Task: Named Entity Recognition
Model: BERT
Lang: IT
Model description
This is a
BERT [1] cased model for the
Italian language, fine-tuned for
Named Entity Recognition (
Person,
Location,
Organization and
Miscellanea classes) on the
WikiNER dataset
[2], using
BERT-ITALIAN (
bert-base-italian-cased) as a pre-trained model.
This is a cased, base size BERT model. If you are looking for a lighter (but slightly less accurate) cased model, you can refer to:
https://huggingface.co/osiria/distilbert-italian-cased-nerIf you are looking for an uncased model, you can refer to:
https://huggingface.co/osiria/bert-italian-uncased-nerTraining and Performances
The model is trained to perform entity recognition over 4 classes: PER (persons), LOC (locations), ORG (organizations), MISC (miscellanea, mainly events, products and services). It has been fine-tuned for Named Entity Recognition, using the WikiNER Italian dataset plus an additional custom dataset of manually annotated Wikipedia paragraphs.
The WikiNER dataset has been splitted in 102.352 training instances and 25.588 test instances, and the model has been trained for 1 epoch with a constant learning rate of 1e-5.
The performances on the test set are reported in the following table:
| Recall | Precision | F1 |
|---|
| 93.35 | 92.22 | 92.78 |
The metrics have been computed at the token level and then macro-averaged over the 4 classes.
Then, since WikiNER is an automatically annotated (silver standard) dataset, which sometimes contains imperfect annotations, an additional fine-tuning on ~3.500 manually annotated paragraphs has been performed.
Quick usage
1from transformers import BertTokenizerFast, BertForTokenClassification
2
3tokenizer = BertTokenizerFast.from_pretrained("osiria/bert-italian-cased-ner")
4model = BertForTokenClassification.from_pretrained("osiria/bert-italian-cased-ner")
5
6from transformers import pipeline
7ner = pipeline("ner", model = model, tokenizer = tokenizer, aggregation_strategy="first")
8
9ner("Mi chiamo Marco Rossi, vivo a Roma e lavoro per l'Agenzia Spaziale Italiana nella missione Prisma")
10
11[{'entity_group': 'PER',
12 'score': 0.99910736,
13 'word': 'Marco Rossi',
14 'start': 10,
15 'end': 21},
16 {'entity_group': 'LOC',
17 'score': 0.9973786,
18 'word': 'Roma',
19 'start': 30,
20 'end': 34},
21 {'entity_group': 'ORG',
22 'score': 0.9987071,
23 'word': 'Agenzia Spaziale Italiana',
24 'start': 50,
25 'end': 75},
26 {'entity_group': 'MISC',
27 'score': 0.9625836,
28 'word': 'Prisma',
29 'start': 91,
30 'end': 97}]
References
Limitations
This model is mainly trained on Wikipedia, so it's particularly suitable for natively digital text from the world wide web, written in a correct and fluent form (like wikis, web pages, news, etc.). However, it may show limitations when it comes to chaotic text, containing errors and slang expressions
(like social media posts) or when it comes to domain-specific text (like medical, financial or legal content).
License
The model is released under Apache-2.0 license
