A preference learning method to estimate consumer preferences from online reviews

Accurate prediction of consumer preferences aids numerous marketing efforts on e-commerce platforms, as it identifies which product attributes have an impact and how they influence a consumer’s purchase decisions. This study proposes a preference learning method for the automated extraction of consumer preferences from online reviews. A preference model, grounded in the multi-attribute value theory, is proposed to delineate the preference structures of consumers. This model bridges overall ratings and attribute-level reviews while incorporating considerations of attribute importance, compensation effects between attributes, and inconsistent sets of attributes across reviews. A classification algorithm is presented based on an optimization model to estimate preference parameters within the preference model. Its prediction accuracy is evaluated using k-fold cross-validation, while its robustness is measured through simulations. Case studies in the hotel, restaurant, and automobile domains validate that the proposed method generates transparent preference models with robust predictive performance and clear interpretations.