[mooman219]

I worked on content recommendation for a bit while I was at Microsoft. People click on low quality articles, spend a lot of time reading said articles, then seek out similar or lower quality articles containing the same named entities. If you recommended a variety of content to those people, they'd most frequently pick articles of equal or lesser quality about named entities they just read.

This is really kind of saddening, but your best bet really is just personalizing on as many user metrics as you can collect to best serve the people who like clickbait as well as the people who are looking for higher quality articles.

Not really relevant, but funny enough, if you prioritize article quality, popular named entities, how relevant those entities are to the article, engagement, and article age, you end up with earnings reports.