Discussing the Value of Automatic Hate Speech Detection in Online Debates
Köffer Sebastian, Riehle Dennis M, Höhenberger Steffen, Becker Jörg
This study discusses the potential value of automatic analytics of German texts to detect hate speech. In the course of a preliminary study, we collected a dataset of user comments on news articles, focused on the refugee crisis in 2015/16. A crowdsourcing approach was used to label a subset of the data as hateful and non-hateful to be used as training and evaluation data. Furthermore, a vocabulary was created containing the words that are indicating hate and no hate. The best performing combination of feature groups was a Word2Vec approach and Extended 2-grams. Our study builds upon previous research for English texts and demonstrates its transferability to German. The paper discusses the results with respect to the potential for media organizations and considerations about moderation techniques and algorithmic transparency.
Natural Language Processing (NLP); Hate Speech; Text Analytics