Predicting star ratings based on annotated reviews of mobile apps

Abstract

This paper presents and evaluates different computational models for review rating prediction. The models rely solely on star ratings from an annotated corpus of customer reviews of mobile apps that were collected from the Google Play Store in a related work. Fine-granular opinions and the classification of their sentiment orientation were already available. The models build upon them to make predictions based on their polarity. Predicting star ratings is of importance to the sentiment analysis community because it can better be understood how customers subjectively rate products. Rating them consistently with corresponding written reviews, however, remains a difficult task for automated predictors. This paper sheds new light in that direction.