Question Type Classification for Comic QA System

Abstract

The objective of our research is to realize a question answering (QA) system for comics. Because comic is a multi-modal content that utilizes texts and illustrations cooperatively, question sentence that should be handled by the comic-QA system varies significantly in comparison with the conventional QA system. To meet this goal, this paper performs type classification of the question for comics as a basic examination. We classified question sentences into query types: bibliographic information type questions (5 types) and content information type questions (6 types). These types are determined by the result of previous works and question sentences collected from Web sites. We performed automatic classification based on the classification. As the result, we observed that accuracy was high in bibliographic information type question, while that in content information type question was low.