Extracting Usable Information from Online Reviews: Methodology and Implications

Abstract

Using 142.8 million online reviews on five product categories on Amazon, and through the lens of dynamic topic modeling, we analyze changes in the text of those reviews over time along two dimensions: aspect fluidity and preference fluidity. Aspect fluidity studies the extent to which product characteristics, be they physical or experiential that are described in reviews, change over time. Preference fluidity examines the dynamics of both the importance that reviewers place on aspects as well as their expressed satisfaction with these aspects. We find evidence that for most aspects, fluidity is rather muted, and more interesting is its gradual evolution with small incremental changes over time. The exception is for a few aspects related to new technologies that inherently change faster, making those aspects much more fluid in reviews. On preference fluidity, we find that certain product aspects become more important over time. We also find that there is a clear separation between aspect importance and the sentiment expressed in positive reviews compared to negative ones, and that 1 and 5-star reviews tend to focus on different aspects. We find that a few aspects dominate the content of reviews: relatively few aspects remain persistently important, and those have persistently strong positive or negative sentiment. Our findings have clear implications for organizing review information in ways that help potential customers navigate the ever-changing content of reviews. We highlight aspects that remain persistently important to consumers, thereby providing a way to identify important reviews from amongst the many available.