Fine Tuning Large Language Models for Accurate Fake News Detection

In the current digital age, false information spreads quickly via social media and other online channels, making it challenging to tell real news from fake. This study introduces a framework for detecting fake news that uses Llama 3 for explanation-based classification in conjunction with conventional machine learning models. Using TF-IDF vectorization and SMOTE for improved balance and feature representation, we experimented with LSTM, SVM, Naive Bayes, Decision Trees, CNN, and Logistic Regression. Because of its capacity for sequential learning, Decision Tree demonstrated impressive results, and Llama 3 produced explanations in natural language to elucidate predictions. A speech-to-text module was also developed, enabling users to speak news statements that are converted into text and analyzed for authenticity.In addition to identifying false information, the framework offers context-aware reasoning, which enhances usability and trust in practical applications.