A measurable view of disinformation in the Slovak information space

SK-2SO is a doctoral project building an autonomous monitoring agent that detects, catalogues and warns about disinformation and manipulative content (FIMI) in the Slovak online space. It is developed under the DIMARL research initiative — the Disinformation and Information Manipulation Analysis and Research Lab.

Canonical taxonomy

Seven annotation layers, every value backed by a citable academic source — Wardle & Derakhshan, EUvsDisinfo, Card et al., SemEval-2023, DISARM, Nimmo.

Human gold standard

The gold corpus is built by independent multi-coder annotation; AI performance is measured against the inter-annotator agreement ceiling (Krippendorff's α), not against arbitrary percentages.

Defensive by design

The tool warns and educates. It does not censor, silence, or track individuals — the ethical boundaries are part of the architecture, not an afterthought.

How the project works

  1. Collection & archiving: items from the Slovak online space are archived with cryptographic integrity (BagIt + sha256), so every conclusion can be traced back to evidence.
  2. Human annotation: trained coders independently classify content along a seven-layer schema in the annotation portal; they cannot see each other's answers (blinding) and a submitted annotation is immutable.
  3. Agreement measurement: the reliability of the taxonomy is measured by inter-annotator agreement, which forms the ceiling against which the AI classifier is evaluated.
  4. AI classification: language models classify content along the same schema (prompt + JSON Schema, no training on sensitive data) with a strong focus on cost efficiency.

Methodology: Design Science Research. The project is run as doctoral research — rigour and citability take precedence over speed.