4 Working Groups 4.1 Working Group on Datasets with Humans and Agents in Groups

Data is, to put it simply, another word for information. Recent years have seen a growth in datasets capturing data relevant for group interactions. While data was predominantly collected to form a quantitative understanding of social interactions, the recent attention for machine learning has resulted in datasets that contain features and annotations which are aimed at classifying or predicting social behavior. Due to the data-hungry nature of machine learning and the low cost of storage, the size of typical datasets is now several orders of magnitude larger than a decade ago.We describe what constitutes a good dataset, and provide a number of resources related to datasets with a specific focus on group interactions, either between human-human, or human-agent groups. We provide dimensions upon which datasets for group interactions involving social agents could be characterized and classify existing datasets accordingly. Such a resource would be helpful for the research community to quickly identify data to work with,