DEEP LEARNING PLUGIN TOOL USING AN HYBRID RNN-LSTM MODEL TO DETECT AESOPIAN PHRASES IN CYBERBULLYING

Abstract

Cyberbullying is a major issue on social media that can cause significant harm to children and teenagers' mental and physical health, in some cases leading to self-harm or suicide. Existing cyberbullying detection systems primarily rely on content moderators built and managed by social media platforms. However, they do not consider complex language features such as slang, multilingualism, or aesopian phrases, which are essential in detecting cyberbullying accurately. This research proposes the world's first AI-based plug-in tool that uses a hybrid RNN-LSTM neural network to automatically detect, anticipate, and classify cyberbullying/prospective incidents while also recognizing false positives. The tool considers patterns across natural language, slang terms, and aesopian phrases that are unique to select groups or communities. The proposed tool can work across languages, browsers, devices, and community-specific phrases to protect children worldwide. The study demonstrates that deep learning techniques can be effectively used for cyberbullying detection, and the proposed tool can have the potential to save over a billion kids on the internet and social media.