Sarcasm Detection

This is a university project to classify sarcasm on the SARC reddit dataset. Sarcasm is very difficult to classify as it is very dependent on the context of the speaker as well as text. We implement a BERT, SVM, Random Forest and CNN model, each in a respective jupyter notebook. Our best model is an augmented CNN that achieves a mean accuracy of 73.1% over 5 folds (see report).

Getting Started

1. Install Python 3.9
2. Install dependencies with pip install -r requirements.txt
3. Run a respective jupyter notebook (*.ipynb)