A Hybrid Question Answering System based on Ontology and Topic Modeling

Abstract

A Question Answering (QA) system is an application which could provide accurate answer in response to the natural language questions. However, some QA systems have their weaknesses, especially for the QA system built based on Knowledge-based approach. It requires to pre-define various triple patterns in order to solve different question types. The ultimate goal of this paper is to propose an automated QA system using a hybrid approach, a combination of the knowledge-based and text-based approaches. Our approach only requires two SPARQLs to retrieve the candidate answers from the ontology without defining any question pattern, and then uses the Topic Model to find the most related candidate answers as the answers. We also investigate and evaluate different language models (unigram and bigram). Our results have shown that this proposed QA system is able to perform beyond the random baseline and solve up to 44 out of 80 questions with Mean Reciprocal Rank (MRR) of 38.73% using bigram LDA.