Seminarios Académicos

Abstract: Online reviews have a significant impact on consumer purchase decisions and on the success of e-commerce platforms. Despite this, review platforms like Amazon, Yelp, or Tripadvisor have struggled since their inception with the problem of fake reviews. We use a novel dataset on a large number of Amazon sellers buying fake reviews to test different approaches for platforms to detect them. We compare methods based on the network structure between sellers to those based on review features, including text and image features. We find that fake review buyers are highly clustered based on having reviews from a common set of reviewers. The result is that network-based detection strategies outperform all others and that even a small number of simple network features are highly accurate for detecting fake review buyers. We replicate these findings on data containing the full set of Amazon reviews using unsupervised clustering methods.