Introducing an Abusive Language Classification Framework for Telegram to Investigate the German Hater Community

Abstract

Because traditional social media platforms continue to ban actors spreading hate speech or other forms of abusive languages (a process known as deplatforming), these actors migrate to alternative platforms that do not moderate user content to the same degree. One popular platform relevant for the German community is Telegram for which limited research efforts have been made so far. This study aimed to develop a broad framework comprising (i) an abusive language classification model for German Telegram messages and (ii) a classification model for the hatefulness of Telegram channels. For the first part, we use existing abusive language datasets containing posts from other platforms to develop our classification models. For the channel classification model, we develop a method that combines channel-specific content information collected from a topic model with a social graph to predict the hatefulness of channels. Furthermore, we complement these two approaches for hate speech detection with insightful results on the evolution of German speaking communities focused on hateful content on the Telegram platform. We also propose methods for conducting scalable network analyses for social media platforms to the hate speech research community. As an additional output of this study, we provide an annotated abusive language dataset containing 1,149 annotated Telegram messages.

Mapping a Dark Space: Challenges in Sampling and Classifying Non- Institutionalized Actors on Telegram

“…Edge-based inclusion criteria differentiate between the types of edges accepted in the unknown underlying network of Telegram entities. Sampling designs considering all kinds of references (e.g., forwards, @-mentions, invite links) as viable network edges (Bovet & Grindrod, 2022;Wich et al, 2022) necessarily discover a different network structure than those that only consider one such reference type (Hoseini et al, 2021;Peeters & Willaert, 2022) or invite links (Curley et al, 2022). Consequently, implicit entity selection needs to be considered, as forwarded messages only refer to the message's sender (either a channel or an individual user); thus, public chat groups cannot be detected if invite links and @-mentions are excluded.…”

“…Curley et al (2022) ensured the representativity of the discovered channels using a review by two experts. Wich et al (2022) filtered channels by language to exclude non-German channels. Urman and Katz (2022b) filtered by word occurrence at the message level to ensure relevance.…”

Toxic comments are associated with reduced activity of volunteer editors on Wikipedia

“…Toxicity has been extensively studied on popular social media websites such as Twitter ( 15 , 16 ), Reddit ( 17 , 18 ), and similar platforms ( 19 , 20 ). However, much of these research focuses on automated toxicity detection and prevalence estimation rather than on evaluating its impact ( 21 ).…”