Debate dialog for news question answering system ‘NetTv’-Debate based on claim and reason estimation-

Abstract

NetTv is a question answering system for net news videos with a voice interface. When watching any news, a question answering system can answer a wide range of questions from users for deeper understanding of news. In this study, in order to help users to understand news deeply, when using NetTv, we propose a debate system which can debate with users about the watching news. The debate system estimates the user’s claim (approval/disapproval/neither) and reason (presence/absence) using CNN, and uses the information to select an opinion stored in debate database and debates by presenting it to the user. In the experiment, eight subjects actually use the system to evaluate naturalness (whether the system response to the user’s utterance was natural), deepness (whether the discussion was established), and user satisfaction (whether new knowledge was obtained) in 5 stages. From the result of subjective evaluation, we show that the debate system gives new knowledge and clarify the present problem of the debate system.