Abuse Detection of Cyberbullying in Tweets Data Using Ensemble DL Model with Advanced Voting Mechanisms

• Published in: 2024 2nd World Conference on Communication & Computing (WCONF) pp. 1 - 6
• Main Authors: R, Mahaveerakannan, Anitha, Cuddapah, Dhar M S, Murali
• Format: Conference Proceeding
• Language: English
• Published: IEEE 12.07.2024
• Subjects: Abusive Words detection; Classification algorithms; Cyberbullying; Cyberbullying detection; Data models; Dynamic Voting; Ensemble learning; Feature extraction; Hate speech; Hybrid Deep Learning; Machine learning algorithms; Modified seagull optimization algorithm; Optimization; Reinforcement learning; Transformers
• DOI: 10.1109/WCONF61366.2024.10692193

In social media, automatic identification of abusive online content- speech, profanity, threats, etc.-has become commonplace. Several initiatives have been devoted to identifying this issue in English. Nonetheless, it is not an easy task to identify hate speech and abuse in languages with limited resources. Transformer-based multilingual pre-trained language models for typologically varied languages are not always efficient due to insufficient labelled data in low-resource languages besides variable generalisation capacity. In order to tackle these problems, the research project compares a dynamic voting classifier in several languages with classic machine learning and deep learning classifiers as part of a hybrid ensemble method for cyberbullying detection. The modified seagull optimisation algorithm (MSOA) selects the most relevant features in the best possible way. The dataset underwent exploratory data analysis prior to implementation in order to obtain more insightful information. The study examines the offensive language detection of several publicly accessible datasets, including 6 datasets in 6 languages, in order to our approach. The studies demonstrate that in most circumstances, suggested deep learning-based representations perform better than transfer learning-based representations.