Iannucci, L., Matakos, A., & Kivelä, M. (2026). Statistical detectability of coordination. arXiv.
Summary
This paper interrogates a foundational assumption in coordination detection on social media: that dense subgraphs or behavioural clusters in account-similarity networks indicate genuine coordination. The authors argue that prevailing pipelines — which project account-entity interactions into similarity networks and then detect dense subgraphs or partition users into clusters — tend to surface co-activated accounts or forcibly group all users regardless of whether true coordination exists. Reframing the problem as one of statistical detectability, the work asks when coordinated activity can actually be distinguished from ordinary or incidental co-activity, and suggests that this distinguishability has statistical limits that should be characterised rather than taken for granted.
Key Contributions
- Reframes coordination detection as a question of statistical detectability rather than mere co-activity identification.
- Critiques the assumptions underlying similarity-network and clustering approaches to coordinated inauthentic behaviour detection.
- Highlights that detecting dense subgraphs or clustering users does not by itself establish the presence of genuine coordination.
Methods
The authors analyse standard coordination-detection pipelines that project account-entity interactions into account-similarity networks based on temporal proximity, shared action counts, or behavioural similarity. They examine dense-subgraph detection and behavioural clustering as the mechanisms used to identify coordinated groups, scrutinising whether these methods can separate coordinated behaviour from noise or incidental similarity.
Findings
- Prevailing methods identify co-activated accounts or forcibly cluster all users, which does not demonstrate that true coordination is present.
- There is a need to assess whether detected coordination is statistically distinguishable from noise or incidental similarity.
- The abstract is truncated, so detailed quantitative results are not available beyond the identification of these methodological limitations.
Connections
This work provides a critical, statistical lens on the similarity-network and clustering methods widely used in coordinated inauthentic behaviour research, connecting directly to related work by the same group in Iannucci2025-eg. Its concerns about distinguishing genuine coordination from incidental co-activity speak to foundational detection frameworks such as Nizzoli2020-cf and Luceri2025-tr, as well as the coordinated-link-sharing tradition represented by Giglietto2020-9d8acdd7 and Giglietto2022-0e951ac5.