A Restaurants Recommendation System: Improving Rating Predictions Using Sentiment Analysis

Abstract

Most users often find themselves in a situation when they need to answer the same question: which product, movie, vacation offer, restaurant or book is the best to choose? In order to answer this question, Recommendation Systems have been developed to generate the best suggestions according to the user's interest. Recommendation Systems play an important role in the user's decision-making process by enriching its experience and satisfaction, considering his peers' actions and preferences. The goal of this paper is to enhance the recommendation process by applying Sentiment Analysis techniques on the input data. Sentiment Analysis is a domain that focuses on classifying information into positive, negative and neutral opinions. The results of a Sentiment Analysis task can be used to define social tendencies, items' popularity and adapting the services for users' needs. The proposed approach combines Sentiment Analysis and Recommendation Systems for defining the best suggestions for a user. Sentiment Analysis is applied for classifying restaurants' text-based reviews into positive and negative. The output of the Sentiment Analysis task is passed to a recommendation system that, using the collaborative filtering algorithm, will predict the rating for a not-visited restaurant and generate a list of top-n restaurants for the user. This approach outperformed the results obtained when the Sentiment Analysis step was not considered in the recommendation process. Therefore, the proposed system increases the accuracy of the recommended items by analyzing, from a sentiment point of view, the text-based reviews offered by users.