To identify relevant content for their users, social media platforms rely on implicit feedback signals to train and deploy recommender systems. However, the algorithms in these systems often amplify low-quality information, like misinformation and conspiracy theories. Current approaches to content recommendation are designed to optimize for user engagement, based on the assumption that aggregating over the individual engagement decisions of large crowds will help quality content “bubble up”. But prior work has shown that ranking the news by (either predicted or achieved) popularity can be problematic for partisan news consumers, as it creates a self-sustaining cycle that prioritizes pro-attitudinal information regardless of quality. How can we recommend reliable, trustworthy information on social media? Using clickstream data from a representative sample of U.S. residents, I have found that this cycle can be broken by prioritizing content that generates engagement in a politically diverse audience. These results highlight a fundamental weakness of current approaches to news recommendation, however the precise link between engagement, recommendation, and pro-attitudinal preferences is still unclear. In this talk I will describe ongoing work aimed at validating these results in an experimental settings and future research directions aimed at extending the concept of audience diversity to non-political attributes, and what are the implications for a better understanding of the socio-algorithmic foundations of more trustworthy news recommedation.
Giovanni Luca Ciampaglia is an assistant professor at the College of Information Studies at the University of Maryland College Park. He holds a Laurea in Computer Science from the Sapienza University of Rome and a Ph.D. in Informatics from the University of Lugano. He is interested in problems originating from the interplay between people and computing systems, in the determinants of information quality in cyberspace, and in how information propagates across social networks, with application to the integrity of information in cyberspace and the trustworthiness and reliability of social computing systems.