Publication

Abstract : With the world moving towards an almost digital lifestyle, especially after the covid-19 pandemic, people are extensively using social media platforms to express their views. The comments and chats on these platforms are often observed to be raw and direct, to the extent that it can become an act of bullying. Increase in suicide attempts, anxiety and depression cases reported on account of cyberbullying highlights the importance of developing automated systems for identifying cyberbullying oriented content. The proposed work deals with the classification of comments on social media oriented as cyberbullying or not. The work explores various deep learning classifiers in combination with varied word embedding inputs on five different social media datasets for classification. The model evaluation reported that Glove embeddings when input to an attention based Bi-LSTM classifier with GRU was the best performer across most of the experimented datasets with a micro-average F-measure ranging from 0.8 to 0.97.