Perhaps one part curation, two parts crowd source and three parts algorithm? I don't think you can entirely automate this because there are nuances and issues of categorization.

I'd like to see it user/community driven. A user generates a topic and one headline. Then other users and the algorithm add headlines to the topic. And if there are two similar topics, it would require a curator intervention to join the topics.

Also, "sourcing + hashing" the articles and headlines so that you can see a "diff" of the articles and headlines in case the media companies change or alter the articles/headlines without informing their customers would be nice.

But if anyone tried this, I'm sure the media companies would attack the startup and of course with the new EU copyright laws, it might open the startup to lawsuits or takedown requests. Not to mention many countries might simply ban it outright.