Overview

• The goal of this project was to predict business’ star ratings based on the given customer business reviews in the Yelp dataset. The reviews were converted to a TF-IDF Matrix which was then given to a neural network.
• Used different structures of neural networks such as activation functions, and optimizers to discover a model that best predicts the star ratings of the businesses.

Methodology:

• Created a dataframe from the JSON file, review.json, containing all the reviews and their associated business ids.
• Created a TF_IDF vectorizer from the reviews and converted this into a dataframe
• A maximum document frequency of 95% was used to eliminate stop words
• Created different neural network models with different layers and early stopping for training.

Result:

• RMSE for each model