Multimodal Cyberbullying Detection Using Deep Learning Techniques: A Review

Abstract

The rise of social networks and online communication, facilitated by internet accessibility and modern technology, has brought numerous benefits. However, it has also given rise to cyberbullying, a harmful phenomenon characterized by disclosing private information and posting hostile content to shame individuals. The repercussions of cyberbullying are severe, impacting victims’ mental health, social lives, and personalities. Given the vast daily data uploads on social media, there is a pressing need for automated cyberbullying detection tools. This paper conducts a review of research in cyberbullying detection, encompassing both traditional machine learning and deep learning studies, spanning unimodal and multimodal approaches. The search involved major academic digital libraries like ACM Digital Library, IEEE Xplore Digital Library, and Springer Link, yielding 250 research articles. A selection process and redundancy checks followed, narrowing down the articles to 45 based on specific criteria: publication between 2019 and 2023, a focus on cyberbullying detection and related online risks like hate speech, use of English language data, and the development or introduction of cyberbullying detection algorithms. The significant contributions of the retained articles were identified, alongside future research directions. The paper also provides summaries of the datasets and algorithms employed. It concludes by highlighting ongoing challenges in the field to be addressed in the future.