Blocking evaluation documentation

MATCHFORGE PLAYGROUND

Blocking evaluation

Configure candidate generation, label known record pairs, and evaluate how effectively your blocking strategy reduces comparisons while retaining genuine matches.

01 / INPUT

Two datasets

Source and target records

02 / GENERATE

Candidate pairs

Apply blocking rules

03 / EVALUATE

Ground truth

Calculate three metrics

01

Input datasets

Provide the source and target records used for candidate generation.

A

Source dataset

5 records · 3 columns

customer_nameemailcity

Edit cell values directly or add and remove records.

B

Target dataset

6 records · 3 columns

full_nameemail_addresslocation

Edit cell values directly or add and remove records.

02

Blocking configuration

Choose the fields and methods used to generate candidate pairs.

Configuration
Configuration
Default settings

OR logic: A pair is retained if it matches at least one blocking rule. Duplicate candidate pairs are removed.

Normalization settings

Standardize values before generating blocking keys.

▼
03

Ground-truth labels

Label known source-target pairs as genuine matches or non-matches.

6 labelled pairs

These labels are used to evaluate candidate generation.

Only explicitly labelled pairs are evaluated. Unlabelled candidate pairs are not automatically treated as false positives. Use representative labels for meaningful candidate precision.

Evaluate blocking strategy

Generate candidates and calculate all three evaluation metrics.