Statistical Co-occurrence Detection for Entity Relationship Discovery in Heterogeneous Knowledge Graphs

Key Points

• The aim is to evaluate statistical methods for edge detection in heterogeneous knowledge graphs to improve relationship discovery.
• Evaluated statistical methods: PMI, NPMI, G-test, and Granger causality.
• Assessed minimum corpus size requirements for effective edge detection.
• Established threshold selection criteria across four professional domains.
• Identified effective edge detection techniques in various domains.
• Determined minimum corpus size required for reliable analysis.
• Proposed criteria for threshold selection improved relationship discovery.