In order for a dialogue system to have a good relationship with humans, it is essential to understand and express human emotions. In previous studies, the expression of emotions has been done in typologies such as joy and anger. In contrast, this research will realize a dialogue system that can better understand and express human emotions by predicting the feelings associated with people like "what does he/she think this situation." This model take "who" and "what they did" as input and estimate the output of "how they feel" according to a particular "person". This allows the dialogue system to use a more flexible way of expressing impressions, depending on the person. In the future, we will realize a method that learns "characteristics of the expression of impressions" from users in real time during the dialogue, and estimates the kind of person from the user's characteristics of the expression of impressions.
/ Interaction Research Group, Innovative Communication Laboratory