Fake News Detection
Classifying news articles as real or fake using deep learning, tested on three public datasets.
What
A grad-school NLP project that tries to separate real news from fabricated news using deep learning. The question was less “can a model do it” and more “how does it fail when you change the dataset under it”.
Approach
Three public datasets of varying provenance. Tokenize, embed, train a few architectures: baseline LSTM, a CNN over word embeddings, and a transformer fine-tune. Held out one dataset at a time to see how models trained on one distribution fared on another.
Stack
Python, PyTorch, TensorFlow, scikit-learn, Jupyter.