Quantifying Media Representation Dynamics Across 25 Years of News Reporting on Policing-related Deaths

2026-06-08 · Source: cs.CL updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Advanced, extended

Summary

A computational analysis of 4,000 Canadian news articles spanning 25 years (2000-2025) reveals that reporting on policing-related deaths historically features state bureaucrat perspectives nearly three times more often than civilian accounts. Researchers developed the PerspectiveGap model, a language model-based computational pipeline, to accurately identify points of view from bureaucrats, civilians, or context. The study found that the median article dedicated 33.3% of its passages to bureaucrat perspectives versus 11.8% to civilian accounts. However, civilian representation significantly increased from 2020-2023, peaking at 30% in 2023, coinciding with heightened scrutiny of police misconduct. Qualitative analysis shows bureaucrat accounts are clinical and procedural, while civilian discourse carries more emotional valence and focuses on justice and complex personhood.

Key takeaway

For media analysts and journalists covering policing-related deaths, this research highlights the historical imbalance in narrative authority, with state bureaucrats dominating discourse. You should actively seek and amplify civilian perspectives, especially given their increased representation post-2020, to foster a more co-productive public opinion formation. Consider how your outlet's content length and production pace might influence reliance on prepackaged official statements versus in-depth civilian accounts.

Key insights

News narratives on policing-related deaths disproportionately amplify state perspectives, though civilian voices gained prominence post-2020.

Principles

• Media representation shapes public opinion.
• Epistemic authority in news is dynamic.
• Bureaucratic accounts are procedural.

Method

The PerspectiveGap model uses language models and coreference resolution to classify paragraphs as bureaucrat, civilian, or context perspectives, trained on 100 annotated articles and applied to 4,000.

In practice

• Contextualize PerspectiveGap to other jurisdictions.
• Track epistemic authority dynamics over time.

Topics

• Media Representation
• Policing Accountability
• Computational Linguistics
• PerspectiveGap Model
• News Narrative Analysis
• Epistemic Authority

Code references

• smfsamir/perspective-gap

Best for: NLP Engineer, AI Scientist, Research Scientist, Domain Expert

Related on AIssential

• Rational Silence and False Polarization: How Viewpoint Organizations and Recommender Systems Distort the Expression of Public Opinion — Online opinion polarization stems from the interplay of user incentives, viewpoint organizations, and AI recommender systems.
• Behind the AI in the Newsroom: The Washington Post’s Vineet Khosla — AI enables news organizations to personalize content and enhance journalistic workflows while maintaining ethical standards.
• More evidence doesn’t mean more justice: The limits of visual technologies in human rights cases — Legal institutions' gatekeeping of digital evidence distorts justice by defining what counts as "truth" and whose knowledge is valid.
• Baptists and Bootleggers: The Hidden Coalition Behind ‘Data-Driven’ Decisions — Unlikely coalitions of principle-driven and self-interested parties often drive "data-driven" decisions.
• Crime As Proxy For Disorder — Perceived increases in crime often reflect concerns about disorder, which data suggests largely tracks historical crime trends.
• A Model of Multi-turn Human Persuadability Using Probabilistic Belief Tracing — Process-level belief tracing and human-faithful simulators are crucial for understanding and safely optimizing LLM persuasion.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CL updates on arXiv.org.