Jev Gallery
← Back to collection

Docs / source reviewed

jlink

Link records across two tables using a natural-language matching rule

How this project uses Jev

Input
Candidate record pairs, selected fields and a user-defined same-entity rule
Jev decides
Whether two records refer to the same company, publication, product or other entity
Code executes
Local code proposes candidates, obtains Jev scores and resolves one-to-one or other matching constraints into a link table with audit sampling

Evidence and limitations

True matches omitted by candidate generation cannot be recovered by the model; normalized exact matches can bypass inference. Author dataset results do not establish performance on other tables. We did not reproduce matching quality or costs.

This project has not been run independently here. Author-reported results are not independently verified results.

Original sources

Explore related use cases

Browse this category →

Learn the approach