A framework for affinity-based personalized review recommendation

Abstract Online review platforms have proliferated thanks to technological advances and consumers’ increased dependence on each other’s opinions for purchase decisions. However, users typically face an enormous number of online reviews and suffer from information overload. Unlike previous research that relies mainly on popularity, crowd-based evaluation, or filtering methods, we propose a framework for personalized review recommendation based on user-review affinity. Indeed, this study seeks to identify and recommend reviews to each user according to the probability that he/she will like (hit the helpfulness vote/like button), comment on, or re-read those reviews, whereby user login time increases, which in turn correlates positively with user affinity toward the platform. We hypothesize a conceptual model, conduct predictive analytics, and perform counterfactual simulations on the log data of a large restaurant review platform in Asia and find that reviewer-user similarity is among the most significant explanatory factors, which is in line with the collectivist culture of the country where platform operates. Built on the results of the explanatory analysis, machine learning-based predictive models are then applied to predict the likelihood that each user will interact with each review for each business. Our counterfactual analysis demonstrates the potential of the resultant affinity-based ranking to increase user engagement with the platform.