Yelp Score Predictor

Read the full report.

The Yelp Rating Predictor is a logistic regression model trained using Yelp's academic dataset. The model is trained on the entire corpus of Yelp's provided review text, and attempts to predict the score rating of a particular business based on its reviews.

Model Architecture

The model consists of three fully connected layers, using tanh, sigmoid, and relu activation functions. The model is compiled using the adam optimizer. Feature extraction was performed over the entire body of Yelp's review text using TF-IDF vectorization. Due to hardware limitations, only 1000 features were able to be vectorized and one-hot encoded. The model achieves an RMSE score of 0.2978, and demonstrated that the model generalized the dataset fairly well.