Abstract

This research thesis aims to address complex problems in Human Behavior Understanding from a computational standpoint: to develop novel methods for enabling machines to capture not only what their sensors are perceiving but also how and why the situation they are presented with is evolving in a certain manner. Touching several fields, from Computer Vision to Social Psychology through Natural Language Processing and Data Mining, we will move from more to less constrained scenarios, describing models for automated behavioral analysis in different contexts: from the individual perspective, e.g. a user interacting with technology, to the group perspective, e.g. a brainstorming session; from living labs, e.g. hundreds of people transparently tracked in their everyday life through smart-phone sensors, to the World Wide Web.