BERT Token Classification - NER OntoNotes (bert_base_token_classifier_ontonote)

Description

BERT Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks.

bert_base_token_classifier_ontonote is a fine-tuned BERT model that is ready to use for Named Entity Recognition and achieves state-of-the-art performance for the NER task. This model has been trained to recognize four types of entities: CARDINAL, DATE, EVENT, FAC, GPE, LANGUAGE, LAW, LOC, MONEY, NORP, ORDINAL, ORG, PERCENT, PERSON, PRODUCT, QUANTITY, TIME, and WORK_OF_ART.

We used TFBertForTokenClassification to train this model and used BertForTokenClassification annotator in Spark NLP 🚀 for prediction at scale!

Predicted Entities

CARDINAL, DATE, EVENT, FAC, GPE, LANGUAGE, LAW, LOC, MONEY, NORP, ORDINAL, ORG, PERCENT, PERSON, PRODUCT, QUANTITY, TIME, WORK_OF_ART

Download

How to use

document_assembler = DocumentAssembler() \
    .setInputCol('text') \
    .setOutputCol('document')

tokenizer = Tokenizer() \
    .setInputCols(['document']) \
    .setOutputCol('token')

tokenClassifier = BertForTokenClassification \
      .pretrained('bert_base_token_classifier_ontonote', 'en') \
      .setInputCols(['token', 'document']) \
      .setOutputCol('ner') \
      .setCaseSensitive(True) \
      .setMaxSentenceLength(512)

# since output column is IOB/IOB2 style, NerConverter can extract entities
ner_converter = NerConverter() \
    .setInputCols(['document', 'token', 'ner']) \
    .setOutputCol('entities')

pipeline = Pipeline(stages=[
    document_assembler, 
    tokenizer,
    tokenClassifier,
    ner_converter
])

example = spark.createDataFrame([['My name is John!']]).toDF("text")
result = pipeline.fit(example).transform(example)
val document_assembler = DocumentAssembler() 
    .setInputCol("text") 
    .setOutputCol("document")

val tokenizer = Tokenizer() 
    .setInputCols("document") 
    .setOutputCol("token")

val tokenClassifier = BertForTokenClassification.pretrained("bert_base_token_classifier_ontonote", "en")
      .setInputCols("document", "token")
      .setOutputCol("ner")
      .setCaseSensitive(true)
      .setMaxSentenceLength(512)

// since output column is IOB/IOB2 style, NerConverter can extract entities
val ner_converter = NerConverter() 
    .setInputCols("document", "token", "ner") 
    .setOutputCol("entities")

val pipeline = new Pipeline().setStages(Array(document_assembler, tokenizer, tokenClassifier, ner_converter))

val example = Seq.empty["My name is John!"].toDS.toDF("text")

val result = pipeline.fit(example).transform(example)

Results

+------------------------------------------------------------------------------------+
 |result                                                                              |
 +------------------------------------------------------------------------------------+
 |[B-PERSON, I-PERSON, O, O, O, B-LOC, O, O, O, B-LOC, O, O, O, O, B-PERSON, O, O, O, O, B-LOC]|
 +------------------------------------------------------------------------------------+

Model Information

Model Name: bert_base_token_classifier_ontonote
Compatibility: Spark NLP 3.2.0+
License: Open Source
Edition: Official
Input Labels: [token, document]
Output Labels: [ner]
Language: en
Case sensitive: true
Max sentense length: 512

Data Source

https://catalog.ldc.upenn.edu/LDC2013T19

Benchmarking

Test:

         precision    recall  f1-score   support

   B-CARDINAL       0.86      0.88      0.87       935
       B-DATE       0.88      0.90      0.89      1602
      B-EVENT       0.71      0.62      0.66        63
        B-FAC       0.75      0.76      0.76       135
        B-GPE       0.97      0.91      0.94      2240
   B-LANGUAGE       0.79      0.68      0.73        22
        B-LAW       0.76      0.65      0.70        40
        B-LOC       0.78      0.83      0.80       179
      B-MONEY       0.88      0.90      0.89       314
       B-NORP       0.92      0.96      0.94       841
    B-ORDINAL       0.81      0.93      0.87       195
        B-ORG       0.87      0.89      0.88      1795
    B-PERCENT       0.92      0.95      0.93       349
     B-PERSON       0.96      0.95      0.95      1988
    B-PRODUCT       0.75      0.78      0.76        76
   B-QUANTITY       0.81      0.82      0.82       105
       B-TIME       0.69      0.70      0.69       212
B-WORK_OF_ART       0.66      0.74      0.70       166
   I-CARDINAL       0.83      0.88      0.86       331
       I-DATE       0.89      0.92      0.90      2011
      I-EVENT       0.70      0.71      0.70       130
        I-FAC       0.82      0.85      0.84       213
        I-GPE       0.96      0.89      0.93       628
        I-LAW       0.85      0.65      0.74       106
        I-LOC       0.85      0.84      0.84       180
      I-MONEY       0.94      0.96      0.95       685
       I-NORP       0.99      0.79      0.88       160
    I-ORDINAL       0.00      0.00      0.00         4
        I-ORG       0.91      0.93      0.92      2406
    I-PERCENT       0.95      0.95      0.95       523
     I-PERSON       0.96      0.96      0.96      1412
    I-PRODUCT       0.81      0.80      0.80        69
   I-QUANTITY       0.83      0.92      0.87       206
       I-TIME       0.71      0.77      0.74       255
I-WORK_OF_ART       0.71      0.66      0.68       337
            O       0.99      0.99      0.99    131815

     accuracy                           0.98    152728
    macro avg       0.82      0.81      0.82    152728
 weighted avg       0.98      0.98      0.98    152728



processed 152728 tokens with 11257 phrases; found: 11537 phrases; correct: 9906.
accuracy:  90.22%; (non-O)
accuracy:  98.00%; precision:  85.86%; recall:  88.00%; FB1:  86.92
         CARDINAL: precision:  83.35%; recall:  86.20%; FB1:  84.75  967
             DATE: precision:  82.23%; recall:  86.95%; FB1:  84.53  1694
            EVENT: precision:  58.06%; recall:  57.14%; FB1:  57.60  62
              FAC: precision:  68.67%; recall:  76.30%; FB1:  72.28  150
              GPE: precision:  95.59%; recall:  90.00%; FB1:  92.71  2109
         LANGUAGE: precision:  78.95%; recall:  68.18%; FB1:  73.17  19
              LAW: precision:  63.16%; recall:  60.00%; FB1:  61.54  38
              LOC: precision:  71.29%; recall:  80.45%; FB1:  75.59  202
            MONEY: precision:  85.40%; recall:  87.58%; FB1:  86.48  322
             NORP: precision:  89.82%; recall:  93.34%; FB1:  91.55  874
          ORDINAL: precision:  81.17%; recall:  92.82%; FB1:  86.60  223
              ORG: precision:  82.80%; recall:  86.35%; FB1:  84.54  1872
          PERCENT: precision:  86.23%; recall:  89.68%; FB1:  87.92  363
           PERSON: precision:  93.93%; recall:  94.22%; FB1:  94.07  1994
          PRODUCT: precision:  70.37%; recall:  75.00%; FB1:  72.61  81
         QUANTITY: precision:  73.28%; recall:  80.95%; FB1:  76.92  116
             TIME: precision:  58.02%; recall:  66.51%; FB1:  61.98  243
      WORK_OF_ART: precision:  52.40%; recall:  65.66%; FB1:  58.29  208