Automated Question Answering System
Oral Defence Date:
Wednesday, December 6, 2017 - 13:45
Asst. Prof. Anagha Kulkarni & Profs. Dragutin Petkovic and Ilmi Yoon
We present an Automated Question Answering system that applies tools and techniques from Natural Language Processing, Information Retrieval, Machine Learning and Data Mining to answer both, factoid and non-factoid, questions from any domain. The system is designed as a series of four modules: (1) Query Formulation Module transforms free-text questions into well-form boolean queries, (2) Document Retrieval Module retrieves relevant documents from multiple sources, (3) Candidate Answer Extraction Module extracts concise candidate answers from the retrieved documents, (4) Answer Selection Module ranks the answers to identify the best answer using Machine Learning algorithms. We conduct a thorough empirical evaluation of the developed system using multiple datasets from TREC LiveQA 2015 and 2016 competitions. The results demonstrate that our system consistently outperforms the best performing Question Answering systems from the LiveQA tracks by at least 14%, while maintaining answer response time under one minute.
Question answering, Query formulation, Answer ranking, Open-domain