Networks of Toxicity: SNA of Communication on Twitch.tv

This work was devoted to the study of communication practices of users of the streaming platform Twitch.tv, with a focus on toxic communication. Despite the fact that the problem of "toxicity" is a widely discussed phenomenon, a full-fledged definition of the phenomenon has not yet been given, due to the contextuality of the phenomena of "toxicity" and the difference in behavior expected from the individual. With the help of the collected chat logs, a thematic model was built. In consequence, topics from the resulting thematic model were assessed by experts on a toxicity scale. This allowed us to more correctly formulate the concept of toxic communication and evaluate the empirical indicators of this phenomenon. Subsequently, users were profiled, based on the definition obtained. Four groups of users were singled out: little "toxic", medium "toxic", strongly "toxic" and users with varying levels of toxicity. A statistically significant difference in the effect of channel characteristics on the change in the level of toxicity of the user was found. In the aftermath, the Valued Exponential Random Graph Model was built for a deeper assessment of the influence of channel characteristics on user preferences. This model allowed us to describe the system of stream co-views, and assess the impact of the environment (channel) toxicity on user choice.