Comprehensive Cyberbullying Detection System : Integrating Text and Image Analysis with Rojak Language

Chan, Wen Ping (2025) Comprehensive Cyberbullying Detection System : Integrating Text and Image Analysis with Rojak Language. Final Year Project (Bachelor), Tunku Abdul Rahman University of Technology and Management.

Text RDS_Chan Wen Ping_Fulltext.pdf Restricted to Registered users only Download (5MB)

Abstract

This project presents a hybrid cyberbullying detection system that utilizes machine learning to identify harmful content in both text and images, specifically tailored to Malaysia’s “rojak” language and meme-based communication. By integrating Recurrent Neural Network (RNN) for multilingual text analysis and a Convolutional Neural Network (CNN) for image classification, the system effectively captures cyberbullying across different content types. A robust preprocessing pipeline was developed to handle text normalization, tokenization, lemmatization, and image OCR extraction. The system is structured using a modular, layered architecture to enhance scalability and maintainability. It features a web-based interface built with HTML, CSS, JavaScript, and Flask, allowing users to submit content for real-time analysis. The machine learning models, developed using TensorFlow, Keras, and Jupyter Notebook, were evaluated with accuracy, precision, recall, and F1-score, achieving reliable performance despite challenges like class imbalance. A feedback and flagging mechanism is included to support moderation and user transparency. The system was successfully deployed and tested locally, confirming its feasibility and effectiveness in detecting cyberbullying in a multilingual, multimedia context. This work offers a comprehensive solution that addresses the complexity of online harm in Malaysia’s digital space and lays the groundwork for future improvements and large-scale deployment.