Coordinated link sharing on Facebook

Summary

This paper proposes a new method for detecting coordinated link-sharing behavior on Facebook that reduces reliance on post-timing signals, which adversaries can easily manipulate. The authors argue that the speed and frequency of link sharing across accounts follow consistent statistical regularities, and that deviations from these regularities provide a more evasion-resistant signature of coordination. They validate the approach on a large corpus of 11.2 million Facebook link posts drawn from roughly 16,000 sources, positioning the work as a methodological contribution to platform integrity research and the broader study of coordinated inauthentic behavior.

Key Contributions

• A coordination detection method that moves beyond timing-based signals, which are trivially gameable by sophisticated actors.
• Identification of statistical regularities in link-sharing speed and frequency that function as robust behavioral signatures.
• Empirical validation at scale using a corpus of 11.2 million Facebook link posts across ~16,000 sources.

Methods

• Assembly of a large-scale Facebook link-sharing dataset (11.2M posts, ~16K sources).
• Statistical characterization of the distributions of sharing speed and frequency across accounts.
• Development of a detection procedure based on departures from these regularities.
• Empirical evaluation of the procedure on the assembled corpus.

Findings

• Sharing speed and frequency exhibit stable, regular statistical patterns across accounts under normal conditions.
• Coordinated link-sharing activity produces detectable deviations from these regularities.
• The proposed signals function effectively at scale, suggesting practical viability for platform-level detection.

Connections

This work sits squarely in the methodological strand of coordinated inauthentic behavior detection that critiques timing-based co-sharing approaches; it relates closely to Graham2025-gp and Graham2026-fb–style network methods, and to robustness/evasion concerns raised in Bouchaud2026-lr and Luceri2025-tr. It also complements broader CIB detection efforts such as Minici2024-tf and Kulichkina2026-zk, which similarly seek behavioral signatures that go beyond temporal co-occurrence.