Bluesky: Network Topology, Polarization, and Algorithmic Curation

TL;DR

Bluesky is analyzed as a multi-layer, decentralized social network with five million users, focusing on Followership, Replies, Reposts, and Likes to map topology and dynamics. The study combines network analytics, multilingual term extraction, and machine-learning stance prediction to characterize polarization and content curation; it finds heavy-tailed, highly clustered, and small-world properties across layers, and a predominantly left-center political leaning with issue-specific polarization around Israel-Palestine. It further reveals that Bluesky’s feeds marketplace enables many user-created algorithms, though uptake remains limited, while the platform maintains minimal exposure to questionable sources. Overall, Bluesky mirrors the topology of larger social networks while offering unique data access for social scientists to study network structure, algorithmic curation, and polarization in a decentralized setting.

Abstract

Bluesky is a nascent Twitter-like and decentralized social media network with novel features and unprecedented data access. This paper provides a characterization of its interaction network, studying the political leaning, polarization, network structure, and algorithmic curation mechanisms of five million users. The dataset spans from the website's first release in February of 2023 to May of 2024. We investigate the replies, likes, reposts, and follows layers of the Bluesky network. We find that all networks are characterized by heavy-tailed distributions, high clustering, and short connection paths, similar to other larger social networks. BlueSky introduced feeds-algorithmic content recommenders created for and by users. We analyze all feeds and find that while a large number of custom feeds have been created, users' uptake of them appears to be limited. We analyze the hyperlinks shared by BlueSky's users and find no evidence of polarization in terms of the political leaning of the news sources they share. They share predominantly left-center news sources and little to no links associated with questionable news sources. In contrast to the homogeneous political ideology, we find significant issues-based divergence by studying opinions related to the Israel-Palestine conflict. Two clear homophilic clusters emerge: Pro-Palestinian voices outnumber pro-Israeli users, and the proportion has increased. We conclude by claiming that Bluesky-for all its novel features-is very similar in its network structure to existing and larger social media sites and provides unprecedented research opportunities for social scientists, network scientists, and political scientists alike.

Figures (10)

• Figure 1: User activity on the BlueSky social media platform from February 2023 to May 20, 2024. Each panel details the number of new and existing active users, ranging from follows (A), likes (B), posts (C), reposts (D), feed generation (E), to blocks (F), showing the number of unique users engaging through these actions. The term 'New Users' refers to individuals interacting for the first time with the platform through the respective activity measure. Blue areas denote new users, while red areas show the number of existing users engaging in activity.
• Figure 2: Social media interaction metrics with posts (log-log scale). Each panel plots a specific metric against its frequency to analyze patterns of user engagement and content spread. The X-axis represents the specific metric, and the Y-axis shows the frequency of occurrences for each metric value. (A) Reposts per Post. (B) Likes per Post. (C) Quotes per Post. (D) Comments per Thread. (E) Posts per User. All plots use logarithmic scales.
• Figure 3: Social media interaction metrics by Users (log-log scale). Each panel represents a specific interaction metric plotted against its occurrence frequency to analyze patterns of user interactions. The X-axis denotes the metric in question, while the Y-axis shows the frequency of occurrences for each metric value. (A) Reposts per Post.(B). Likes per Post. (C) Comments per Thread. (D) Posts per User.
• Figure 4: Metrics capturing changes in the network structure from 2023 to May 2024. These metrics are computed across four networks. (A-D) Count of unique nodes (in millions) active per week for each network. (E - H) Number of unique Edges in the network per Week. (I - L) Ratio of edges to unique edges, capturing the activity of nodes in each week. The black dashed line in each graph denotes the date of the public opening of Bluesky.
• Figure 5: Metrics capturing changes in the network structure from 2023 to May 2024. Clustering-coefficient, density, and average shortest path are computed for four networks. Replies, Reposts, and Likes capture non-persistent interactions, thus all metrics are calculated individually for each week's edges. The followership network is persistent. (A-D) Normalized average clustering coefficient. The dashed red line represents an equal value for the random and original graph. (E-H) Density of the networks. (I - L) Average shortest path length for all networks. The black dashed line in each graph denotes the date of the public opening of Bluesky.