Tracking and Detecting Fake News and Deepfakes in social media

Abstract

The spread of fake news and deepfakes on social media platforms has become a significant issue affecting public opinion and decision-making. Recent advances in artificial intelligence, particularly deep learning architectures, have shown promising results in addressing this problem. This report examines the effectiveness of machine learning, deep learning, and pre-trained models, including BERT, RoBERTa, FakeBERT, LSTM, Bi-LSTM, GRU, Bi-GRU, SVM, LR, and RF, for detecting fake news, and the Gemini API for deepfake detection on ISOT, AFND, and AFND TweepFake datasets. The proposed approach leverages the strengths of these models to enhance accuracy in detecting fake news and deepfakes. Evaluation metrics such as precision, recall, and F1 score are used to assess performance, where RoBERTa and FakeBERT achieved the best results for ISOT with an accuracy of 99.98%, and BERT achieved the best result for AFND with an accuracy of 92.17%.