FactKG: Fact Verification via Reasoning on Knowledge Graphs

• DOI: 10.18653/v1/2023.acl-long.895
• Publication Date: 2023-08-05T00:57:42Z
• Authors: Jiho Kim; Sungjin Park; Yeonsu Kwon; Yohan Jo; James Thorne; Edward Choi
• Keywords: Negation; Conjunction (astronomy); Knowledge graph; Baseline (sea)

In real world applications, knowledge graphs (KG) are widely used in various domains (e.g. medical applications and dialogue agents). However, for fact verification, KGs have not been adequately utilized as a source. can be valuable source verification due to their reliability broad applicability. A KG consists of nodes edges which makes it clear how concepts linked together, allowing machines reason over chains topics. there many challenges understanding these machine-readable map information text. To enable the community better use KGs, we introduce new dataset, FactKG: Fact Verificationvia Reasoning on Knowledge Graphs. It 108k natural language claims with five types reasoning: One-hop, Conjunction, Existence, Multi-hop, Negation. Furthermore, FactKG contains linguistic patterns, including colloquial style well written increase practicality. Lastly, develop baseline approach analyze reasoning types. We believe advance both practicality KG-based verification.