Telugu Hate Comments Classification

This work studies various approaches to classifying hate speech in Telugu Text, using a dataset containing Telugu and English characters. Three pre-processing approaches were considered – Direct, Transliteration, and Transliteration (which is present in the preprocessing files folder). The pre-processed data form each approach was used independently to train ten models, broadly classified under three approaches – Machine Learning, Deep Learning, and Ensemble Learning. It was found that the transliteration pre-processing approach yielded better classification performance, with the single cell Bi-LSTM model most accurately performing the given classification task