Olle Zachrison, the News Commissioner and Head of AI at Swedish Radio, sheds light on how this public broadcaster has revolutionised audio news production, curation, and consumption in the digital age.
Transformation of back-office systems
Swedish Radio initiated a transformative journey to consolidate its back-office systems, culminating in the creation a seamless editorial ecosystem. Until recently, the broadcaster employed external software for news production while relying on a distinct, in-house content management system for digital content publication. However, recognising the need for heightened integration, flexibility, and security, Swedish Radio implemented an in-house broadcast system.
Olle Zachrison highlights the rationale behind this strategic shift:
"For us, it's better if we can control both systems ourselves, not to mention the cost optimisation."
A pivotal factor in this decision was Swedish Radio's unique operational landscape. With 55 offices and 26 local stations across Sweden, frequent live broadcasts from diverse locations presented a significant challenge. This necessitated a tailored content management solution supporting high-quality, live reporting from any location.
Recognising the limitations of off-the-shelf content management systems, the organisation opted for a customised CMS that precisely aligned with its specific requirements. Furthermore, the desire to reduce dependence on external providers was a driving force in this transformative endeavour.
Audio-first strategy supported by automation
With a strong focus on audio content, Swedish Radio prioritises an audio-first strategy across all platforms. Its digital offering revolves around a concept where news clips are treated like songs in a Spotify playlist and updated around the clock.
"We use an editorial algorithm to put the clips in a certain order, and we have playlists for local news, national news, sports, science, culture, etc. Like any traditional radio station, we broadcast a lot on a set schedule, but we have also added a live news team that goes live quickly when there is breaking news. At the same time, we start a live stream in our app and send a push notification to users. When a user clicks on the notification, they are taken directly to the live audio stream," says Olle.
News clips in Swedish Radio app are treated like songs in a Spotify playlist and updated around the clock (image by Swedish Radio)
Fusing technology and values
Swedish Radio produces about 350 news clips daily, which human editors evaluate according to three criteria: general news value, distinct public service value, and lifespan. In this way, the broadcaster aims to provide news that aligns with their vision of more voices and powerful stories for greater understanding.
"The overall news value is the story's magnitude: is it small or extraordinary? The second dimension is particularly important because it relates to our mission as a public broadcaster: We want to be pluralistic and give a voice to people with different perspectives.
We do not just sit in the office and report the news; we want to be out there talking to people and giving the full context of the story.
Finally, the third dimension – lifespan – describes how long a story will last: A minor traffic accident has a short lifespan, while an investigation about a high police chief can last weeks or months," Olle explains.
These three dimensions are assigned through metadata tags associated with each piece of content. Based on this, the algorithm decides where the story should appear in a news playlist and on the website. Crucially, this relationship between human judgement and technology works. Analysis shows that stories that include Swedish Broadcasting's unique public service values achieve 28% more engagement.
In 2021, the News Values project was awarded the European Broadcasters Union (EBU) Technology & Innovation Award.
Personalisation and geographic relevance
While Swedish Radio's news playlists are primarily curated based on their editorial algorithm, they offer some personalisation options. Users can choose their location to receive a separate playlist of news clips from that area. In addition, SR has prototyped a more personalised mixed playlist that allows users to select specific news sources or topics of interest. While this feature is not yet in production, it underscores the commitment to customise the news experience to individual preferences.
"We also give our listeners personalised podcast recommendations based on the AI system (provided by the EBU) called Peach. At its basic level, it's a recommendation engine, but it's also a self-learning system that improves over time."
Swedish Radio's news playlists offer personalisation options based on the user's location (image by Swedish Radio)
Pilot innovations
Swedish Radio's focus on media innovation and digital transformation manifests itself in other initiatives, such as the automatic transcription of news clips thanks to an internal system based on AI technology. The broadcaster is also experimenting with synthetic speech technology in some pilot projects (currently used in the SR to guide listeners from one clip to the next).
Another innovation Swedish Radio is working on is automatic tag extraction within the transcription system. This feature uses natural language processing (NLP) techniques and algorithms to automatically identify and extract key terms, entities, or topics from the transcription, improving searchability and allowing for easier categorisation and indexing of specific information within the transcriptions. The broadcaster is also working on innovations related to audio car systems and time-shifted radio, which allows rewinding and accessing previously aired content, including songs and broadcast news.
Olle Zachrison, News Commissioner and Head of AI at Swedish Radio
The text was originally published on Autentika's blog on June 27, 2023.