Detecting violent texts on social networks using machine learningalgorithms and lexical features

Social interaction is facilitated by any online environment that leads to increased antisocialbehavior. Incidents of cyberbullying, trolling and hate speech have increased significantly acrossthe globe. Recognizing hate and aggression has become an important part of cyberbullying.Cyberbullying refers to aggressive behavior by making rude, abusive, insulting, hateful andmocking comments to harm other people on social media. Human moderation is slow and costly,and even in rapidly growing data, only automated detection can stop trolling. In this study, weaddressed the challenge of automatic detection of violence in social networks. We appliedmultilayer perceptron by feeding important hand-engineered features and also experimented onadvanced combination of CNN-LSTM and CNN-BiLSTM in deep neural network. The results ofthis study showed ۹۱% accuracy for detecting verbal violence.