A Longitudinal, Multinational, and Multilingual Corpus of News Coverage of the Russo-Ukrainian War
A new dataset of 246,000 news articles from 11 outlets across 5 nations reveals how media constructs competing realities.
A team of researchers has published DNIPRO, a major new dataset designed to computationally analyze media narratives around the Russo-Ukrainian war. The corpus contains 246,000 news articles collected from February 2022 to August 2024, sourced from 11 major outlets across five critical nation-states: Russia, Ukraine, the U.S., the U.K., and China. The articles are in three languages (English, Russian, Ukrainian) and come with rich, human-evaluated annotations for key attributes like political stance, sentiment, and topical framing. This structure transforms a vast amount of news text into a structured resource ready for empirical study.
DNIPRO is uniquely positioned to power research on narrative divergence and information warfare. Initial exploratory analyses using the dataset reveal how media outlets construct incompatible realities not by directly refuting opposing narratives, but through subtle differences in attribution of events and selection of which topics to cover. For AI and NLP researchers, this corpus enables the development of models that can detect implicit contradictions and track the evolution of cross-lingual information flow in fragmented digital ecosystems. It will be presented at the Language Resources and Evaluation Conference (LREC) in 2026.
- Contains 246,000 articles from 11 outlets across 5 nations (Russia, Ukraine, U.S., U.K., China) in 3 languages.
- Features human-annotated metadata for stance, sentiment, and framing, enabling systematic analysis of narrative construction.
- Initial analysis shows media creates incompatible realities through divergent attribution and topic selection, not direct refutation.
Why It Matters
Provides a foundational dataset for building AI tools to detect media bias, analyze information warfare, and understand geopolitical narrative shifts.