Automatic detection of “enthusiasm” in non-task-oriented dialogues using word co-occurrence

A method is proposed for automatically detecting “enthusiastic” utterances in text-based dialogues. Using conditional random fields, our proposed method distinguishes between the enthusiastic and non-enthusiastic parts of a dialogue. Testing demonstrated that it performs as well as human detection. Being able to distinguish between the enthusiastic and non-enthusiastic parts makes it possible to quantitatively analyze the phenomenon of enthusiasm, which should lead to a practical approach to the creation of non-task-oriented agents that can help generate enthusiastic dialogues.