More than words: Understanding how valence and content affect review value

This paper combines sentiment analysis with latent Dirichlet allocation model to study how review valence and content categories (i.e., the specific way in which a review describing positive or negative sentiment) affect its value. We find five categories of customer complaints toward hotels, namely, facility complaints, environment complaints, food complaints, service complaints and money complaints. Econometric analyses show that reviews with the same valance while describing distinct experiences have different impacts on reader's evaluations of review value. For example, service complaints are perceived as more helpful compared with facility complaints or money complaints. Furthermore, we investigate how hotel grades moderate the impact of review content characteristics on its perceived value. Results show that service complaints posted on high-end hotels’ pages receive significantly more helpful votes than those posted on low-end hotels’ pages. These nuanced yet important differences can only be revealed by mining meaning from text data, going beyond valence.