抄録
近年,人間と会話のできる対話システムが社会に広がりつつある.しかし,人間と雑談をし,話を盛り上げるための非タスク指向型対話システムの研究は少なく,知的な応答が可能な非タスク指向型対話システムは皆無である.本論文では知的な応答が可能な非タスク指向型対話システムに必要となる知識データ構築を目指し,意味ネットワークを対話から構築する.さらにその応用として非タスク指向型対話システムの客観的・定量的な評価法を提案する.実験の結果,複数の非タスク指向型対話システムの性能を比較評価できることが確認された.また,人間同士の対話についても比較評価できる可能性が確認された.
Recently, computerized dialogue systems have been actively investigated and used in various fields. However, There are few researches of the systems to talk with a human, and there are no systems that can make an intellectual reply. In this paper, we build a semantic network from a dialogue aiming at the knowledge data construction that is necessary for non-task-oriented dialogue systems. Furthermore, we propose a new criterion which can evaluate non-task-oriented dialogue systems objectively and quantitatively using semantic network. As a result of an experiment, it was clarified that our method can evaluate dialogue systems. In addition, we also carried out a similar experiment with human-human dialogues. The result suggested the possibility that our method can evaluate the dialogues using semantic network.
