Analyzing hate speech dynamics on Twitter/X: Insights from conversational data and the impact of user interaction patterns

Abstract

This paper investigates the pervasive issue of hate speech within Twitter/X Portuguese network conversations, offering a multifaceted analysis of its characteristics. This study utilizes a mixed-method approach, combining several methodologies of network analysis (triad census and participation shifts) over the network of interaction between users. Qualitative manual content annotation was applied to the dataset to dissect different patterns of hate speech on the platform. Key findings reveal that the number of users followed by an individual and potentially reads is a relevant predictor for a user's propensity to post aggressive content. We concluded also that during a conversation thread, hate speech happens significantly more within the first 2 h of interaction. Transitivity of interactions and individual expression are considerably lower as more hate speech is prevalent in conversations. Our research confirms that hate speech is usually expressed by external individuals who intrude into conversations. Conversely, the expression of hate speech of indirect type by third parties interfering in conversations is uncommon. We also found that counter-speech discourse is strongly correlated with a type of discourse that typically avoids conflict and is not privately held.