Cyberbullying detection on TikTok using a deep learning approach

• Published in: Scientific Bulletin. Series C, Electrical Engineering and Computer Science no. 1; p. 5
• Main Authors: Stoleriu, Razvan, Nascu, Andrei, Pop, Florin
• Format: Journal Article
• Language: English
• Published: Bucharest University Polytechnica of Bucharest 01.01.2025
• Subjects: Artificial neural networks; Bullying; Deep learning; Feature maps; Internet; Machine learning
• ISSN: 2286-3540

Nowadays, the Internet has penetrated all aspects of the humans' lives. Despite providing many advantages and benefits, it can also produce a series of social problems such as spam, Internet crime, and Internet addiction. Unfortunately, negative information has become more common in modern society. In recent years, cyberbullying, one of the representatives of abnormal behavior on the Internet has been prominent in many countries across the globe. Different organizations and institutions tried to formulate and take some measures against it. In this paper, we propose a solution for cyberbullying detection in TikTok videos using a deep learning approach. We employ a Transformer-based model that operates on Convolutional Neural Network (CNN) feature maps. Moreover, we utilize the DenseNet121 model pre-trained on the ImageNet-1k dataset. We evaluated the accuracy of the model and observed we got an accuracy of 94.16%.