Cyberbully: Aggressive Tweets, Bully and Bully Target Profiling from Multilingual Indian Tweets

The present work proposes an end-to-end solution to identify a potential bully and bully-targets from multilingual aggressive tweets in Indian Twitter-o-sphere. The proposed work uses two LSTM based classifiers in pipeline to detect the tweet’s language and aggressiveness. The model was trained with over 150,000 tweets of Hindi, English, Bengali, and Hinglish languages. F1 scores achieved for English, Hindi, Bengali, and Hinglish are 0.73, 0.83, 0.69, and 0.91, respectively. The paper further reported the patterns identified for several different attributes such as followers count, friends count, frequencies of tweets, and percentage of aggressive tweets of such potential bully and target users.

Conference paper