Pre-trained Model

[travellertea]

Hello There,

Question 1: url = 'https://storage.googleapis.com/chakki/datasets/public/models.zip' What is this model trained on for NER tagging?

Question 2: Why is the trained model performing significantly better with the trained model, compared to when I train a model with Conll2003 data and evaluate on the test set?

Cheers, TZ

[Hironsan]

Thank you for your question. My answers are as follows:

Answer 1:

This model is based on Bidirectional LSTM-CRF.

• Neural Architectures for Named Entity Recognition

This is trained by CoNLL2003 dataset.

Answer 2:

Because the published model is trained by CoNLLL2003 all(train + valid + test) datasets.

In anaGo 1.0.0, I released new trained dataset. This is trained by CoNLL2003 train + valid datasets. I think it is comparable to some papers score.

• https://storage.googleapis.com/chakki/datasets/public/ner/models_en.zip

with best regards