Media Reporting on International Organizations: Using Machine Learning to Identify Worldwide Patterns

Coverage of international organizations (IOs) in news media is essential for the long-term build-up of their legitimacy. This article argues that news reporting on IOs systematically differs between powerful and rich countries, which dominantly control global IOs, and weaker and low-income countries, which are the primary recipients of global IOs’ deliveries. Reporting on IOs in the former is likely to be heavily oriented to central IOs politics and processes; reporting in the latter to local implementation work of the IOs. At the core of the empirical analysis is a dataset mapping news reporting on IOs in more than 4.6 million articles across 166 states and 64 languages, in the years 2018-2021, and a supervised machine learning model that classifies news on whether they relate 1) to central IO politics, 2) to ‘boots-on-the-ground’ local policy implementation by IOs, or 3) to IOs’ expertise and provision of information. It is based on the large pretrained semantic model ERNIE fine-tuned with 1654 human-labelled news articles. On the core classification task, the model achieves very high accuracy of 0.89. The analysis covers 45 IOs and IO bodies jointly forming the entire United Nations System. Empirical evidence strongly supports the theoretical argument: news reporting on local IO implementation activity is more than twice as frequent in weaker and low-income states than in powerful and high-income states, ceteris paribus. On the other hand, reporting on central IO politics is 30% more prominent in rich and powerful states than in low-income weaker states.