Multi-featured Cyberbullying Detection Based on Deep Learning

Cyberbullying is a common and harmful social problem. The existing cyberbullying detections used traditional machine learning and deep learning algorithms to extract text features and then classify users’ emotions, but the classification results can hardly meet expectations. This paper proposes a BiGRU-CNN sentiment classification model (GCA: BiGRU+CNN+ATTETION) for cyberbullying detection, which consists of BiGRU layer, attention mechanism layer, CNN layer, full connection layer and classification layer. BiGRU is a bidirectional GRU layer that has the function of combining context to gain the global features. With the join of CNN layer based on a convolution function with 128 kernels of length 5, it can not only obtain the local features more comprehensively, but also improve the learning rate of the model greatly. The attention mechanism layer can grasp representative words and assign weight to words better. We use the text data set from Kaggle and the emoji data set crawled from social networks to train and test the model. The results show that the classification accuracy of the model is better than that of the traditional model, which reaches 91.07%.