ARCHIVES

With the rapid proliferation of social media, fake news has been spread more rapidly and detected less efficiently in the era of social media. Traditional machine learning techniques are often limited in their sophistication and encoding of the semantic and linguistic depth that language carries with it, which calls for more sophisticated NLP. This work presents an AI-based platform for fake news classification and sentiment analysis based on Transformer models (BERT, DistilBERT) combined with Recurrent Neural Networks (LSTM, Bi-LSTM). We test the framework on standard benchmark data sets such as Kaggle's Fake News Dataset and Twitter Sentiment datasets, considering preprocessing steps (e.g., tokenization, stop-word removal, Word2Vec, or GloVe). Experimental findings indicate that Transformer models excel at capturing contextual relationships. LSTM variants, on the other hand, reach similar accuracy while needing less computing power. Furthermore, sentiment analysis is used to assess the tone of news content. This provides insights into the emotional aspect and potential bias tied to misinformation. The proposed framework not only makes fake news detection more reliable but also improves our understanding of social media sentiment. This has consequences for media monitoring, policy-making, and building digital trust.