HIIwiStJS at GermEval-2018: Integrating Linguistic Features in a Neural Network for the Identification of Offensive Language in Microposts

This paper describes our submission for the GermEval-2018 shared task on the identification of offensive language. We use neural networks for both subtasks: Task I — Binary classification and Task II— Finegrained classification. We comparatively evaluate the use of typical textual features with extensions also considering metadata and linguistic features on the given set of German tweets. Our final system reaches 73.69% macro-average F1-score in a crossvalidation evaluation for the binary classification task. Our best performing model for the fine-grained classification reaches an macro-average F1-score of 43.24%. Furthermore, we propose methods to include linguistic features into the neural network. Our submitted runs in the shared task are: HIIwiStJS coarse [1-3].txt for Task I and HIIwiStJS fine [1-3].txt for Task II.