Tools
Tools
Jose, Abin (2024) Enhancing Hate Speech Detection In Social Media using XLNet and Graph Convolutional Networks: Sentiment Analysis. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (514kB) | Preview |
Preview |
PDF (Configuration Manual)
Download (596kB) | Preview |
Abstract
Hate speech on social media has become a ubiquitous problem, causing real-life harm and upending online spaces. Such concerns lead to the exploration of advanced methods for accurately identifying hate speech, which often exhibits context-sensitive language, sarcasm, or coded phrases. We performed a lot of preprocessing such as removing noise, tokenizing text, doing sentiment analysis on datasets from Hatebase and Kaggle. We combine traditional machine learning models (Logistic Regression, SVM) with deep learning architectures (RNN, CNN) and transformer-based models (BERT, XLNet). To account for implicit hate speech, we introduced sentiment analysis, while graph convolutional networks facilitated the exploration of word relationships. Transformer models obtained higher performance under accuracy, F1-score, and ROC-AUC. Employing XLNet detected complex hate speech patterns, outperforming other approaches with an accuracy rate of 91%. This most recent research highlights the need for context-aware models and sentiment analysis to address hate speech. Yet the limitations of data and demands on computational resources remain hurdles. This work provides a valuable contribution to the field, proposing a strong framework that incorporates state-of-the-art methodologies, paving the way for further research, and practical use-cases to promote more secure online spaces.
Actions (login required)
 |
View Item |