In both, human-human interactions and human-machine interactions between several partners, communication strategies and paradigms need to be used to achieve a successful communication. In this context, anticipation of human behaviour is a challenging issue but is also an option to improve the communication. This manuscript aims for a better understanding of automatic analyses of interaction partners in human-machine interactions, especially in terms of dispositions, as well as a handling of classification approaches used for these investigations. For this, ideas and methods, which increase the ability to anticipate an interlocutor in an interaction, are discussed in relation to interaction at large, action modalities' analysis, and recognitions in depth. These aspects as well as utilised methods and approaches are further considered in a group perspective, which is currently in the focus of (automatic) interaction analyses. Finally, the manuscript discusses novel, neural-based recognition approaches and, in addition, data generation methods are emphasised. Given the research, results, and discussions presented in the manuscript, a roadmap for further investigations towards group interactions is developed.