Graph based approach for question answering - improving efficiency in natural language processing for small corpora

Abstract

This work presents the process of building a Question/Answer System (QAS) for Brazilian Portuguese based on a corpus extracted from a tech book on the dairy farming domain. Throughout the work, One common problem in Question/Answer Systems is identified and analyzed, namely, the cases where there’s a distance between the terms present in the question and the terms present in the answer - this distance is called “question answering gap”. To elaborate a Question/Answer System capable of overcoming such distance, a query expansion mechanism was developed through the use of a domain Knowledge Network, providing an effective way to give answers even though they do not share the same lexicon of the questions.