Streaming Behaviour: Livestreaming as a Paradigm for Analysis of Emotional and Social Signals

Abstract

We argue that video game livestreams constitute an invaluable paradigm towards building multi-view, data-driven models of human behaviour. The interactive setting under which a stream operates is enriched with social signals, conveyed between streamers and viewers via facial expressions, body movement, vocal cues, as well as written language. We consider the data sources involved in a typical broadcast (e.g., camera stream, game footage, text) as data-views that carry inherent correlations, since they all describe events occurring during a stream. We argue that this unique interactive setting facilitates the joint, multi-view analysis of human behaviour in groups, utilizing the various heterogeneous data sources involved in a coherent and self-contained manner. We elaborate on the emergence of social signals in this setting, while discussing close links and potential research directions related to areas such as affective computing, machine learning, computer vision and intelligent game design.