After finding a model that performed reasonably well, I decided to integrate it into an app using Streamlit and then deploy it via Heroku. Although the BERT model achieved the highest accuracy, the model size was much larger than similar performing models, thus, in order to speed up inference, I opted to use the Stacking Classifier model which had similar accuracy despite being much smaller in size. The deployed app allows users to enter a review and receive a predicted rating for it.