Automatic Algerian offensive language detection in social media networks

Abstract

This research focuses on the detection of offensive and abusive content in Facebook comments in Algerian dialect Arabic, we created a corpus from scratch with over 8.7k texts written in both Arabic and Roman scripts, and annotated with three categories offensive, abusive or normal. We performed a series of automatic classification tests with state-of-the-art algorithms. In addition to rule-based classification with an identification algorithm. The results showed an acceptable performance, but the identification algorithm can still be improved with further investigation.