Краткое описание

This paper discusses the problem of assessing aggressive behavior, presented on videos that are publicly available on the Internet, of users of the sociocyberphysical systems in various subject situations: a game video stream and a conversational and entertainment video stream. This work presents the technique of preliminary selection of video samples containing manifestations of aggression in various situations. This technique is based on the naturalistic behavior observation approach. It allows to evaluate the presence of aggressive behavior in the video and to identify the type of aggression, as well as additional behavioral, emotional and other manifestations that accompany the aggressive behavior of users. An assessment based on the analysis of contingency tables of the difference in the types of aggression shown in the situations under consideration is made. The assessment of the difference between the accompanying emotional and behavioral manifestations in different types of aggression is performed. The differences in specific aggressive actions for the considered object situations are also highlighted. The results obtained make it possible to clarify the criteria for assessing aggressive behavior on video for various subject situations, which, in turn, makes it possible to clarify the methodology for marking up video datasets for machine learning systems.