Aggression detection through deep neural model on Twitter

摘要

Social interaction is being facilitated by every online environment that results in rise of antisocial behavior. Incidents like cyberbullying, trolling and hate speech have been grown significantly across the globe. Hate and aggression detection had become a crucial part of cyberbullying and cyberharassment. Cyberbullying refers to aggressive behavior with rude, offensive, insulting, hateful and teasing comments to harm other individuals on social media. Human moderation is slow and expensive and even not feasible in speedily growing data, only automatic detection can lead to put a stop on trolling. In this paper we address the challenge of automatic identification of aggression detection on tweets of cyber-troll dataset. We deploy Multilayer Perceptron by feeding important manually engineered features and also experimented on state-of-the-art combination of CNN-LSTM and CNN-BiLSTM in deep neural network, both perform well. Statistical results proved that our proposed model performed best and detects aggressive behavior with 92% accuracy.