AN ENHANCED FAKE NEWS DETECTION SYSTEM EMPLOYING FUZZY DEEP LEARNING

Because fact-checking procedures are inherently imprecise and nuanced, detecting fake news is still a crucial and difficult task in the field of information verification. The growing amount and speed of false information calls for automated systems that can efficiently handle this uncertainty, as fact-checking has historically relied on the experience of qualified fact-checkers. In this study, we provide a novel deep learning framework based on Long Short-Term Memory (LSTM) that is specifically designed to handle the intricacies and ambiguities inherent in the identification of fake news. Our strategy greatly enhances classification performance by taking advantage of LSTM's capacity to grasp contextual nuances and represent long-range dependencies in textual data. With a remarkable accuracy of 99%, we test the suggested model on the popular LIAR dataset, a benchmark that presents significant difficulties because of its varied and complex claims. We propose LIAR2, a freshly curated benchmark dataset enhanced with further annotations and insights from the academic community, to address the shortcomings of the LIAR dataset, such as its size and representativeness, and better depict the complex nature of fake news. We give a deeper knowledge of the dataset characteristics that affect model efficiency and create new baseline results for LIAR2 through comprehensive experimental research, including ablation analyses and comparative performance evaluations on both LIAR and LIAR2. Our results highlight the capacity of LSTM architectures to manage uncertainty in textual untruth detection and provide insightful direction for further studies targeted at enhancing automated fact-checking systems and boosting their dependability in practical settings.