fg-zettelkasten

Mapping Nefarious Social Media Actors to Speed-up Covid-19 Fact-checking

3 min read

Mapping Nefarious Social Media Actors to Speed-up Covid-19 Fact-checking

Summary

This report describes the MINE-FACTS project, which built and operationally tested a content-agnostic prototype tool to help Italian fact-checkers (notably Facta.news) surface problematic COVID-19 information on Facebook faster than manual triage allows. Building on Coordinated Link Sharing Behavior (CLSB) detection via the CooRnet R package, the authors map Italian “nefarious” coordinated networks at macro, meso, and micro scales, trace the rise of an Italian covid-skeptic coordinated network grafted onto pre-existing political clusters, and document the tactics these networks use to evade platform enforcement. A one-month live test showed the prototype flagged a substantially higher share of problematic content than routine third-party fact-checking workflows.

Key Contributions

  • An operational fact-checking prototype integrating CooRnet, CrowdTangle, and the IFCN CoronaVirusFacts database into Italian newsroom workflows.
  • A multi-scale (macro/meso/micro) map of Italian coordinated networks spreading COVID-19 misinformation on Facebook.
  • Documentation of novel evasion tactics: first-comment link placement, image-macro “link laundering,” shared Google AdSense IDs across domains, and Page rebranding.
  • Empirical evidence that behavior-based, content-agnostic detection outperforms content-based triage in identifying problematic content.
  • A methodological demonstration of iterative CooRnet seeding, where links from already-identified coordinated accounts are used to discover new ones.
  • OSINT-driven attribution (WHOIS, reverse WHOIS, AdSense/Analytics IDs, Wayback Machine) linking specific domains and operators to disinformation operations.

Methods

The authors combine Meta’s URL Shares dataset and CrowdTangle queries seeded from 212 IFCN-listed Italian false claims to build URL inventories, then apply CooRnet iteratively to detect CLSB across Facebook Pages, groups, and verified profiles. Macro-level structure is visualized with Force Atlas 2 network layouts; meso-level dynamics are examined through four qualitative case studies framed by Camille François’s Actors-Behavior-Content (ABC) model; micro-level analysis assesses individual flagged posts. OSINT techniques attribute domains and actors, and a one-month November 2021 deployment with Facta.news tests the prototype in production.

Findings

  • February–June 2020 top Italian Facebook links favored partisan, oversimplified narratives, including a Tgcom24 piece amplifying a bioweapon claim (15M+ views) and removed 5G-COVID YouTube videos.
  • Initial CooRnet runs identified 30 Pages (2.1M followers) and 308 groups (2.73M members) engaged in CLSB; conspiracy-group interactions roughly tripled during lockdown.
  • Iterative seeding eventually surfaced 2,151 coordinated accounts (242 Pages, 1,900 groups) across 89 components, with the largest covid-skeptic cluster overlapping League- and Five Star Movement-affiliated accounts and bridging groups linking political, religious, and covid-skeptic communities.
  • Case studies revealed (i) a Blogspot conspiracy-blog network monetized via shared AdSense IDs, (ii) a 62-group Catholic network occasionally injected with covid-skeptic content, (iii) a 31-Page network laundering traffic to reputable GEDI-group outlets, and (iv) the Mag24 image-macro network traced via WHOIS/AdSense to a specific operator.
  • In the November 2021 test, ~40% of prototype-surfaced links/posts were rated problematic, versus ~28% in routine third-party fact-checking.
  • The comments-to-shares ratio is a useful ranking signal: non-problematic content is dominated by shares, problematic content shows relatively more commenting.
  • Unprincipled “celebrities” (Montagnier, Viganò, Kennedy) and covid-skeptic communities form a symbiotic amplification loop, and quoted false claims often evade fact-check labeling because the quote itself is authentic.

Connections

This paper sits squarely in the CooRnet/CLSB research program developed by the same group, extending Giglietto2023-fa71a001 and Marino2024-a678b03f with COVID-specific Italian network mapping and a fact-checker-facing tool; it also resonates with Ghezzi2023-8bebc91f on Italian problematic health information. Its content-agnostic, behavior-based detection logic connects to broader work on coordinated inauthentic behavior such as Luceri2025-tr, Minici2024-tf, and Graham2025-gp, and to evasion-and-adaptation themes explored in Kuznetsova2025-nu and Schroeder2026-im.

Graph View

Backlinks