Using a named entity tagger and a syntactic parser to improve web-based answer extraction

The amount and the quality of the available information on the web make it an interesting resource for seeking quick answers to simple questions. Question answering (QA) systems have proven to be helpful to users because they can provide accurate answers that do not require users to go through a large number of documents for an answer. However, despite the recent advances in QA research, the accuracy of the extracted answer is still an open domain that needs more investigation from the researchers to achieve a high accuracy. In this project, we implemented Named-entity tagging ( Gate-NE ) and Grammar Parsing ( Link parser ) as two different approaches to improve the extracted answer accuracy of an existing web-QA system. TREC-8 (200 questions) was used as a training set and a total of 1693 questions of TREC-9, 10, and 2002 were used for the validation process. Our approach shows a 11% MRR increase, from 0.101 to 0.113, compared to the original system and a 9% increase in the number of correct answers extracted, from 276 to 300. Although the increase is not as high as we had hoped, we believe that our results are encouraging and this work should be considered a base and starting point for future work to achieve more performance enhancements