Regional Remix
User-centric news with AI
How artificial intelligence is helping us rethink regional news
--
What if radio news was tailored to you personally? With the ability to skip and on demand? The AI + Automation Lab is working on just such user-centric regional news: Try out the BR Regional Update live. Here we’re outlining the idea behind the project, the method and our learnings.
Regional news is an integral part of Bayerischer Rundfunk’s DNA and has long been anchored in its linear radio programming: short news blocks, each on the hour and half hour, individually targeting five Bavarian regions:
Lower Bavaria & Upper Palatinate, Swabia, Upper Bavaria, Franconia and Main-Franconia. In this way, BR provides its listeners with news and weather from these regions several times a day.
Our “Remix Regional” project, which emerged as an idea at the end of 2020 as part of a BBC Hackathon, has set itself the goal of supplementing this linear offering in three key respects:
- Through even more granular regionalization
- Through better individualization with metadata
- Through nonlinear and on-demand availability
In other words: a digital news broadcast whose structure is individually aligned with our users’ location — whenever the users want it. An hourly updated news briefing, tailored to the individually perceived “home region”.
Remix Regional is thus a central component of our “Data Driven Publishing” strategy in the AI + Automation Lab: We use technology and (meta-)data to offer our users a tailored service — at the same time, projects like this help us to relieve and digitize processes in BR.
What do the users say?
A qualitative user test in mid-2021 with our first, static prototype provided the following clues for further development:
- The idea is good, but not within another separate app
- Regional news are relevant for an average of two, maximum three days
- The ability to navigate through the news is appreciated
- The additional map view is perceived as added value
- Geographical area of interest varies greatly from person to person
- Often several places are of interest
(current place of residence, place of origin, etc.)
In addition to this user test, we found out with the help of a quantitative survey that 40 to 50% of the 732 participants are strongly to very strongly interested in news reports from their immediate surroundings or their county and would like to see more BR news from their area.
For context: The administrative district of Lower Bavaria alone has nine counties.
A product home
In line with the user feedback “Please don’t do another app!”, Remix needed a home in the established BR world if possible: As a native app, the BR Radio app has everything it takes to serve as a home for Remix Regional and, above all, already has a grown, active user base that is interested in audio content.
We therefore developed a web dummy of the BR Radio app to send the idea into a second and more concrete user test: Would a feature like Remix Regional make the radio app more attractive to existing users of the app?
And is our thesis correct that segmented, modular audio content generates added value for BR and its audience? (Spoiler alert: Yes and yes.)
Behind the scenes
The biggest challenge we had to master was behind the scenes: While we still used manually cut and tagged audio snippets from the archive as sample content for the first prototype, our requirement now was to work with current live data and fully automated segmentation. To do this, we needed a technology — called a “segmenter” for short — that was able to
- automatically capture each new regional
news broadcast as it appears, and do so - multiple times a day and without
interrupting live radio operations, - match those with the corresponding information
from the BR broadcast scheduling tool, - identify the start and end of each news item
based on this information, and - subsequently cut the recordings into
individual short audio clips, - automatically add recognized location and
metadata to these data packages and - store them in an online memory where
they can be retrieved.
We succeeded — beyond the status of a mere proof of concept. The most critical part of the project was the correct recognition of the cut marks in the audio files, because: The more precise the cuts, the better the user experience in the end. Two aspects proved to have a decisive influence on this:
- The code of the segmenter itself: The more training data we provide to the model in the form of broadcast material and the more fine-tuning takes place, the better the algorithm learns to tailor the individual messages optimally.
- The workflows in the regional studios: The more uniformly and completely the people in the regional studios fill the broadcast scheduling tool, the better the algorithm is able to recognize the beginning and end of an individual message.
Thus, through a combination of technical and cultural work, we were able to successively raise the quality of the segmenter by about 20% already in this experimental status:
Several optimization loops on the algorithm itself, workshops with the regional editorial teams and a small but effective conversion of the broadcast scheduling software towards better metadata helped a lot.
Of course, our current prototype is not a bug-free product that meets the usual BR standards. However, since it is much more than just a test run, we decided to expand the web dummy frontend, which was initially intended only for user testing, into a publicly accessible website called “BR Regional Update” and thus show what we are working on internally to advance personalization and regionalization in BR.
What’s next?
After the second user test in cooperation with pub’s Userlab went extremely well, it was clear that we were on the right track and that Remix could create added value for BR’s radio products and our users.
With the new findings from the second user test and from the operation of the live demo, we will now send Remix into the third phase. Our goals are:
- Looking at the product and our listeners, the goal is a native integration of Remix into the digital BR radio landscape. Specifically, this means an integration into BR Radio’s public web offerings and the beta version of the BR Radio app.
- Looking at the bigger picture, the goal is to build an ARD-wide, universally conceived segmenter solution that makes it possible to personalize attractive products with modular content.
We are currently finding out whether both will work and will keep you up to date. :)
