A Deep Learning Approach to Detect Abusive Bengali Text

Abstract

Day by day, Social media sites, online news portals and blogs commenting sections are getting saturated with abusive contents in Bangladesh. Detecting different types of abusive contents in online will not only improve these websites discussion sections but will also ensure user's safety. In this paper, several machine Learning and deep learning based algorithms e.g. Linear Support Vector Classifier (LinearSVC), Logistic Regression (Logit), Multinomial Naïve Bayes (MNB), Random Forest (RF), Artificial Neural Network (ANN), Recurrent Neural Network (RNN) with a Long Short Term Memory (LSTM) cell have been tested to detect multi-type abusive Bengali text. Besides, there has been introduced new stemming rules for Bengali language which help to achieve better performance of algorithms. Deep learning based algorithm RNN outperforms other algorithms by gaining highest accuracy 82.20%.