Machine Tools Reveal News Bias Patterns

News bias does not only come from opinionated wording. It can also come from what gets chosen, what gets skipped, and the lens used around a story. The Media Bias Detector tries to measure those patterns with a computational system built for large-scale news analysis.

Instead of asking people to read thousands of pieces by hand, the tool uses large language models together with fast-running news collection. It can sort hundreds of news items in a day and tag them with structured labels. Those labels include political lean, tone, topics, story type, and major events.

Those tags can then be compared at several scales. A single sentence can show a biased framing choice. A full news item can reveal broader patterns. A publisher can be studied across many days to see how its coverage differs from others. This gives researchers more room to examine two big questions: who gets attention, and how that attention is shaped.

Alongside the method, the team provides an interactive web platform so users can explore the collected data more easily. The release also includes a dataset made up of more than 140,000 news items from 2024, drawn from 10 prominent publishers. Early findings from that dataset show how the tool can point to patterns connected with bias in news coverage, including differences in topic selection and framing.