For people wondering how this can be neutral, I found their method quite interesting. From what I gathered, they deploy some bot accounts that follow one initia

For people wondering how this can be neutral, I found their method quite interesting. From what I gathered, they deploy some bot accounts that follow one initial political account, and then all interactions afterwards (liking posts, following other people) are random. From this, their main finding was that > We find no strong or consistent evidence of political bias in the news feed But while this study focuses on algorithmic bias, I thought one of the main accusations about platform bias was that twitter banned/suspended more conservative accounts than liberal ones. Also something I found interesting: > Conservative accounts also find themselves in denser communities and are exposed to more low-credibility content With the definitions of "low credibility" based on previous research papers: > A low-credibility source is one that exhibits extreme bias, propaganda, conspiratorial content, or fabricated news. We use a list of low-credibility sources compiled from several recent research papers. Specifically, we consider a source as low-credibility if it fulfills any one of the following criteria: (1) labeled as low-credibility by Shao et al.33; (2) labeled as “Black,” “Red,” or “Satire” by Grinberg et al.49; (3) labeled as “fake news” or “hyperpartisan” by Pennycook et al.62; or (4) labeled as “extreme Left,” “extreme Right,” or “fake news” by Bovet et al.63. This provides us with a list of 570 sources.