Creating AI-generated stories for the Bundesliga channels
16 July 2024 – Through its app, website and social media channels, the DFL supplies football aficionados around the world with a constant stream of news from the Bundesliga and Bundesliga 2 as well as insights into German professional football. AI-based processes are increasingly applied to further expand the range of services and attract new user groups.
On the road to success with the Story format
The so-called Story format is well-established and very popular among users of social media platforms. A Story consists of a sequence of images or videos in smartphone-optimised vertical format. Every image is overlaid with a brief, descriptive text, and users browse through the story image after image.
“We consider this form of content usage as an established best practice, so we want to gradually embed it in our digital offerings. With our Bundesliga Shorts, this journey got off to a successful start,” explains Björn Rosenthal, Head of Product at DFL Digital Sports GmbH. “We are convinced that this will make our content much more appealing to our users, as it is optimised for mobile consumption.”
At the DFL, a team of editors is in charge of writing professional reports about what is going on in German professional football. They create articles, analyses and commentary for publication on Bundesliga digital media.
AI optimises editorial processes
Since the Story format requires much more visual material than conventional articles, along with a different layout, it comes with an increased need for editorial resources. “This has prompted us to search for an AI-based approach to supporting the production of Stories,” says Rosenthal.
This is how it works: Generative AI first analyses the content of an article written by the editorial team, then creates the desired Story format based on pre-defined rules. This concept has been implemented successfully.
AWS provides the toolset
Today, the DFL’s new content management system Contender provides a full range of generative AI functionality enabling automated creation of Stories from existing articles. The generative AI thus reduces the additional time and effort to a bare minimum. For example, an editor can ask Contender to generate a Story from a set of photos and add suitable titles and brief descriptive texts. Before any of the draft Stories generated automatically by the system can be published, human editors must review them, make the necessary adjustments and perform some fine-tuning.
“We have a number of use cases for generative AI. These include machine learning, computer vision as well as generation of Stories.”
Björn Rosenthal, Head of Product at DFL Digital Sports GmbH
All this is made possible by the technology partnership between the DFL and Amazon Web Services (AWS) which was renewed and expanded last spring and now includes generative AI as an additional set of services.
“We have a number of use cases for generative AI,” explains Rosenthal. “These include machine learning, computer vision – which means automated image interpretation – as well as generation of Stories.” To generate text, the content management system uses the large language model Claude offered by AWS partner Anthrophic. “All this runs on AWS’ AI service AWS Bedrock, which we are using,” says Rosenthal. This service is embedded in the content management system and will supply the Bundesliga app with Stories on a regular basis from the beginning of the 2024/25 season.
The challenge of image management
“Automated text creation was a major challenge which we have been able to tackle successfully,” confirms Rosenthal. “And while text generation may appear like a relatively simple task at first, it was a complex process requiring attention to many details and a substantial amount of time to make sure the results are fully satisfactory.”
A core task where AI can deliver massive support is generating the indispensable meta data for each image. These are a critical prerequisite for AI processes because they help the machine ‘understand’ what to search for. Every season, every match and every player is assigned a unique DFL identifier that can be used to automatically categorise new photos so they can be identified automatically and efficiently based on a pre-defined system at the DFL. This enables very fast searches.
AI optimises image quality
Finding suitable photos for Stories quickly after every Bundesliga and Bundesliga 2 match is another hurdle the DFL has been able to overcome with the help of AI: Using the match statistics provided by DFL subsidiary Sportec Solutions, and the timestamp on the digital TV video, AI can quickly extract individual frames showing specific match events. “This approach allows the AI solution to even analyse specific player poses,” says Rosenthal. “So we have created a system that can automatically search for images using computer vision and AI, generate text components, and provide the editor with a finished Story which the editor can either publish directly or revise as needed.”
“So we have created a system that can automatically search for images using computer vision and AI, generate text components, and provide the editor with a finished Story which the editor can either publish directly or revise as needed.”
Björn Rosenthal, Head of Product at DFL Digital Sports GmbH
Another benefit: “To ensure satisfactory image quality, especially when using frame clippings, AI algorithms upscale the selected clippings and sharpen the contours,” Rosenthal points out. Here again, the editor has the last word: as the ‘human in the loop’, she or he checks whether the result meets the DFL’s quality criteria, makes editorial adjustments, and puts on the finishing touches.
The DFL and AWS announced the expansion of their collaboration with a focus on AI at SportsInnovation in March 2024. Now these major AI accomplishments demonstrate the potential of the partnership. Thy create synergies at the editorial office while delivering multifaceted, modern content to Bundesliga enthusiasts.