- Mystem
- TreeTagger
- Marmot
- Hunpos
Ruscorpora PoS markup was taken as a golden standard
len = 17000 words, good_answers = 0.70 (answers like data : [X,X,X,X] -> label : [X], where X is a PoS tag)
len = 499 words, good_answers = 0.69 (answers like data : [X,X,X,X] -> label : [X], where X is a PoS tag)
models/mymodel_SVC.pkl
accuracy = 0.9879
models/mymodel_DT.pkl
accuracy = 0.9859
models/mymodel_NB.pkl
accuracy = 0.9418
models/NER
accuracy = 0.9299