Improving Hotel Customer Sentiment Prediction by Fusing Review Titles and Contents

Abstract

The large volume of online customer reviews is a valuable source of information for potential customers when making decisions and for companies seeking to improve their products and services. While many researchers have focused on the content of reviews and their impact on customers’ opinions using deep learning approaches, the mechanism by which review titles and contents influence sentiment analysis (SA) has received inadequate attention. This study proposes a deep learning-based fusion method that reveals the importance of reviewing titles and contents in predicting customer opinions. Our experiments on a crawled TripAdvisor dataset showed that the performance of the document-level SA task could be improved by 2.68% to 12.36% compared to baseline methods by effectively fusing informative review titles and contents.