Case Study: How Bayerischer Rundfunk Used Modular Journalism to Personalize Radio News Based on Location


Published October 1, 2024

AI in the Newsroom is a series of case studies from the Online News Association (ONA) that highlight specific ways journalists are building and using AI tools.

Launched by Germany’s public broadcaster Bayerischer Rundfunk (BR), BR Regional Update is an innovative service that delivers personalized radio news content based on listeners’ locations. This service represents a significant shift in how radio news is produced and consumed, breaking down traditional broadcasts into modular, geo-tagged segments.

Opportunity

BR delivers radio news segmented across five regions in Bavaria in a linear format. The challenge was to make regional news a more granular, personalized, and on-demand news experience while maintaining the audio-first approach of radio broadcasting. This aligned with BR’s vision for a data-driven publishing strategy, aiming to transform static, one-size-fits-all broadcasts into flexible, reusable content modules tailored to create location-specific audio news experiences.

Solution

The initial idea, conceived at a BBC hackathon three years ago, was to split linear radio news into individual audio files, each tagged with geolocation data. This led to the development of a static prototype that manually segmented audio and geolocated it on a map. Following positive user feedback, the project advanced to create a more dynamic and automated solution.

Key features of BR Regional Update

Audio Processing: The system extracts audio files and host scripts from regional studios and uses a toolchain that includes Whisper AI to create the transcripts. These transcripts are compared with the scripts to accurately locate the start and end points of each news item. Ongoing efforts aim to fine-tune a model for more precise cut marks, ensuring smooth audio segmentation.

Location Detection: Once audio segments are defined, the next step is to determine their locations. This can happen either manually or automatically.

  • Manual Metadata Input: Editors and hosts add location metadata to news items in Open Media, the newsroom management software. This allows the segmented news items to automatically include the corresponding metadata. However, this step is sometimes missed due to the fast-paced nature of the newsroom.

  • Automated Location Detection: For items lacking manual metadata, the system searches for location names within the news items, matching them to a pre-existing list of about 2,100 locations with fixed latitude/longitude data. If no match is found, a separate mapping service is used to try to determine the location. If no location is explicitly mentioned, AI algorithms infer the location from the news content. However, in 10% of cases, insufficient information may render an audio clip unusable.

Finally, metadata, including geolocation and content summary, is attached to each segmented audio file. These enriched audio files are stored in a database for easy retrieval and personalized news delivery.

User Interface: Audience can access this personalized news through the BR Regional Update web app by entering their location or using their device’s geolocation feature. The app dynamically adjusts the news feed based on their preferences, such as the age and radius of news coverage.

Results so far

Mobile view of BR Regional Update shows a recommended radio story based on a location on the map.

The project has transitioned to a public beta website, receiving positive feedback for its personalized news experience. Audience tests highlighted appreciation for navigable audio news items and interest in features such as subscribing to multiple locations (e.g., current residence, hometown) as well as making it available via RSS feed. These features are being considered for future implementation. This project is also helping digitize the newsroom infrastructure and helping stakeholders in understanding the value of metadata and personalization.

Lessons learned

  • Metadata is Key: The project underscored the importance of having modular, regionalized, and personalized content items with rich metadata. While it required building a complex product to prove this point, it demonstrated the value of metadata in enabling personalized content delivery.

  • Audience-Centric Design is Essential: Multiple rounds of audience testing were crucial in refining the product. The team learned to value audience feedback and iterate based on audience needs. The team learned to be adaptable, starting with a static prototype and evolving to a dynamic system based on audience feedback and technological capabilities.

  • Show, Don’t Just Tell: Tangible prototypes are vital. As Max Brandl, the product manager puts it, “You can talk so much about this maybe great idea but it’s nearly impossible to convince people if you can’t show what could be at least one version of a possible product.” Creating a working prototype was crucial in demonstrating the concept’s potential and opening doors within the organization.The team also found value in keeping the project “beneath the surface” for as long as possible which allowed them to operate quickly and implement ideas without premature skepticism. They only revealed the project when it was nearly complete, which proved to be a “door opener” as people could immediately understand its potential.

Keep learning

To learn more about this product and other AI work happening at BR, listen to the Newsroom Robots podcast episode with Uli Köppen, now the Chief AI Officer of BR.

This resource is part of the AI in the Newsroom series. Read other case studies you might have missed:

Related Topics
Author
Nikita Roy
ICFJ Knight Fellow and Newsroom Robots Podcast Host