Learning judgment benchmarks of customers from online reviews

Online reviews play an important role for the purchasing decision of customers. One challenge is that different reviewers have different judgment benchmarks when making online reviews, which can mislead purchasing decisions. Specifically, the same star rating may correspond to different levels of sentiment for different reviewers because of the explicit preference differences in individuals. This study explores the personal judgment benchmarks through a preference learning process. Considering the nonlinear cognition of reviewers, we propose a marginal value function with smooth shapes and clear parameters to model the scores of online reviews. A mathematical programming model is established to predict the specific marginal value function for each reviewer. Two kinds of performance accurateness are defined to measure the performance of the learning model. We evaluate two empirical data sets extracted from TripAdvisor.com to deepen the understanding of personal judgment benchmarks. A simulation study is conducted to validate the proposed model. The results have important theoretical and practical implications for purchasing decisions based on online reviews.