I was looking for ways of doing NER (Named Entity Recognition) in Portuguese and discovered that the spaCy guys have developed an open-source NER system for Portuguese using our work on Universal Dependencies for Portuguese (paper in https://www.aclweb.org/anthology/W17-6523/)
This was a great surprise! I love it!!!
I worry as I don't think the corpus Bosque is good enough. It is small and somewhat old and full of mistakes, but still, the numbers for NER are very good:
| NER ACCURACY | |
|---|---|
| NER F | 89.18 |
| NER Precision | 89.32 |
| NER Recall | 89.03 |
No comments:
Post a Comment