GAME-ON: A Multimodal Dataset for Cohesion and Group Analysis

This paper presents GAME-ON (Group Analysis of Multimodal Expression of cohesiON), a multimodal dataset specifically designed for studying group cohesion and for explicitly controlling its variation over time. Cohesion is here addressed according to the Severt and Estrada's theoretical multidimensional integrative framework. More specifically, GAME-ON focuses on the social and task dimensions of the instrumental function of cohesion. The dataset consists of over 11 hours of synchronized multimodal recordings (audio, video, and motion capture data) of 17 small groups (3 persons) playing a social game, i.e., an escape game. The game consists of several tasks designed to manipulate the variation of cohesion over time. GAME-ON includes annotations consisting of self-assessment of cohesion and other constructs such as emotions, leadership, and warmth and competence. A first statistical analysis of these annotations shows that we successfully manipulated all the relative variations of cohesion (between tasks) over time. This holds for all tasks except for one where we observed a significant variation of cohesion in the opposite direction than expected. The dataset will be publicly available for research purposes. The motivation of our work is to provide the scientific community with an asset for studying cohesion and other group phenomena.