skip to content

[ paper · 2026 ]_

Engineered Homophily in Higher-Order Network Ties: A Case Study of Bluesky Starter Packs

[ what it found ]

  • Disciplinary homophily in academic starter packs is extreme (H = 0.85); geographic homophily is moderate (H = 0.33).
  • 88.4% of users appear in only one pack, sharply limiting the cross-disciplinary weak ties Granovetter identified as the useful ones.
  • Starter packs make homophily a platform affordance rather than an emergent property — the paper terms this "engineered homophily".
  • Measured over 387 packs: 22,026 users and 25,281 memberships, via bipartite projection and Louvain community detection.

[ abstract ]

Bluesky's "starter packs" — curated lists of up to 150 accounts a new user can follow in a single click — create many social connections at once, a higher-order network tie with no equivalent on older platforms. Using a sample of 387 academic starter packs (22,026 users, 25,281 memberships), this study builds bipartite and projected networks, runs Louvain community detection, and measures disciplinary and geographic homophily. Disciplinary homophily is extreme (H = 0.85), 88.4% of users appear in only one pack — sharply limiting the cross-disciplinary weak-tie bridges Granovetter identified — and geographic homophily is moderate (H = 0.33). The paper argues that starter packs operationalise homophily as a platform affordance rather than an emergent property, a pattern it terms "engineered homophily," and asks how classical concepts in social network analysis must be rethought when a single curated act bundles many connections at once.

[ document ]

[ the full paper, in the browser ]