Customer preference identification from hotel online reviews: A neural network based fine-grained sentiment analysis

Abstract

As a kind of user-generated information, online reviews contain customers’ preferences for different aspects of hotels, which not only influence customers’ booking decisions but also help hotel managers to improve service quality of hotels timely. The key of deriving customers’ preferences from hotels’ online reviews is to identify fine-grained sentiment towards hotel attributes. In general, fine-grained sentiment analysis involves multiple fundamental tasks such as sentiment element extraction, aspect-opinion pair (i.e., AOP) identification and sentiment orientation analysis. However, existing fine-grained sentiment analysis methods cannot efficiently identify AOPs, especially when dealing with Chinese reviews. To this end, we construct an improved convolutional neural network (i.e., CNN) model, which can comprehensively utilize unstructured features and structured features, to improve the performance of AOP identification. We further propose a refined fine-grained sentiment analysis methodology to calculate accurate customer sentiment intensity value for each evaluated aspect rather than positive or negative sentiment, integrated with aspect term clustering algorithm, to identify customers’ specific preferences for different hotel attributes. Finally, to illustrate the reasonability and advantages of the proposed methodology, we conduct an empirical study with hotels’ online reviews crawled from Ctrip.com. Empirical results indicate that our proposed method can indeed improve the performance of AOP identification, and can effectively identify customer preferences from hotels’ online reviews. Furthermore, we find that customers show different preferences for different hotel attributes, and these vary across the types of customers.