AI for Good

Chapter 21The Proliferation of Misinformation Online

—Will Fein, Mayana Pereira, Jane Wang, Kevin Greene, Lucas Meyer, Rahul Dodhia, and Jacob Shapiro

Executive Summary

Over the course of a very fruitful collaboration with researchers from Princeton's Empirical Studies of Conflict Project, we conducted an analysis that shed light into how users navigate to and from unreliable news sites online and then used our findings to motivate a machine learning model with practical use in identifying other unreliable sites.

In the initial analysis, we identified factors that contribute to the phenomenon of rabbit holes—specifically, the browsing patterns of users who, after encountering misinformation, tend to delve even deeper into engaging with untrustworthy content. This analysis highlights the stark differences in the way users reach reliable and unreliable news sites and how engagement with these sites differs as a result of linking structures.

We found that readers on unreliable sites click more internal links—links that refer readers to other articles within the site—making these sites “stickier.” We also found compelling evidence that reliable and unreliable news sites are largely disconnected from each other; unreliable sites provide few exits to reliable sources of news and reliable sources of news rarely send readers to unreliable sites.

We then used these findings to motivate a machine learning model capable of identifying unreliable sites based on patterns of ingoing and ...