Semantic and data-driven hierarchies for personalized models of affect

Abstract

When building personalized models of affect, hierarchical structures are important in creating levels of separation and sharing between models. Past studies have indicated that semantic hierarchies along demographic divisions perform well in estimating affect. This work focuses on comparing these semantic groupings to data-driven hierarchies. A key question is whether data-driven hierarchies can provide additional ways of understanding affect, outside of semantic boundaries. The experiments are conducted in the context of therapy sessions between personal robots and children with autism. The results reveal novel data-driven hierarchies that could grant better understanding of autism and facilitate more versatile interactions between child and robot.