Cyberbully Detection Using 1D-CNN and LSTM

摘要

This paper proposes deep learning-based solutions for cyberbullying issue that is becoming increasingly common in the modern era of social media and digital connect. The early detection and identification of such events can curb the menace of this unethical practice. Toward this, 1D-CNN, LSTM and bidirectional LSTM (BiLSTM) networks are utilized in this work that are able to detect and classify texts into six different cyberbully classes. The dataset used in our training and testing procedure contains 159k input examples comprising a variety of texts belonging to both non-bullying and bullying sentiments. Our results show that the proposed models achieve an overall test accuracy of 0.9633, 0.9412 and 0.9745 using 1D-CNN, LSTM and BiLSTM networks, respectively, thereby making BiLSTM a suitable network for cyberbully detection purpose.