What Airbnb Reviews can Tell us? An Advanced Latent Aspect Rating Analysis Approach

Luo, Yi
Major Professor
Rebecca(Liang) Tang
Committee Member
Journal Title
Journal ISSN
Volume Title
Research Projects
Organizational Units
Journal Issue
Apparel, Events and Hospitality Management

There is no doubt that the rapid growth of Airbnb has changed the lodging industry and tourists’ behaviors dramatically since the advent of the sharing economy. Airbnb welcomes customers and engages them by creating and providing unique travel experiences to “live like a local” through the delivery of lodging services. With the special experiences that Airbnb customers pursue, more investigation is needed to systematically examine the Airbnb customer lodging experience. Online reviews offer a representative look at individual customers’ personal and unique lodging experiences. Moreover, the overall ratings given by customers are reflections of their experiences with a product or service. Since customers take overall ratings into account in their purchase decisions, a study that bridges the customer lodging experience and the overall rating is needed. In contrast to traditional research methods, mining customer reviews has become a useful method to study customers’ opinions about products and services. User-generated reviews are a form of evaluation generated by peers that users post on business or other (e.g., third-party) websites (Mudambi & Schuff, 2010).

The main purpose of this study is to identify the weights of latent lodging experience aspects that customers consider in order to form their overall ratings based on the eight basic emotions. This study applied both aspect-based sentiment analysis and the latent aspect rating analysis (LARA) model to predict the aspect ratings and determine the latent aspect weights. Specifically, this study extracted the innovative lodging experience aspects that Airbnb customers care about most by mining a total of 248,693 customer reviews from 6,946 Airbnb accommodations. Then, the NRC Emotion Lexicon with eight emotions was employed to assess the sentiments associated with each lodging aspect. By applying latent rating regression, the predicted aspect ratings were generated. With the aspect ratings, , the aspect weights, and the predicted overall ratings were calculated.

It was suggested that the overall rating be assessed based on the sentiment words of five lodging aspects: communication, experience, location, product/service, and value. It was found that, compared with the aspects of location, product/service, and value, customers expressed less joy and more surprise than they did over the aspects of communication and experience. The LRR results demonstrate that Airbnb customers care most about a listing location, followed by experience, value, communication, and product/service. The results also revealed that even listings with the same overall rating may have different predicted aspect ratings based on the different aspect weights. Finally, the LARA model demonstrated the different preferences between customers seeking expensive versus cheap accommodations.

Understanding customer experience and its role in forming customer rating behavior is important. This study empirically confirms and expands the usefulness of LARA as the prediction model in deconstructing overall ratings into aspect ratings, and then further predicting aspect level weights. This study makes meaningful academic contributions to the evolving customer behavior and customer experience research. It also benefits the shared-lodging industry through its development of pragmatic methods to establish effective marketing strategies for improving customer perceptions and create personalized review filter systems.