🤖 AI Summary
Research on social media news ecosystems has long prioritized public-domain interactions, while private-domain user behaviors—and their impact on news credibility assessment and bias propagation—remain critically underexplored. Method: This study introduces the first “public–private dual-domain comparative framework,” grounded in a manually annotated dataset of over 19,000 news items. It integrates full-domain (public + private) and public-only datasets, employing multi-source data fusion, statistical modeling, and CrowdTangle-assisted cleaning and attribution. A fairness-aware data processing pipeline is also designed. Results: Users exhibit significantly deeper engagement with news in private domains than in public ones; moreover, interaction patterns for news varying in reliability and ideological bias differ systematically across domains. This work fills a foundational empirical gap in private-domain news behavior research and establishes a novel paradigm for platform governance and algorithmic auditing.
📝 Abstract
The rapid growth of news dissemination and user engagement on social media has raised concerns about the influence and societal impact of biased and unreliable information. As a response to these concerns, a substantial body of research has been dedicated to understanding how users interact with different news. However, this research has primarily analyzed publicly shared posts. With a significant portion of engagement taking place within Facebook's private sphere, it is therefore important to also consider the private posts. In this paper, we present the first comprehensive comparison of the interaction patterns and depth of engagement between public and private posts of different types of news content shared on Facebook. To compare these patterns, we gathered and analyzed two complementary datasets: the first includes interaction data for all Facebook posts (private + public) referencing a manually labeled collection of over 19K news articles, while the second contains only interaction data for public posts tracked by CrowdTangle. As part of our methodology, we introduce several carefully designed data processing steps that address some critical aspects missed by prior works but that (through our iterative discussions and feedback with the CrowdTangle team) emerged as important to ensure fairness for this type of study. Our findings highlight significant disparities in interaction patterns across various news classes and spheres. For example, our statistical analysis demonstrates that users engage significantly more deeply with news in the private sphere compared to the public one, underscoring the pivotal role of considering both the public and private spheres of Facebook in future research. Beyond its scholarly impact, the findings of this study can benefit Facebook content moderators, regulators, and policymakers, contributing to a healthier online discourse.
