Amazon’s search algorithm gives preferential treatment to books that promote false claims about vaccines, according to research by UW Information School Ph.D. student Prerna Juneja and Assistant Professor Tanu Mitra.
Meanwhile, books that debunk health misinformation appear lower in Amazon’s search results, where they are less likely to be seen, the researchers wrote in a paper that was recently accepted to CHI, the top annual conference on human-computer interaction.
In their paper, Juneja and Mitra noted that Amazon has faced criticism for not regulating health-related products on its platform. They conducted audits to determine how much health misinformation is present in Amazon’s recommendations and how the personalization of its search results affects the amount of misinformation users receive.
The researchers tested a list of 48 search terms on vaccine-related topics, such as “immunization” and “MMR vaccine and autism” over a 22-day period and found that products promoting misinformation consistently outperformed those that debunk false claims. Misinformative products performed especially well when results were sorted by certain filters — “average customer reviews” and “price low to high.”
Juneja and Mitra used bots to build user histories and test whether Amazon pushes shoppers toward misinformation after they’ve shown an interest. They found that after users click on misinformative products or place them in their carts, they are more likely to see similar items on their Amazon home pages, creating a misinformation feedback loop.
To combat misinformation on its platform, the researchers recommended several actions Amazon could take. Those include displaying informative links alongside the product listings, introducing bias indicators, and, in the long term, modifying its algorithms and policies to stop promoting misinformative books.
Read the full paper, “Auditing E-Commerce Platforms for Algorithmically Curated Vaccine Misinformation,” and reports on the research from the Seattle Times and Venture Beat, as well as Juneja's blog post on the study.