Auditing E-Commerce Platforms for Algorithmically Curated Vaccine Misinformation

Abstract

There is a growing concern that e-commerce platforms are amplifying vaccine-misinformation. To investigate, we conduct two-sets of algorithmic audits for vaccine misinformation on the search and recommendation algorithms of Amazon-world's leading eretailer. First, we systematically audit search-results belonging to vaccine-related search-queries without logging into the platformunpersonalized audits. We find 10.47% of search-results promote misinformative health products. We also observe ranking-bias, with Amazon ranking misinformative search-results higher than debunking search-results. Next, we analyze the effects of personalization due to account-history, where history is built progressively by performing various real-world user-actions, such as clicking a product. We find evidence of filter-bubble effect in Amazon's recommendations; accounts performing actions on misinformative products are presented with more misinformation compared to accounts performing actions on neutral and debunking products. Interestingly, once user clicks on a misinformative product, homepage recommendations become more contaminated compared to when user shows an intention to buy that product. CCS CONCEPTS• Information systems → Personalization; Content ranking; Web crawling; • Human-centered computing → Human computer interaction (HCI).

A review of product safety regulations in the European Union

“…Even information itself can today be seen as a safety issue-if not a hazard. For instance, recent results indicate a substantial amount of vaccine misinformation (some of which may also be intentional disinformation) associated with potentially dangerous products sold on electronic commerce platforms [41]. These and other similar results restate the general platform problem.…”

A Review of Product Safety Regulations in the European Union

“…Even information itself can today be seen as a safety issue-if not a hazard. For instance, recent results indicate a substantial amount of vaccine misinformation (some of which may also be intentional disinformation) associated with potentially dangerous products sold on electronic commerce platforms [36]. These and other similar results restate the general platform problem.…”

Estimating and Penalizing Induced Preference Shifts in Recommender Systems

“…A variety of previous work considers how RSs algorithms might affect users: influencing user's preferences for e-commerce purposes (Häubl & Murray, 2003;Cosley et al, 2003;Gretzel & Fesenmaier, 2006), altering people's moods for psychology research (Kramer et al, 2014), changing populations' opinions or behaviors (Matz et al, 2017), exacerbate (Hasan et al, 2018) cases of addiction to social media (Andreassen, 2015), or increase polarization (Stray, 2021). There have been three main types of approaches to quantitatively estimating effects of RSs' policies on users: 1) analyzing static datasets of interactions directly (Nguyen et al, 2014;Ribeiro et al, 2019;Juneja & Mitra, 2021;Li et al, 2014), 2) simulating interactions between users and RSs based on hand-crafted models of user dynamics (Chaney et al, 2018;Bountouridis et al, 2019;Jiang et al, 2019;Mansoury et al, 2020;Yao et al, 2021;Ie et al, 2019a), or 3) using access to real users and estimating effects through direct interventions (Holtz et al, 2020;Matz et al, 2017). We see our approach as an improvement on 2), in that we propose to implicitly learn user dynamics instead of hand-specifying them.…”