Uncovering Synergy and Dysergy in Consumer Reviews: A Machine Learning Approach

Abstract

Massive online text reviews can be a powerful market research tool for understanding consumer experiences and helping firms improve and innovate. This research exploits the rich semantic properties of text reviews and proposes a novel machine learning modeling framework that can reliably and efficiently extract consumer opinions and uncover potential interaction effects across these opinions, thereby identifying hidden and nuanced areas for product and service improvement beyond existing modeling approaches in this domain. In particular, we develop an opinion extraction and effect estimation framework that allows for uncovering customer opinions’ average effects and their interaction effects. Interactions among opinions can be synergistic when the co-occurrence of two opinions yields an effect greater than the sum of two parts, or as what we call dysergistic, when the co-occurrence of two opinions results in dampened effect. We apply the model in the context of large-scale customer ratings and text reviews for hotels and demonstrate our framework’s ability to screen synergy and dysergy effects among opinions. Our model also flexibly and efficiently accommodates a large number of opinions, which provides insights into rare yet potentially important opinions. The model can guide managers to prioritize joint areas of product and service improvement and innovation by uncovering the most prominent synergistic pairs. Model comparison with extant machine learning approaches demonstrates our improved predictive ability and managerial insights. This paper was accepted by Gui Liberali, marketing.