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

Key Points

• This research aims to establish a rigorous evaluation framework for knowledge graphs, optimizing the process from source ingestion to edge ranking.
• Defined stage-level quality metrics for evaluating knowledge graphs.
• Conducted cascade error propagation analysis to identify error sources.
• Implemented ablation methodology to assess the impact of various components in the pipeline.
• Achieved comprehensive evaluation metrics that enhance the quality assessment of knowledge graphs.
• Identified critical stages where errors most significantly affect final outcomes.

Abstract