Configuring a spotter
Spotters are configured in pipeline YAML. Each entry declares a vocabulary with optional aliases and a link target.
pipeline:
language: auto
spotters:
- kind: dictionary
name: products
entries:
- value: MacBook Air
aliases: [MBA, "MacBook Air 2024"]
link_to: Device:MBA-2024 # links to this node key on match
- value: ThinkPad X1
aliases: [X1C]
link_to: Device:TPX1
min_confidence: 0.85
exclusions:
- context_includes: ["e.g.", "for example"] # suppress false positives
Confidence thresholds — a practical guide:
| Threshold | Use for |
|---|---|
| ≥ 0.60 | Show entity highlight in UI |
| ≥ 0.70 | Use as a facet |
| ≥ 0.85 | Auto-link to graph node |
| ≥ 0.95 | Auto-merge / deduplicate |
Start conservative (0.85 for linking). Lower once you've validated precision on real data.