Fake News Detection Based on a Bidirectional GRU Deep Learning Modelx

In a world where misinformation spreads at lightning speed, being able to spot fake news has never been more important. This research proposes a Bidirectional Gated Recurrent Unit (BiGRU) deep learning model for classifying news articles as real or fake. The two publicly available datasets, True. csv and Fake. csv, were combined, preprocessed, and split up into test (25%) and train (75%) subsets. The sequential and contextual dependencies in the text are captured by training a BiGRU model on word embeddings. They tested its effectiveness against traditional machine learning models, specifically the Support Vector Machine (SVM) and K-Nearest Neighbors (KNN). The BiGRU model is far better than the earlier models as it shows near perfect performance (≈ 99.88%) accuracy, precision, recall and F1-score on the test set. The results of further testing on a new external dataset (8,117 samples) indicated strong generalization accuracy of ~98.53% along with few misclassifications and no overfitting. The model's training dynamics suggest that early stopping is effective and learning is efficient. The results show the benefits of BiGRU in being able to understand the context of the text for fake news detection. Further studies also hint at the possibility of hybrid and multilingual extensions.