Evaluation on PER, LOC, ORG tag scheme and BIOES tag sheme and how to use BERT emeddings

jiesutd / NCRFpp

NCRF++, a Neural Sequence Labeling Toolkit. Easy use to any sequence labeling tasks (e.g. NER, POS, Segmentation). It includes character LSTM/CNN, word LSTM/CNN and softmax/CRF components.

Apache License 2.0

1.88k stars 446 forks source link

Evaluation on PER, LOC, ORG tag scheme and BIOES tag sheme and how to use BERT emeddings #129

Open alimghl opened 5 years ago

alimghl commented 5 years ago

Scenario: use My own RNN model (very simple model ) on my dataset by using (ORG, PER, LOC, O)tag scheme my model gives score about (F1 = 0.775) but when i use NCRFpp to train my model and it gives me F1 score = -1 and after debugging I changed my data format according to the requirement (BIOES tag sheme used) and then it gives my very poor results (F1 = 0.554) Q-1 how do you compute f1-Score replacing tag with original or just weighted average? Q-2 if I want to use bert Embedding How can I use?

baiziyuandyufei commented 4 years ago

The tag Schema needs to be consistent with the format in the repository. For example: Zine B-PER al-Abidine I-PER Ben I-PER Ali E-PER