Social media has become an indispensable part of daily life for many youngsters. The widespread presence of hate speech, image and videos on these platforms has led to interest in developing automated cyberbullying detection systems. Traditional text-based detection approaches fail to capture multimodal nature of online interactions, where harmful content may be expressed through a combination of text, images, memes, audio, and video. Recent studies have integrated information from multimodal data to improve the detection performance. This work provides an overview of eight recent AI-driven approaches, covering homogeneous and heterogenous graph neural networks along with their variants. The models are designed to handle various tasks, including aggressive message detection, message type classification, bullying session detection, and victim-perpetrator target detection. The application domains of different graph models are compared and summarized. In addition, a unified framework for developing an effective multimodal detection system is presented. Overall, this work aims to help readers identify suitable models for addressing different aspects of cyberbullying leverage multimodal data.

Authors:
Chun-fai Chu, The Hang Seng University of Hong Kong, Hong Kong
Chun-ho Tong, The Hang Seng University of Hong Kong, Hong Kong
Po-kin Chan, The Hang Seng University of Hong Kong, Hong Kong

About the Presenter(s)
Dr Carlin Chu is a University Assistant Professor/Lecturer at The Hang Seng University of Hong Kong in Hong Kong

Connect on Linkedin
https://hk.linkedin.com/in/chun-fai-carlin-chu-354263a2

See this presentation on the full schedule – Monday Schedule

Bookmark