CAREER: Socio-Algorithmic Foundations of Trustworthy Recommendations

University Of Maryland, College Park, College Park MD

Investigators

Abstract

Social media are now the main source of news for the majority of Americans and for billions of people worldwide. The recommendation algorithms employed by these platforms are designed to maximize views and clicks by finding engaging content, which in turn often ends up amplifying stories from dubious news sources, conspiracy theories, and unverified rumors. Older adults are a demographic increasingly turning to social media as a way to stay informed and are especially vulnerable to online misinformation. This situation is especially concerning for partisan news consumers, who have a tendency to engage with information that conforms to their beliefs, regardless of its accuracy. The objective of this research is to build a simulated social media platform that will allow researchers to test more robust news recommendation algorithms designed to improve the news consumption of social media users while recommending content that is still relevant to them. The project will include studies of how older adults engage with the news on major social media platforms and whether alternative algorithms would improve the quality of their news consumption. This project is based on the observation that ranking the news by (either predicted or achieved) popularity creates a self-sustaining cycle that prioritizes pro-attitudinal information regardless of its quality. To break this loop, the project tests the hypothesis that prioritizing content that generates engagement in diverse audiences will improve the trustworthiness of recommendations. The project will explore the relevant dimensions under which the heterogeneity of an audience provides a good signal for news quality. The technical contributions will include a set of new regularization techniques to incorporate audience diversity of news sources into content recommendation methods, a quantitative evaluation of the effect of different re-ranking methods on the quality of the information diet of news consumers, especially older consumers, and a new experimental methodology on counterfactual ranking with high ecological validity. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.