The Fault in the Stars: Understanding the Underground Market of Amazon Reviews

Reviews play a crucial role in a product's visibility and eventually customers' buying decisions on e-commerce sites. While there is suspicion and anecdotal reports about the prevalence of fake reviews on popular e-commerce sites, little is known about how reputation manipulation services continue to operate despite attempts to thwart them. We investigate an emerging fraud ecosystem of incentivized reviews -- fake reviews written by real customers. In this kind of review fraud, sellers give away products for free to complicit buyers in exchange for a positive review. Like other fraud services, these incentivized review services are operated through specialized agents that mediate the exchange between buyers and sellers. In this paper, we conduct an in-depth quantitative and qualitative study of such services for Amazon.com. We infiltrate Facebook groups that are used for recruitment. We are able to identify more than 1600 products seeking incentivized reviews. We first evaluate off-the-shelf machine learning classifiers as well as Amazon's own countermeasures against incentivized reviews and corresponding products. Our results demonstrate the ineffectiveness of these countermeasures -- even when incentivized reviews are removed, new ones are promptly added back to make up for the removed reviews. To better understand the inner workings of incentivized reviews services, we survey 38 agents and 36 buyers. Our analysis reveals that the reviews service spans multiple countries, with sellers typically in China, agents in Pakistan and Bangladesh, and buyers in the US, UK, Canada, France, and Germany. We find that groups of agents work together using sophisticated training materials, incentive structures, and reporting mechanisms. Buyers intentionally use tactics such as spreading out the reviews over time and simulating an organic search to circumvent detection.