RNN and CNN Deep Neural Models for Hate Speech Classification

Abstract: The importance of creating a social media environment void of abuse, bullying and threats cannot be overemphasized. Using Twitter as a case study, this project intends to employ current technology to perform hate speech detection on social media. I propose a classification method that combines an embedding algorithm with Convolution Neural Networks (CNN) and Recurrent Neural Networks (RNN), such as Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Multichannel combinations of these networks, using a dataset from Twitter for the training, validation and testing of the neural network. This study implemented nine deep neural models with different architectures, and the multichannel RCNN model performed best among the other models.

Keywords: CNN, C-LSTM, GRU, LSTM, NLP, RNN, RCNN, ReLU.