DataTerminology.comNEW
Also known as: Probabilistic Matching, Fuzzy Matching
A scoring-based record linkage technique that weights attribute similarities to estimate the likelihood of a true match.
DESCRIPTION
Weighted approximate matching computes a match score by combining attribute-level similarities with learned or expert-defined weights. It stems from probabilistic record linkage, notably the Fellegi–Sunter model, and balances precision and recall via thresholds. Typical implementations use similarity functions like Jaro–Winkler or edit distance, plus frequency-based weights. Blocking keys or indexing reduce candidate pairs for scalability. It is widely used in MDM, KYC, deduplication, healthcare registries, and customer 360. Governance defines thresholds, clerical review, and survivorship rules to resolve conflicts. Modern systems may learn weights from labeled pairs.
ALTERNATIVE DEFINITIONS
Wikipedia
Record linkage is the task of finding records in a data set that refer to the same entity across different data sources.
Defines the overarching task that weighted approximate matching supports.
Wikipedia
In computer science, approximate string matching is the technique of finding strings that match a pattern approximately (rather than exactly).
Highlights the string-similarity foundation used in many weighted matching systems.
RELATED TERMS
Entity Resolution Fellegi–Sunter Model Approximate String Matching Jaro–Winkler Distance Levenshtein Distance Blocking Entity Matching Survivorship
COMPETING TERMS
Deterministic Matching Exact Matching
RELATIONS
- Broader: Matching
KNOWLEDGE AREAS
Data Quality Data Management Data Integration & Interoperability Data Science AI
TARGET AUDIENCE
Data Engineer Data Scientist Data Steward Data Architect Data Analyst
EXTERNAL SOURCES
- Wikipedia - Record linkage
- python-record-linkage - Open-source record linkage toolkit documentation
- NIST DADS - Edit distance definition and context
- Microsoft Learn - DQS matching concepts in SQL Server
TOP QUESTIONS
What is Weighted Approximate Matching in practice?
It assigns weights to attributes and combines their similarity scores into a composite match score, enabling probabilistic decisions on whether two records refer to the same entity.
Category: Definition
Why is Weighted Approximate Matching used in MDM?
It improves matching accuracy across messy, heterogeneous sources by leveraging probabilistic evidence rather than brittle exact rules, yielding better deduplication and golden records.
Category: Relevancy
How do you tune Weighted Approximate Matching thresholds?
Use labeled pairs to plot precision-recall curves and choose thresholds that balance false matches and misses; incorporate clerical review for borderline cases.
Category: Requirements
What algorithms support Weighted Approximate Matching?
Common components include Jaro–Winkler, Levenshtein distance, cosine similarity on token sets, frequency-based weighting, and blocking or indexing strategies.
Category: Overview
What are common mistakes with Weighted Approximate Matching?
Ignoring frequency effects, over-weighting sparse fields, skipping standardization, and failing to evaluate on labeled data often degrade match quality.
Category: Mistakes
How does Weighted Approximate Matching compare to deterministic matching?
Deterministic rules are simple but rigid; weighted matching is probabilistic and robust to noise, though it requires data preparation and evaluation.
Category: Comparison
What data preparation improves Weighted Approximate Matching?
Normalize casing, remove noise, standardize formats (addresses, phones), parse compound fields, and create blocking keys to reduce comparisons.
Category: BestPractice
What risks exist when applying Weighted Approximate Matching?
False matches can merge distinct identities, while misses fragment entities; both can impact compliance and customer experience if unmitigated.
Category: Risk
How can machine learning enhance Weighted Approximate Matching?
Supervised models can learn optimal weights or decision boundaries from labeled pairs, improving accuracy over static, expert-defined weights.
Category: Topic
Where should a team start with Weighted Approximate Matching?
Begin by labeling a representative sample, selecting similarity functions per attribute, and iteratively tuning weights and thresholds with validation metrics.
Category: GettingStarted