This is a Polish NER system trained using a Frustratingly Easy Domain Adaptation architecture. It is based on Polish BERT and supports different tagsets all using IOBES formats:
- Wikiann (LOC, PER, ORG)
- SlavNER 19/21 (EVT, LOC, ORG, PER, PRO)
- KPWr (EVT, LOC, ORG, PER, PRO)
- NKJP (DATE, GEOPOLIT, LOC, ORG, PER, TIME)
PER: person, LOC: location, ORG: organization, EVT: event, PRO: product, MISC: Miscellaneous, MEDIA: media, ART: Artifact, TIME: time, DATE: date, GEOPOLIT: Geopolitical
You can select the tagset to use in the output by configuring the model. This model manages differently uppercase words.
More information about the model can be found in the paper (https://aclanthology.org/2021.bsnlp-1.12.pdf) and GitHub repository (https://github.com/EMBEDDIA/NER_FEDA).
- Downloads last month
- 28