Consumer preference analysis integrating online reviews: a multiple criteria group approach considering individual stochastic behavior

To find a rational basis underlying the holistic judgments provided by customers, a data-driven preference model is constructed by indirectly inducing preferences from online reviews. Compared with direct questioning, the indirect induction of preferences needs to pay attention to two aspects: the expressive ability of the preference model and the robustness of the decision calculated using it. To improve the expressive ability of the preference model, a nonparametric preference-learning approach is proposed to obtain the preference model, which considers individual stochastic behavior and customers’ psychological factors, e.g., reference level dependence. In addition, to improve the robustness of the decision results, the parameters in the most representative preference model are obtained via a robust ordinal regression model. To verify the effectiveness and superiority of the proposed method, this study selects five hotels from Tripadvisor website for a case study, applies the preference models of different customer groups obtained based on online reviews to the recommendation of the same star rating hotels and compares them with those of other methods.