Liner2.6 NER NKJP model
The package contains a pre-trained Liner2 (https://github.com/CLARIN-PL/Liner2) model for recognition named entities according to NKJP guidelines. The model was trained on the NKJP corpus (http://nkjp.pl/) and evaluated in the PolEval 2018 Task 2 (http://poleval.pl/tasks/).
The model won third place with the following results: Exact — 0.778, Overlap — 0.818, Final — 0.810.
* NKJP corpus in TEI format — http://clip.ipipan.waw.pl/NationalCorpusOfPolish?action=AttachFile&do=view&target=NKJP-PodkorpusMilionowy-1.2.tar.gz
* PolEval 2018 Task 2 evaluation corpus — http://mozart.ipipan.waw.pl/~axw/poleval2018/