Fake News Detection

Misinformation spreads faster than corrections. This model learns what fake news language looks like from raw text instead of hand-made features.

I built a deep learning classifier for fake news trained on more than 20,000 labelled articles. The README compares five approaches, a TF-IDF Naive Bayes baseline, LSTM, GRU, CNN and a Bidirectional LSTM, and the BiLSTM did best at roughly 99% accuracy.

The final model tokenizes and pads text, uses pretrained GloVe embeddings, then a 128-unit bidirectional LSTM with dropout and a sigmoid output. Reading both directions helps with cases like “not good” versus “good”.