- Zane Alumbaugh (zanedma@gmail.com)
- Farhan Saeed (fasaeed@ucsc.edu)
- Ilay Raz (ilraz@ucsc.edu)
- Nikhil Sheth (npsheth@ucsc.edu)
This was a final project for our Machine Learning class at UCSC. Given a movie review, the classifier attempts to detect the review sentiment (either positive or negative). The project was directly taken from this kaggle challenge however instead of using the "Bag of Words" approach we decided to explore neural network classifier options. Once we discovered the BERT SOTA classifier it was clearly a good candidate for our needs. The BERT classifier achieves around 93% accuracy on unlabled test data which the group was very happy. While we could have used a simpler Support Vector Machine and gotten around 89% accuracy, we wanted to expand our horizons and try to squeeze a bit more accuracy out of the classifier because there was a small competition bonus for this assignment.
* Note: The BERT_colab.ipynb was created and tested using Google Collaboratory and has not been tested elsewhere