osiria
/

minilm-l6-h384-italian-cased-ner

 ---
 license: mit
+language:
+- it
+widget:
+- text: "Mi chiamo Marco Rossi, vivo a Roma e lavoro per l'Agenzia Spaziale Italiana"
+  example_title: "Example 1"
 ---
+--------------------------------------------------------------------------------------------------
+<body>
+<span class="vertical-text" style="background-color:lightgreen;border-radius: 3px;padding: 3px;"> </span>
+<br>
+<span class="vertical-text" style="background-color:orange;border-radius: 3px;padding: 3px;">    Task: Named Entity Recognition</span>
+<br>
+<span class="vertical-text" style="background-color:lightblue;border-radius: 3px;padding: 3px;">    Model: MiniLM</span>
+<br>
+<span class="vertical-text" style="background-color:tomato;border-radius: 3px;padding: 3px;">    Lang: IT</span>
+<br>
+<span class="vertical-text" style="background-color:lightgrey;border-radius: 3px;padding: 3px;">  </span>
+<br>
+<span class="vertical-text" style="background-color:#CF9FFF;border-radius: 3px;padding: 3px;"> </span>
+</body>
+--------------------------------------------------------------------------------------------------
+<h3>Model description</h3>
+This is a <b>MiniLMv2</b> <b>[1]</b> cased model for the <b>Italian</b> language, fine-tuned for <b>Named Entity Recognition</b> (<b>Person</b>, <b>Location</b>, <b>Organization</b> and <b>Miscellanea</b> classes) on the [WikiNER](https://figshare.com/articles/dataset/Learning_multilingual_named_entity_recognition_from_Wikipedia/5462500) dataset <b>[2]</b>, using <b>BERT-ITALIAN</b> ([bert-base-italian-cased](https://huggingface.co/osiria/bert-base-italian-cased)) as a pre-trained model.
+<h3>Training and Performances</h3>
+The model is trained to perform entity recognition over 4 classes: <b>PER</b> (persons), <b>LOC</b> (locations), <b>ORG</b> (organizations), <b>MISC</b> (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 |
+| ------ | ------ |  ------ |
+| 80.84  | 85.32 | 82.85 |
+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.
+<h3>Quick usage</h3>
+```python
+from transformers import AutoTokenizer, AutoModelForTokenClassification
+import string
+tokenizer = AutoTokenizer.from_pretrained("osiria/minilm-l6-h384-italian-cased-ner")
+model = AutoModelForTokenClassification.from_pretrained("osiria/minilm-l6-h384-italian-cased-ner")
+from transformers import pipeline
+ner = pipeline("ner", model = model, tokenizer = tokenizer, aggregation_strategy="simple")
+text = "Mi chiamo Marco Rossi, vivo a Roma e lavoro per l'Agenzia Spaziale Italiana, dove mi occupo del controllo del satellite Prisma"
+for p in string.punctuation:
+    text = text.replace(p, " " + p + " ")
+ner(text)
+[{'entity_group': 'PER',
+  'score': 0.8044705,
+  'word': 'Marco Rossi',
+  'start': 10,
+  'end': 21},
+ {'entity_group': 'LOC',
+  'score': 0.8467228,
+  'word': 'Roma',
+  'start': 32,
+  'end': 36},
+ {'entity_group': 'ORG',
+  'score': 0.72725743,
+  'word': 'Agenzia Spaziale Italiana',
+  'start': 54,
+  'end': 79},
+ {'entity_group': 'MISC',
+  'score': 0.7243446,
+  'word': 'Prisma',
+  'start': 126,
+  'end': 132}]
+```
+You can also try the model online using this web app: https://huggingface.co/spaces/osiria/minilm-l6-h384-italian-cased-ner
+<h3>References</h3>
+[1] https://arxiv.org/abs/2012.15828
+[2] https://www.sciencedirect.com/science/article/pii/S0004370212000276
+<h3>Limitations</h3>
+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).
+<h3>License</h3>
+The model is released under <b>MIT</b> license