Expand description
Edge model for the context graph with Bayesian confidence scoring.
Each edge carries a confidence score that represents our belief in the relationship’s existence and current validity. Confidence is computed via multi-source Bayesian fusion and decays over time using temporal decay (TempValid framework, ACL 2024).
Design decisions:
- Noisy-OR for multi-path confidence (Google Knowledge Vault, KDD 2014)
- Temporal decay rates vary by discovery source
- Each edge tracks its full evidence provenance chain
Structs§
- Evidence
Source - A single piece of evidence supporting an edge’s existence.
- UcmEdge
- An edge in the context graph with confidence scoring and provenance.
Enums§
- Confidence
Tier - Display tier for confidence scores. Dual representation: continuous internally, three-tier for UI.
- Relation
Type - The type of relationship between two entities.