Thanks for the reference. It's good to know that there is research backing SNS (social network sites) exposes us to more diverse news contrary to popular belief. Size of the data set seems to be big enough to rely on the results.

I'm wondering if the same holds true when that experiment is conducted in isolation. For example, if we take any particular news site, if a user is repeatedly visiting the same news site, then my hypothesis is that the kind of news one sees will converge onto a specific political view point. It might vary based on each news site - google news vs apple news based on how they are optimizing for users.

In reality, the isolated experiment might be moot as users do end up on news site from SNS which ultimately leads to recommendation algos picking up diverse signals to show diverse results. Given that, whether the algo itself is diverse in isolation or not wouldn't matter as the environment that feeds it is diverse. Does it make sense?

Good questions. I suspect diverse news may not actually result in diverse perspective anyway. I wouldn't be surprised significant fraction of people simply don't change their opinion, even while consuming news from diverse source. So it may be doubly moot in the end.