Reproduce “FactKG: Fact Verification via Reasoning on Knowledge Graphs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiho Kim, Sungjin Park, Yeonsu Kwon, Yohan Jo, James Thorne, Edward Choi
Researchers at KAIST introduced FACTKG, the first large-scale fact verification dataset leveraging natural language claims and complex reasoning over DBpedia. Their experiments showed that models using knowledge graph evidence attained 77.65% overall accuracy, outperforming text-only baselines and improving verification for claims involving existence and negation by a considerable margin.
Key insights
- Fact verification systems explicitly reasoning over knowledge graph evidence generally outperform models relying solely on natural language claims.
- The structured nature of KGs enables clearer inferential links and potentially more explainable verification decisions.
- Training models on claims with diverse linguistic styles, including colloquial forms, can improve generalization rather than hinder it.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.