can be calculated with help of the precision and recall.