RuG at GermEval: Detecting Offensive Speech in German Social Media

This paper reports on the systems the RuG Team submitted to the GermEval 2018 - Shared Task on the Identification of Offen- sive Language in German tweets. We sub- mitted three systems to Task 1, targeting the problem as a binary classification task, and only one system for Task 2, address- ing a fine-grained classification of offen- sive tweets in different categories. Prelim- inary evaluation of the systems has been conducted on a fixed validation set from the training data. The best macro-F1 score for Task 1, binary classification, is 75.45, obtained by an ensemble model composed by a Linear SVM, a CNN, and a Logistic Regressor as a meta-classifier. As for Task 2, multi-class classification, we obtained a macro-F1 of 40.75 using a multi-class Linear SVM.