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

Key Points

• The research aims to explore statistical methods for detecting relationships in heterogeneous knowledge graphs.
• Evaluated statistical techniques: PMI, NPMI, G-test, and Granger causality.
• Focused on observation-level edge detection.
• Derived minimum corpus size requirements and threshold selection criteria.
• Analyzed data across four professional domains.
• Identified effective statistical methods for relationship detection in knowledge graphs.
• Provided insights on minimum corpus size and threshold criteria.
• Validated the robustness of methods across multiple domains.