CBR tagging of emotions from facial expressions

Abstract

Mobility and context-awareness are two active research directions that open new potential to recommender systems. Usage of dynamically enriched information from the user context leads the system to find better solutions that are adapted to the specific situations. In this paper we focus on the difficult problem of dynamically acquiring the emotional context about the user during a recommendation process. We use the fact that emotions are tightly connected with facial expressions and it is difficult for people to hide emotions in facial expressions. We describe PhotoMood, a CBR system that uses gestures to identify emotions in faces, and present preliminary experiments with MadridLive, a mobile and context aware recommender system for leisure activities in Madrid. In the experiments, the momentary emotion of a user is dynamically detected from pictures of the facial expression taken unobtrusively with the front facing camera of the mobile device.