Repo · Data & extraction

JevSpan

Zero-shot named entity recognition library on Jev: text is split at punctuation, candidate windows are offered as one Choice question per entity…

Open github.com ↗

From the repository

Zero-shot named entity recognition on TypeSafe Jev. Define entity types in plain words — no training data, no GPU. 73.7 strict F1 across 12 Chinese/English NER benchmarks, with calibrated scores and auditable traces. 零样本命名实体识别。

benchmarkcalibrationchinesechinese-nlpclicrossnerentity-extractionfastapiinformation-extractionjevllmnamed-entity-recognitionnernlppythontext-miningtypesafezero-shotzero-shot-ner

How builders describe it

Zero-shot named entity recognition library on Jev: text is split at punctuation, candidate windows are offered as one Choice question per entity type, then verified and boundary-polished by further Choice calls. Types are defined in plain words; no training data or GPU is needed.

The decision Jev makes

Keep or drop a row, assign a label, or select one field value at a time.

Where it fits

Putting Jev next to the data: filtering rows with plain-English conditions, classifying records, labelling datasets and pulling structure out of text one small decision at a time. All 98 data & extraction projects →

Related projects