Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Blue Start: A large-scale pairwise and higher-order social network dataset

Alyssa Smith, Ilya Amburg, Sagar Kumar, Brooke Foucault Welles, Nicholas W. Landry

TL;DR

We present a large-scale Bluesky dataset that links pairwise following relationships with higher-order group structures via user-curated starter packs, addressing a gap in validating higher-order dynamics. The dataset aggregates 26.7 million users, 1.6 billion follows, and 301.3 thousand starter packs, collected from Bluesky APIs with careful anonymization and ethical considerations. Two data records are released in multiple formats: a deidentified starter-pack corpus with hyperedge and HIF representations, and a deidentified follows network with millions of directed edges, enabling joint analyses of pairwise and higher-order interactions. Technical validation confirms rich, heterogeneous degree distributions, dense higher-order overlap, and complementary information between the two representations, highlighting the dataset's utility for higher-order network science and graph learning applications. Data availability through SOMAR and accompanying code further support reproducible research and methodological development in higher-order social networks.

Abstract

Large-scale networks have been instrumental in shaping the way that we think about how individuals interact with one another, developing key insights in mathematical epidemiology, computational social science, and biology. However, many of the underlying social systems through which diseases spread, information disseminates, and individuals interact are inherently mediated through groups of arbitrary size, known as higher-order interactions. There is a gap between higher-order dynamics of group formation and fragmentation, contagion spread, and social influence and the data necessary to validate these higher-order mechanisms. Similarly, few datasets bridge the gap between these pairwise and higher-order network data. Because of its open API, the Bluesky social media platform provides a laboratory for observing social ties at scale. In addition to pairwise following relationships, unlike many other social networks, Bluesky features user-curated lists known as "starter packs" as a mechanism for social network growth. We introduce "A Blue Start", a large-scale network dataset comprising 26.7M users and their 1.6B pairwise following relationships and 301.3K groups representing starter packs. This dataset will be an essential resource for the study of higher-order network science.

A Blue Start: A large-scale pairwise and higher-order social network dataset

TL;DR

We present a large-scale Bluesky dataset that links pairwise following relationships with higher-order group structures via user-curated starter packs, addressing a gap in validating higher-order dynamics. The dataset aggregates 26.7 million users, 1.6 billion follows, and 301.3 thousand starter packs, collected from Bluesky APIs with careful anonymization and ethical considerations. Two data records are released in multiple formats: a deidentified starter-pack corpus with hyperedge and HIF representations, and a deidentified follows network with millions of directed edges, enabling joint analyses of pairwise and higher-order interactions. Technical validation confirms rich, heterogeneous degree distributions, dense higher-order overlap, and complementary information between the two representations, highlighting the dataset's utility for higher-order network science and graph learning applications. Data availability through SOMAR and accompanying code further support reproducible research and methodological development in higher-order social networks.

Abstract

Large-scale networks have been instrumental in shaping the way that we think about how individuals interact with one another, developing key insights in mathematical epidemiology, computational social science, and biology. However, many of the underlying social systems through which diseases spread, information disseminates, and individuals interact are inherently mediated through groups of arbitrary size, known as higher-order interactions. There is a gap between higher-order dynamics of group formation and fragmentation, contagion spread, and social influence and the data necessary to validate these higher-order mechanisms. Similarly, few datasets bridge the gap between these pairwise and higher-order network data. Because of its open API, the Bluesky social media platform provides a laboratory for observing social ties at scale. In addition to pairwise following relationships, unlike many other social networks, Bluesky features user-curated lists known as "starter packs" as a mechanism for social network growth. We introduce "A Blue Start", a large-scale network dataset comprising 26.7M users and their 1.6B pairwise following relationships and 301.3K groups representing starter packs. This dataset will be an essential resource for the study of higher-order network science.
Paper Structure (17 sections, 6 figures, 1 table)

This paper contains 17 sections, 6 figures, 1 table.

Table of Contents

  1. Background & Summary
  2. Methods
  3. Obtaining the list of users
  4. Collecting starter packs
  5. Collecting network data
  6. Anonymizing user data
  7. Limitations
  8. Ethical Considerations
  9. Data Records
  10. Bluesky starter packs
  11. Bluesky following network
  12. Technical Validation
  13. Starter pack data
  14. Following data
  15. Comparing the starter pack and following network data
  16. ...and 2 more sections

Figures (6)

  • Figure 1: Basic higher-order starter pack statistics. Panel (a) plots the probability distributions of both the number of starter packs to which a user account belongs and the number of starter packs that each user account has created. Panel (b) plots the starter pack size distribution, with the minimum and maximum starter pack sizes illustrated. Panel (c) plots the sizes of the connected components with the minimum and maximum starter pack sizes illustrated.
  • Figure 2: Temporal starter pack statistics. Panel (a) plots the number of starter packs created each day and panel (b) plots the distribution of time elapsed between when a Bluesky account was created and when that account created a starter pack. The dashed lines in panel (a) illustrate the following notable events: (1) Bluesky announces the release of starter packs as a feature on June 26th, 2024 blueskypbc_introducing_2024, (2) the Brazilian government bans the X platform on August 30th, 2024, (3) changes to the blocking feature on X are announced on October 16th, 2024 allowing blocked accounts to view posts of the blocking account, and (4) X announces new terms of service on November 15th, 2024 requiring users to consent to their posts being used to train X's artificial intelligence models.
  • Figure 3: $s$-line graph statistics of the starter pack network. The number of nodes and edges in the $s$-line graph of the starter pack network for all non-trivial values of $s$. In panels (a) and (b), the dashed line denotes the maximum starter pack size of 150.
  • Figure 4: Mesoscale starter pack statistics. Panel (a) plots the distribution of $k$-coreness, Panel (b) plots the probability of two user accounts appearing in $m$ different starter packs together, and Panel (c) plots the distribution of normalized entropy calculated for all starter packs. The illustrations of starter packs indicate at $s=0$, a starter pack completely contained in an inferred community and at $s=1$, a starter pack where every member belongs to a different community.
  • Figure 5: Basic following network statistics. Panel (a) plots the in-degree and out-degree distributions and panel (b) plots the frequency of strongly and weakly connected component sizes.
  • ...and 1 more figures