Cross Source Evidence Correlation
Cross Source Evidence Correlation is a systemic layer in GEO.or.id that measures relational consistency between evidence items originating from independent sources. It is designed to detect convergence, divergence, and hidden dependency patterns across distributed information streams.
This layer operates on top of the Evidence Graph and feeds directly into the Evidence Weighting Engine for dynamic trust recalibration.
1. Core Objective
The main objective is to determine whether multiple independent sources are genuinely reinforcing the same signal or simply repeating shared lineage.
- Detect multi-source agreement patterns
- Identify hidden source dependency chains
- Separate true convergence from duplication noise
- Strengthen high-confidence evidence clusters
Connected systems: Semantic Matching Layer, Evidence Retrieval Binding
2. Correlation Mechanism
The system evaluates correlation across three structural dimensions: semantic similarity, provenance independence, and temporal alignment.
2.1 Semantic Correlation
Measures how closely two evidence items align in meaning space after normalization and entity resolution.
See: Entity Relationships Ontology
2.2 Provenance Independence
Evaluates whether sources are structurally independent or derived from shared upstream datasets.
Connected: Evidence Provenance Model
2.3 Temporal Alignment
Checks whether correlated signals appear within meaningful time windows or are artificially synchronized.
Related: Evidence Lifecycle Management
3. Correlation Scoring Model
Correlation strength is computed using a weighted composite function:
C(e1, e2) = (S × α) + (P × β) + (T × γ)
- S = Semantic Similarity Score
- P = Provenance Independence Score
- T = Temporal Alignment Score
These weights are dynamically adjusted based on domain volatility and retrieval density.
4. Correlation Types
Not all correlations indicate agreement. The system distinguishes multiple correlation types.
- Positive Correlation: Independent sources reinforce same claim
- Negative Correlation: Sources systematically contradict each other
- Spurious Correlation: Similarity caused by shared upstream origin
- Contextual Correlation: Alignment only under specific conditions
Handled by: Evidence Conflict Detection
5. Cluster Formation
Correlated evidence is grouped into dynamic clusters representing shared informational intent. These clusters are not static and evolve as new evidence arrives.
Cluster outputs feed into: Entity Prioritization Layer, Re-ranking System
6. Failure Modes
- False convergence due to syndicated content duplication
- Hidden correlation masking contradictory signals
- Overweighting high-volume but low-diversity sources
Mitigation is handled via: Source Selection Analysis, Query Variation Testing
7. Role in GEO Architecture
Cross Source Evidence Correlation functions as a trust amplification layer. It does not create truth; it strengthens or weakens confidence in existing evidence clusters based on structural independence.
It is tightly integrated with: Retrieval Authority Model, AI Ground Truth Framework