MasterThesis-Lies-DeceptiveText

Social media plays a vital role in connecting people worldwide and fostering interpersonal connections. The vast reach of social media platforms means that any information circulated through them carries the potential to influence a substantial global audience. Consequently, Individuals find themselves frequently encountering potentially deceptive content, Whether in the form of fabricated news, biased product appraisals, or lies regarding various events and activities. While addressing this challenge, Several studies have ventured into automated text-based deception detection using traditional Machine learning algorithms, Recurrent neural networks and recently introduced Transformer models. Recent breakthroughs in Natural language processing have introduced more deceptive content which is even undetectable at human level. The significant obstacle in this scenario is the inherent difficulty in deciphering and comprehending the changing underlying logic behind deceptive context. This thesis will focus on utilizing various Transformer models and fine-tuning them for the target objective of detecting misinformation and deceptive content. The thesis will explore different scenarios with utilization of Linear classifier for supervised learning and Generative Adversarial Network classifier for semi-supervised learning for deceptive statements and finally Question and Answer classifier to tackle lies and misinformation content. These approaches will capitalize on the versatility and robustness of Transformer models, Enabling a comprehensive exploration of their potential in addressing various challenges related to deceptive content.