To search, Click below search items.


All Published Papers Search Service


Performance comparison of Deep Learning architectures for Cyberbullying detection on Multi-modal data


Subbaraju Percherla and E Ilavarasan


Vol. 22  No. 11  pp. 681-691


Cyberbullying drastically increased with the increase of Internet and Social media networks from the last one decade. The various forms of bullying are increasing with the increase of smart phones. Now a days bullies are targeting the victims not only through text, but they may also send images, videos, graphics, emojis. In this paper, we compared various types of deep learning architectures for cyberbullying detection on multi-modal data. The selected framework could be able to handle bullying detection for text and image combinational data. We mainly focused on extracting (computing) image embeddings using advanced architectures such as Inception, VGG19, ResNet, Xception, MobileNet, Desnet and EfficentNet. We used RoBERTa deep learning architecture to generate word embeddings from the text data. We have employed various machine learning classifiers such as Logistic regression (LR), XGBoost(XG), LightGBM(GBM), Decision Tree(DT), Bernoulli Navie Bayes(BNB) and Gaussian Navie Bayes(GNB) to classify bullying and non-bullying messages. The experiments were conducted on 2100 samples of combined data of text and image. The XCeption and LightGBM classifier combination performed well as compared to other combinations of deep networks and classifiers.


Cyberbullying, Social Networks, Natural Language Processing, RoBERTa, EfficientNet.