Check out the paper at PlosOne journals.plos.org/plosone/arti...

Even though political leanings on Bluesky are relatively homogeneous, we find issues-based divergence by studying opinions related to the Israel-Palestine conflict where opposing viewpoints seldomly interact.

Users predominantly share left-center news sources (63.4%), with very few right-leaning outlets (7.9%). Interestingly, the spread of questionable sources is minimal - only 0.14% of shared links.

Bluesky's custom feeds feature has created a diverse algorithmic ecosystem with almost 40,000 unique feeds! From professionally curated content like the "Science" feed to niche interest categories and alternative home timelines, users have embraced feed creation.

We found that despite its novel features, Bluesky's network structure resembles other social networks: heavy-tailed distributions, high clustering, and short connection paths. Sign-ups were strongly correlated with controversial events at Twitter/X!

Hi Luc, thank you! The model was trained on 2,000 annotated posts. For the neutral class 39% of cases were misclassified—with more of the errors leaning towards Palestine. The F1 for neutral was 65%, Palestine 67%, and Israel 62%. However, we did not look at formal calibration measures.

Thanks & Noted - will fix when we submit this!

🙋‍♂️Just published a preprint about our little website 🟦☁️

Whether your Twitter data has dried up, or whether you want to be able to observe the graph of a novel social media site - Bsky is an exciting opportunity for researchers. Check out our paper arxiv.org/pdf/2405.17571 and reach out for any questions (firstname[.]lastname@uzh.ch) or send me a DM.

Feed algorithms have been blamed for polarisation, misinfo, and radicalisation. However, research has been hindered by the fact that feeds are backboxes – we don’t know how they rank content and cant study exposure. Bsky users create their own feeds and we can see which feeds user like!

Israel & Palestine (Cont): Fascinating patterns in the relative proportion of stances on the conflict over time. While interest in the conflict has decreased over time this decrease is mainly driven by supporters of Israel while voices of Palestine have remained vocal and now dominate the discourse.

But does that mean that there is no controversy? We train a Transformer model to infer the stance of Users on the Israel – Palestine Conflict and repeat the analysis. Two large modular clusters of users emerge! This patterns show clear signs of polarisation.

Who uses Bluesky? We look at the political leanings of Users, and investigate polarisations and political leanings on Bluesky. Bluesky users are almost exclusively center-left leaning with a unimodal distribution of political opinions.

Millions of users have created hundreds of millions of posts, likes and follows – all accessible via the AT Protocol. We find that initial signups to Bluesky were driven by Technical Troubles on Twitter. Remember receiving rate limit errors last year? That lead to a surge in singups on Bluesky!

hey - have the same issue right now - what do you mean by legacy firehose? I'm essentially using github.com/MarshalX/atp...

Thanks for your help!

@pfrazee.com would that change show up in the firehose api?