Created
February 14, 2023 11:38
-
-
Save vanatteveldt/f07b8ca5778469879455066362404d0c to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Text: Wij zijn Wouter van Atteveldt en Nel Ruigrok. | |
10.000 nieuwe stamcel- en bloeddonoren na oproep PSV-perschef Thijs Slegers. | |
Ongeneeslijk ziek Opvallend veel mannen meldden zich aan als donor na een oproep van de ongeneeslijk zieke Slegers, Matchis kreeg 7.000 nieuwe aanmeldingen, Sanquin 3.000. | |
Model: pdelobelle/robbert-v2-dutch-ner | |
NER output: | |
{'entity_group': 'PER', 'score': 0.9998577, 'word': ' Wouter van Atte', 'start': 9, 'end': 24, 'full_word': 'Wouter van Atteveldt'} | |
{'entity_group': 'PER', 'score': 0.9999995, 'word': ' Nel Ruig', 'start': 33, 'end': 41, 'full_word': 'Nel Ruigrok'} | |
{'entity_group': 'MISC', 'score': 0.99988997, 'word': ' PSV', 'start': 96, 'end': 99, 'full_word': 'PSV'} | |
{'entity_group': 'PER', 'score': 0.99999857, 'word': ' Thijs Sleg', 'start': 109, 'end': 119, 'full_word': 'Thijs Slegers'} | |
{'entity_group': 'PER', 'score': 0.92822623, 'word': ' Sleg', 'start': 231, 'end': 235, 'full_word': 'Slegers'} | |
{'entity_group': 'PER', 'score': 0.75320077, 'word': ' Match', 'start': 240, 'end': 245, 'full_word': 'Matchis'} | |
{'entity_group': 'PER', 'score': 0.79860216, 'word': ' Sanquin', 'start': 281, 'end': 288, 'full_word': 'Sanquin'} |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
from transformers import pipeline | |
from transformers import AutoTokenizer | |
import unicodedata | |
model = "wietsedv/bert-base-dutch-cased-finetuned-conll2002-ner" | |
model = "pdelobelle/robbert-v2-dutch-ner" # geeft zelfde probleem | |
text = """Wij zijn Wouter van Atteveldt en Nel Ruigrok. | |
10.000 nieuwe stamcel- en bloeddonoren na oproep PSV-perschef Thijs Slegers. | |
Ongeneeslijk ziek Opvallend veel mannen meldden zich aan als donor na een oproep van de ongeneeslijk zieke Slegers, Matchis kreeg 7.000 nieuwe aanmeldingen, Sanquin 3.000. | |
""" | |
class WordFinder: | |
def __init__(self, model, text): | |
self.model = model | |
tokenizer = AutoTokenizer.from_pretrained(self.model) | |
revocab = {id: word for (word, id) in tokenizer.vocab.items()} | |
tokens = tokenizer(text, return_offsets_mapping=True) | |
self.words = [revocab[id] for id in tokens['input_ids']] | |
self.offsets = tokens['offset_mapping'] | |
self.tokenmap = {start: i for i, (start, end) in enumerate(tokens['offset_mapping']) if start != end} | |
def start_of_word(self, word): | |
if self.model == "wietsedv/bert-base-dutch-cased-finetuned-conll2002-ner": | |
return not word.startswith("##") | |
if self.model == "pdelobelle/robbert-v2-dutch-ner": | |
return unicodedata.category(word[:1]).startswith("P") or word.startswith("\u0120") | |
def get_full_word(self, start, end): | |
end_of_name = end | |
for j in range(self.tokenmap[start], len(self.words)): | |
end_of_token = self.offsets[j][1] | |
if (end_of_token > end) and self.start_of_word(self.words[j]): | |
break | |
end_of_name = end_of_token | |
return text[start:end_of_name] | |
classifier = pipeline("ner", model = model, aggregation_strategy='simple') | |
output = classifier([text]) | |
wf = WordFinder(model, text) | |
print("\nText:", text) | |
print("\nModel:", model) | |
print("\nNER output:") | |
for token in output[0]: | |
token['full_word'] = wf.get_full_word(token['start'], token['end']) | |
print(token) | |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment