A dataset of abusive comments on the Nigerian web

Abstract

As the online comments keep growing in an exponential manner, the need for effective filtering or detecting abusive language in generated web textual content becomes increasingly important. In addition, the geometric increase in the textual content of the web has made the use of methods like word-bag and pattern matching for detection of hate speech less effective. The purpose of this work is to develop a corpus of online users' comments annotated for abusive language and develop a classification machine learning based model to detect hate speech on the Nigerian web. The model is evaluated on real-time comments and in different settings to further increase the learning rate of our model.