JLink
Link research records using a plain-English matching rule.
What Jev decides
The probability that a record pair represents the same entity.
JLink asks Jev whether two records describe the same entity under a researcher-defined rule and preserves a probability for every pair.
- Jev role
- Assess whether two records match under the supplied rule.
- Input
- Two candidate records and the researcher’s plain-English match definition.
- Output
- A match probability retained for thresholding and audit.
- Decision frequency
- Once per candidate record pair.
- Fallback
- Researchers can review sampled pairs and set explicit thresholds or margins.
How it works
Overview
Researchers state what counts as a match, such as whether a parent and subsidiary should be treated as different firms. JLink compares candidate pairs with Jev, stores probabilities, and offers thresholds, audit samples, and methods text for reproducible analysis. Interfaces include Python, CLI, Stata, and R; the README reports benchmark results on selected linkage datasets.
- Language
- Python
- License
- MIT
- Author
- keltokhy
Sources
Related projects
pg-jev
Use Jev decisions directly inside PostgreSQL filters, rankings, and classifications.
What Jev decides
Whether a row matches, how strongly it matches, or which supplied label best describes it.
jev-spam-eval
A reproducible email-classification study comparing zero-shot Jev decisions with trained TF-IDF baselines.
What Jev decides
Whether an email is ham, spam, or phishing in the main experiment.
Jev World
Ask Jev whether map coordinates are land or water, then replay the answers.
What Jev decides
Whether each supplied map coordinate is land.