Reference

Entity resolution with explicit uncertainty

Behavior

WDC held-out result	Value	Decision implication
Pair precision	0.623	Roughly four in ten automatic links were wrong
Pair recall	0.507	About half of true matches were linked
Pair F1	0.559	URL fields produced too many errors for unattended merging
Review rate	0.476	The review band sends uncertain pairs to review

The run uses URLs from the compact WDC gold files. Product titles and identifiers belong in the next field-evidence run. Attributes provide another signal to measure there.

Example resolution trace

Feature	Agrees	Match probability	Non-match probability	Contribution
Email	Yes	0.99	0.01	Positive log-likelihood
Name	No	0.90	0.20	Negative log-likelihood

The summed score is compared with separate link and review thresholds. Ambiguous records remain review candidates.

The cascade spends expensive work after cheap deterministic checks. Ambiguous candidates enter a merge through a configured threshold or a review decision.

How it works

The caller assembles the cascade: block by tenant, scope, and entity type, compare normalized aliases, and optionally add fuzzy or embedding candidates. resolve_entity scores one supplied field-agreement comparison. It returns LINK, REVIEW, or DISTINCT and each field’s contribution. Canonical-ID selection and merging remain separate caller decisions.

Scores sum natural-log likelihood ratios. They are comparison scores, distinct from calibrated match probabilities. Estimate field probabilities on labeled matches and distinct pairs, then select thresholds for your tolerated merge error. Correlated fields can repeat the same evidence. Inspect their individual contributions and the resulting review volume.

Papers

Fellegi–Sunter: probabilistic record linkage

scope + type block normalized exact field/fuzzy score embedding candidates link · reject · review

resolution.py

from mari_kit.graph import FieldAgreement, ResolutionDecision, resolve_entity

resolution = resolve_entity([
    FieldAgreement(field="email", agrees=True,
        match_probability=0.99, nonmatch_probability=0.01),
    FieldAgreement(field="name", agrees=False,
        match_probability=0.90, nonmatch_probability=0.20),
], link_threshold=4.0, review_threshold=1.0)

if resolution.decision is ResolutionDecision.REVIEW:
    review_queue.put(candidate, resolution.contributions)