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.