Hybrid AI framework for detecting deep-fake tweets on social media

Deepfake technology is developing quickly, which brings both potential and concerns, especially on social media where misleading content can sway public opinion and disseminate false information. Deepfake tweets are dangerous because they spread misleading information and sway online debates. They are made to look like real individuals. This paper suggests a Hybrid AI Framework that successfully detects deepfake tweets by combining sentiment analysis, feature engineering, transfer learning, and ensemble learning. In order to improve robustness, the framework uses a majority voting classifier in conjunction with Random Forest for classification and BERT (Bidirectional Encoder Representations from Transformers) for contextual feature extraction. This method enhances interpretability and detection accuracy by integrating linguistic, semantic, and environmental data. The study shows that the suggested model outperforms more conventional classifiers like Decision Tree, SVM, and LSTM by evaluating it using common performance criteria. By offering a scalable, effective, and interpretable method for detecting deepfake tweets, this research helps fight disinformation by enhancing the legitimacy of digital platforms and encouraging reliable online conversation.