Deep Learning for toxic comment classification

Last month, I had the pleasure to take part in the GermEval 2018 workshop on the identification of offensive comments in German language microposts. During my stay at KAUST, me and my supervisors took part in Task 1 (binary classification) of the shared task for German language classification.

For our participation, we decided to use Deep Learning techniques (namely LSTM and CNN models) for the classification of offensive microposts. As a result, our best model reached a F1 score of 69.15% (with 76.77% F1 score being the best result in the task). The paper is available under the following link for further details: Offensive Comment Classification on German Language Microposts

Personally, I consider the identificaton of offensive comments (and sentiment analysis in NLP in general) to be a highly intersting research topic to work on. It first catched my attention when I came accross the Conversation AI project by Google Jigsaw, and I am very eager to learn more about the recent developments in the NLP field (especially the idea to use transfer learning for NLP).

If you are interested in discussions and exchanging your thoughts on this topic, please drop me a line!