This paper describes our approach (ur-iw-hnt) for the Shared Task of GermEval2021 to identify toxic, engaging, and fact-claiming comments. We submitted three runs using an ensembling strategy by majority (hard) voting with multiple different BERT models of three different types: German-based, Twitter-based, and multilingual models. All ensemble models outperform single models, while BERTweet is the winner of all individual models in every subtask. Twitter-based models perform better than GermanBERT models, and multilingual models perform worse but by a small margin.

This paper presents the contribution of the Data Science Kitchen at GermEval 2021 shared task on the identification of toxic, engaging, and fact-claiming comments. The task aims at extending the identification of offensive language, by including additional subtasks that identify comments which should be prioritized for fact-checking by moderators and community managers. Our contribution focuses on a feature-engineering approach with a conventional classification backend. We combine semantic and writing style embeddings derived from pre-trained deep neural networks with additional numerical features, specifically designed for this task. Ensembles of Logistic Regression classifiers and Support Vector Machines are used to derive predictions for each subtask via a majority voting scheme. Our best submission achieved macro-averaged F1-scores of 66.8%, 69.9% and 72.5% for the identification of toxic, engaging, and fact-claiming comments.

At the same time, social media sites have 2020). It is well-known that training large neural increasingly become more prone to offensive content transformer models often result in long processing (Hada et al., 2021; Zhu and Bhat, 2021; Bucur times. As GermEval-2021 features three related et al., 2021). As such, identifying the toxic language tasks, from a performance standpoint, we pose that in social media is a topic that has gained, training a model jointly on three tasks is likely to be and continues to gain traction. Research surrounding computationally more efficient than training three the problem of offensive content has centered models in isolation. Moreover, as GermEval-2021 around the application of computational models provides a single dataset for the three tasks, MTL that can identify various forms of negative content can also be used to help improving performance such as hate speech (Malmasi and Zampieri, 2018; across tasks. As such, we introduce multitask learning Nozza, 2021), abuse (Corazza et al., 2020), aggression whereby one model can predict all three tasks (Kumar et al., 2018, 2020), and cyber-bullying as an alternative approach.