An Ensemble Based Approach to Detect Hate Speech

In recent times, we see numerous hate speech incidents related to different domains like: gender, caste, religion, etc. The sudden outburst of users over the online forums has aggravated this issue. Given the relentless stream of content posted on online sites, the task of manually filtering hateful posts is a difficult task. Furthermore, in ML models there exist problems of generalization. In order to tackle this problem, we have proposed an ensemble model based on CNN+LSTM - a deep learning technique and RoBERTa - transformer based model. We have trained our model on three popular datasets namely HASOC, Davidson and HatEval and also on the combination of these three to get more generalized results. The outcome of the research indicates that the presented ensemble model gives better accuracy and F1 score on the combined data set as compared to the individual method.