Stratified Evaluation of Evidence-Graded Knowledge Graph Edges: A 1200-Edge Multi-Domain Assessment

Key Points

• The aim is to evaluate 1,200 edges in knowledge graphs across different domains and evidence levels.
• Conducted a stratified evaluation across four domains and four evidence levels.
• Applied Cochran-Mantel-Haenszel analysis to investigate domain effects.
• Used Firth logistic regression for analysis of sparse cells.
• Calculated Wilson confidence intervals for reliability of estimates.
• Identified significant domain effects in the evaluation.
• Provided insights into the reliability of knowledge graph edges across multiple domains.

Abstract