How can online marketplaces reduce rating manipulation? A new approach on dynamic aggregation of online ratings

Abstract

Retailers' incentives to manipulate online ratings can undermine consumers' trust in online marketplaces. Finding ways to avoid fake ratings has become a fundamental problem. Most marketplaces update product ratings immediately, i.e., display new ratings as soon as they are submitted. Some platforms have proposed to reduce the frequency of rating updates, as hiding ratings for a certain amount of time allows identifying and eliminating bursts of suspicious ratings. Reducing the update frequency also allows aggregating ratings and displaying only a summary statistic (e.g., mean of ratings). Although such aggregation helps to reduce the amount of fake ratings, as multiple fake ratings get represented by only one value, it might also distort legitimate ratings from real customers and hence have negative impact on honest retailers. In the present study, we propose and evaluate a novel method that instead of displaying every new rating immediately, aggregates a sequence of most recent ratings to k-values, with k determined dynamically based on the distribution of the recent ratings. In a simulation, we demonstrate that our proposed method outperforms state-of-the-art aggregation methods – it effectively reduces the impact of fake ratings on sales, and at the same time only marginally affects sales of honest retailers. Our proposed method can be easily integrated in online rating systems and can be especially used for designing fraud-resistant ranking algorithms.