On most landing pages users will see a sort option for “Subtopics”. This returns the given lineup of stories grouped by subtopics and only displaying groups that have at least two stories within it.
These groups are automatically discovered by using the words within the articles, ranking their importance, and then clustering those stories based on the identified important works. The result is the groups as you see them presented. There’s no guarantee that stories actually about quite different topics won’t have overlapping words, so some results may not be perfect, but hopefully this view gives users a quick way to scan for emerging themes in the news cycle.
If you’re curious for more details on the exact algorithms used, you can read more about tf-idf here, the means of ranking works, and read more about Jaccard distance here, the means of clustering the groups.