CBR Tagging of Emotions from Facial Expressions

In our current research we use CBR to identify the emotional state of an user during her interaction with a recommender system by analysing pictures of her momentary facial expression. In a previous work [2] we introduced PhotoMood, a CBR system that uses gestures to identify emotions from the user face self-pictures, and presented preliminary experiments analysing only the external mouth contour of the faces. In this paper we extend the experiments to analyse other gestures in the faces, like eyes and eyebrows. We demonstrate that different people express emotions with different gestures and apply a GA to compute the specific set of weights that maximize the precision results for each user. We experiment with different configurations of case bases and set of weights and demonstrate that the system obtains good precision results even in the cold start situation using only a general case base made of pictures of anonymous people.