Inferring User Preferences Using Reviews for Rating Prediction

Abstract

Nowadays, E-commerce websites have been developed and are very popular among online users. Users deal with the problems to choose the right items that meet with their needs. Recommender systems try to suggest the right items to the user by applying different recommendation approaches. Collaborative Filtering recommendation (CF) approach makes recommendations to users using the user-item matrix which has the ratings on each item given by users. Data sparsity problems may occur when a user-item matrix is built based on the ratings of users (one to five stars). User reviews on the products contain more information and opinions than user ratings. This paper proposes the rating prediction approach that infers the user preferences from textual reviews of hotels by performing sentiment analysis. The preference scores obtained from the sentiment analysis are integrated into the rating prediction process which applies two approaches named memory-based CF and model-based CF. The performance of the proposed system is tested on the Myanmar hotel reviews which are crawled from TripAdvisor site and hotel reviews which are downloaded from the Kaggle site. The resulted rating prediction accuracy of two approaches on two data sets is compared by using Root Mean Square Error (RMSE).