Composite Weight Functions for Evidence-Graded Knowledge Graphs: Evaluating Temporal Decay, Source Credibility, and Cross-Source Confirmation

Key Points

• This research assesses a composite weight function for edges in knowledge graphs, focusing on factors that influence their reliability.
• Evaluated composite weight function w(e) incorporating factors like temporal decay and source credibility.
• Analyzed different types of temporal decay: exponential, hyperbolic, and bounded exponential.
• Used shrinkage estimation to assess source credibility and cross-source confirmation inspired by Dempster-Shafer theory.
• Conducted Monte Carlo simulations with 50,000 synthetic edges to test ranking stability.
• Achieved high ranking stability with Kendall's τ greater than 0.91 across simulations.
• Demonstrated distinct impacts of different temporal decay models on knowledge graph reliability.
• Validated the effectiveness of cross-source confirmation on enhancing credibility measures.