How News Framed Duterte's Drug War: A Computational Look at Media Narratives

A new study digs into how English-language news outlets talked about Rodrigo Duterte's controversial anti-drug campaign in the Philippines. The campaign ran from June 2016 to June 2025, and researchers wanted to see how different media outlets framed the story over time.

The researchers used advanced computer tools to analyze 1,133 headlines pulled from Google News. They applied BERTopic topic modeling and VADER sentiment analysis to sort through the data. Their goal was to understand how coverage changed based on institutional pressures like norms, rules, and peer influence.

Ten main themes popped up in the headlines. The biggest ones focused on human rights concerns, legal action by the International Criminal Court (ICC), and what happened after Duterte left office. Early on, Philippine-based outlets covered human rights issues more often than international ones. But after 2022, that flipped. By 2024 and 2025, international outlets mentioned human rights in about 64% of their headlines, while domestic outlets did so in roughly half.

Sentiment analysis showed most headlines were negative. Out of all the headlines studied, 807 were negative, 203 were neutral, and 123 were positive. Negative coverage spiked in early 2025, right around when Duterte was arrested on an ICC warrant. The language in headlines also shifted during key moments: the 2018 ICC filing, the 2020 UN Human Rights Report, and the 2021 ICC investigation authorization.

The study breaks down the forces behind these changes using Institutional Theory. Normative pressure explains why local Philippine outlets led the human rights conversation before any international group stepped in. Coercive pressure from the ICC and UN lines up with sudden shifts in tone. Mimetic pressure shows how media outlets copied each other's framing styles over time, first from local to global sources, then back again. The researchers also note that Google News's own choices about what to show may have influenced which voices and angles appeared in the data.