End-to-End Evaluation Framework for Evidence-Graded Knowledge Graphs: From Raw Sources to Ranked Edges

Key Points

• The aim is to develop a comprehensive evaluation framework for evidence-graded knowledge graphs from raw sources to ranked edges.
• Defined stage-level quality metrics for evaluating data
• Conducted cascade error propagation analysis to identify error impacts
• Implemented ablation methodology for testing the framework.
• Establishment of a reliable framework for assessing knowledge graph quality
• Identification of critical error propagation stages
• Demonstrated improved ranking of edges through the evaluation process.

Abstract