Using approximately 24,000 AirBnB reviews for 7,500 listings & feature vectors such as bedrooms, neighbourhood, response ratings, etc. I created a Linear Regression model with polynomial features & a K-NN model to predict AirBnB ratings. Bag-of-words with one-hot-encoding using Term Frequency-Inverse Document Frequency is used to quantify reviews.

Both models marginally beat a baseline predictor which predicts the most frequent rating for each listing and the the linear regression model is found to be superior over the kNN model.

Furthermore, the feasibility of predicting individual listings' ratings, such as Location score, Communication score, and Value score, is evaluated. It has been discovered that the location score can be effectively predicted using a kNN model with the longitude and latitude coordinates of the listing. The value score can be predicted using a linear model with the varying amenities as feature vectors. Additionally, the communication score can be predicted based on the owner's response rating.