This is something some colleagues in my research team (Reuters Institute for the Study of Journalism, part of the University of Oxford) are studying.

The evidence doesn't seem to show an effect: while there's definitely partisan selective exposure as a result of Facebook's algorithm, our longitudinal audience data suggests that before Facebook people mostly only read or watched a single news source. So even the attenuated diversity you get through a selective feed still seems to result in wider incidental exposure than the pre-social media world.

No paper to link, as the research is still ongoing. I understand they look like pretty robust results, though - it's certainly changed the way we look at things.