Balancing Domestic and Global Perspectives: Evaluating Dual-Calibration and LLM-Generated Nudges for Diverse News Recommendation

A team of researchers from the University of Minnesota and other institutions conducted a novel experiment to see if AI could help break users out of their news consumption echo chambers. They designed two types of interventions: a 'topic-locality dual calibration' algorithm that actively balances recommendations between domestic and world news, and a presentation nudge where a large language model (LLM) generates personalized explanations for why a user might want to read a particular article. The 5-week study was run on the POPROX news recommendation platform with 120 active U.S. news readers.

The results, detailed in the arXiv paper 'Balancing Domestic and Global Perspectives,' showed that the algorithmic nudge was a clear success. It effectively increased both the exposure to and consumption of a more diverse set of news, specifically pushing users toward a better balance of domestic and international coverage. The impact of the LLM-generated text nudges was more varied, though highlighting an article's relevance to a user's prior reading history performed better than generic topic-based prompts. Crucially, the study found evidence that longitudinal exposure to this calibrated feed can actually shift readers' habits, making them value a more balanced news digest.

This research provides a concrete framework and evidence for platform designers aiming to build 'pro-social' algorithms. Instead of purely optimizing for engagement clicks, the study demonstrates it's possible to design systems that gently expand user horizons. The findings offer a roadmap for future work on using AI not just to reflect user preferences, but to positively shape them toward a healthier information diet.