Cheap eats or fine diner? Discovering social media signals for restaurant price level prediction

Abstract

In this paper, we discover signals from user-generated contents about post-purchase customer experience on the Yelp platform for predicting restaurant price level. We combine business, textual and visual signals extracted from a large-scale dataset with reviews and photos, by per-forming topic modeling to identify thematic content related to customer perceived experience, as well as employing an aesthetics assessment model to evaluate visual characteristics. Our results show that social media signals from reviews and photos may significantly improve the model’s predictive power and help explain the differences in customer perceived value between budget restaurants and fine-dining restaurants.