Predicting User Ratings Using Status Models on Amazon.com

1. ABSTRACT Amazon is the world’s largest online retailers where millions of customers purchase and review products on its website. However, many Amazon customers do not review and rate products after the purchase, and tons of research projects work on producing better prediction strategies for customer ratings. In this research project, we show a new approach to enhance the accuracy of the rating prediction by using machine learning methods that learn from a graph based on user status, defined by features such as the helpfulness votes and total votes received by the users. We discussed how the training performance of our model changes as we change the training method, the dataset used for training and the features used in the model. We compared this graphical model with another non-graphic model that is also based on customers’s review and status with different settings, and we achieved over 95% prediction accuracy using our graphical status model.