1P1-C21 複数人への聞き込みから記事を作成するロボットシステム

概要

論文の詳細を見る
In previous work we proposed a "Journalist Robot" that writes a local news article based on a single interview. Because of limitations in the robot's ability to understand the world directly, it instead asks bystanders questions about a witnessed event. News was defined as an "unexpected truth". We analyzed basic components of a news article obtained through multiple interviews, and evaluated the questions based on the above definition. The results indicate that the robot should use observable information to generate initial questions, extract information based on morphological analysis, form text clusters, and verify the truth and predictability of events through interviews. We implemented the proposed method on a real world robot, and conducted an experiment. The experiment shows that the robot can collect local news using the proposed method.