Abstract
- Social media is not only a channel for the appropriate sharing of personal opinions and enriching discussions, but also facilitates the dissemination of inappropriate and aggressive statements. Those are especially concerning when they actively incite harmful actions such as violence or attacks on the government. Against this background, this paper presents the Ger-mEval Shared Task for Harmful Content Detection , which addresses three subtasks that have been largely neglected in previous competitions and research projects: the detection of 1) calls to action, 2) attacks against the liberal democratic basic order and 3) violence-supporting statements. For this pilot task, 11,551 tweets from a German Twitter network belonging to an extremist group were annotated. A total of eleven teams participated in at least one of the three subtasks, with nine teams submitting a system paper. Overall, macro-average F 1-scores of up to 87% were achieved for subtasks 1 and 3 and up to 71% for subtask 2. Content warning: We show illustrative examples of harmful content.