DS@GT ARC at CheckThat! 2026: A Verdict-Anchored Pipeline for

DS@GT ARC at CheckThat! 2026: A Verdict-Anchored Pipeline for Generating Fact-Checking Articles


Authors: Hoang Thanh Thanh Truong, Shuyu Tian

arXiv: 2611.05382 [cs.CL]

Submitted: 30 July 2026

Venue: CLEF 2026 Working Notes, 21–24 September 2026, Jena, Germany

Subjects: Computation and Language (cs.CL)

ACM Classes: I.2.7; H.3.3

DOI: https://doi.org/10.48550/arXiv.2611.05382


Abstract


Misinformation spreads faster than professional fact-checkers can manually verify it and write articles explaining their verdicts. This creates a pressing need to automate the production of fact-checking articles. This paper details work by the DS@GT ARC team for CLEF 2026 CheckThat! Lab Task 3: Fact-Checking Article Generation. The goal is to generate a full fact-checking article from a claim, its original verdict, and a set of scraped source documents. Our team built a four-stage pipeline that (1) cleans noisy evidence, (2) retrieves and clusters relevant sources, (3) generates the article by conditioning Llama-3.1-8B-Instruct on the original fact-checker verdict, and (4) recovers inline citations through deterministic post-processing. We achieved a mean score of 0.428 and ranked 3rd out of 9 teams for CLEF 2026 CheckThat! Lab Task 3, compared to a baseline mean score of 0.272. Our code is available on GitHub at https://github.com/dsgt-arc/checkthat-2026-task3.


Method Overview


The proposed system is a four-stage, verdict-anchored pipeline:


  1. Evidence cleaning — noisy scraped documents are preprocessed to remove irrelevant or low-quality content.
  2. Retrieval and clustering — relevant sources are retrieved and grouped to support coherent article construction.
  3. Verdict-conditioned generation — Llama-3.1-8B-Instruct generates the article conditioned on the original fact-checker verdict, ensuring alignment with the established conclusion.
  4. Deterministic citation recovery — inline citations are restored through post-processing, improving traceability and verifiability.

  5. Results


    • Mean score: 0.428
    • Ranking: 3rd out of 9 teams
    • Baseline mean score: 0.272

    Additional Metadata


    • Comments: CLEF 2026 Working Notes, 21–24 September 2026, Jena, Germany
    • Cite as: arXiv:2611.05382 [cs.CL]
    • Submission history: [v1] Submitted 30 July 2026

    Code


    Available on GitHub: https://github.com/dsgt-arc/checkthat-2026-task3

    via ArXiv CL+LG

Related