Bullying Detection Solution for GIFs Using a Deep Learning Approach

Abstract

Nowadays, technology allows people to connect and communicate with each other even from miles away, no matter the distance. With the increased use of social networks that were rapidly adopted in human beings’ lives, they can chat and share different media files. While the intent for which they have been created may be positive, they can be abused and utilized in a negative way. One form in which they can be maliciously used is represented by cyberbullying. This is a form of bullying where an aggressor shares, posts, or sends false, harmful, or negative content about someone else by electronic means. In this paper, we propose a solution for bullying detection in GIFs. We employ a hybrid architecture that comprises a Convolutional Neural Network (CNN) and three Recurrent Neural Networks (RNNs). For the feature extractor, we used the DenseNet-121 model that was pre-trained on the ImageNet-1k dataset. The obtained results give an accuracy of 99%.