Description
XLM-RoBERTa 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.
xlm_roberta_base_token_classifier_ontonotes is a fine-tuned XLM-RoBERTa 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 TFXLMRobertaForTokenClassification to train this model and used RoBertaForTokenClassification 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
How to use
document_assembler = DocumentAssembler() \
.setInputCol('text') \
.setOutputCol('document')
tokenizer = Tokenizer() \
.setInputCols(['document']) \
.setOutputCol('token')
tokenClassifier = XlmRoBertaForTokenClassification \
.pretrained('xlm_roberta_base_token_classifier_ontonotes', '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 = XlmRoBertaForTokenClassification.pretrained("xlm_roberta_base_token_classifier_ontonotes", "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)
import nlu
nlu.load("en.ner.xlm_roberta.base").predict("""My name is John!""")
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: | xlm_roberta_base_token_classifier_ontonotes |
| Compatibility: | Spark NLP 3.3.0+ |
| 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
precision recall f1-score support
B-CARDINAL 0.86 0.86 0.86 935
B-DATE 0.88 0.90 0.89 1602
B-EVENT 0.64 0.51 0.57 63
B-FAC 0.73 0.75 0.74 135
B-GPE 0.97 0.96 0.96 2240
B-LANGUAGE 0.80 0.55 0.65 22
B-LAW 0.78 0.62 0.69 40
B-LOC 0.78 0.80 0.79 179
B-MONEY 0.87 0.91 0.89 314
B-NORP 0.94 0.96 0.95 841
B-ORDINAL 0.81 0.92 0.86 195
B-ORG 0.92 0.90 0.91 1795
B-PERCENT 0.93 0.95 0.94 349
B-PERSON 0.94 0.96 0.95 1988
B-PRODUCT 0.74 0.75 0.75 76
B-QUANTITY 0.79 0.82 0.80 105
B-TIME 0.72 0.66 0.68 212
B-WORK_OF_ART 0.68 0.72 0.70 166
I-CARDINAL 0.79 0.89 0.83 331
I-DATE 0.90 0.92 0.91 2011
I-EVENT 0.70 0.81 0.75 130
I-FAC 0.84 0.86 0.85 213
I-GPE 0.94 0.92 0.93 628
I-LAW 0.84 0.65 0.73 106
I-LOC 0.85 0.83 0.84 180
I-MONEY 0.94 0.94 0.94 685
I-NORP 0.95 0.92 0.94 160
I-ORDINAL 0.00 0.00 0.00 4
I-ORG 0.92 0.93 0.93 2406
I-PERCENT 0.95 0.96 0.96 523
I-PERSON 0.95 0.96 0.96 1412
I-PRODUCT 0.73 0.78 0.76 69
I-QUANTITY 0.83 0.93 0.88 206
I-TIME 0.70 0.77 0.74 255
I-WORK_OF_ART 0.70 0.68 0.69 337
O 0.99 0.99 0.99 131815
accuracy 0.98 152728
macro avg 0.81 0.81 0.81 152728
weighted avg 0.98 0.98 0.98 152728
processed 152728 tokens with 11257 phrases; found: 11498 phrases; correct: 10038.
accuracy: 90.96%; (non-O)
accuracy: 98.12%; precision: 87.30%; recall: 89.17%; FB1: 88.23
CARDINAL: precision: 82.65%; recall: 84.06%; FB1: 83.35 951
DATE: precision: 83.75%; recall: 87.52%; FB1: 85.59 1674
EVENT: precision: 56.90%; recall: 52.38%; FB1: 54.55 58
FAC: precision: 69.44%; recall: 74.07%; FB1: 71.68 144
GPE: precision: 95.46%; recall: 94.87%; FB1: 95.16 2226
LANGUAGE: precision: 80.00%; recall: 54.55%; FB1: 64.86 15
LAW: precision: 66.67%; recall: 60.00%; FB1: 63.16 36
LOC: precision: 70.35%; recall: 78.21%; FB1: 74.07 199
MONEY: precision: 81.63%; recall: 86.31%; FB1: 83.90 332
NORP: precision: 92.51%; recall: 95.48%; FB1: 93.97 868
ORDINAL: precision: 81.36%; recall: 91.79%; FB1: 86.27 220
ORG: precision: 87.76%; recall: 87.08%; FB1: 87.42 1781
PERCENT: precision: 87.68%; recall: 89.68%; FB1: 88.67 357
PERSON: precision: 93.19%; recall: 95.62%; FB1: 94.39 2040
PRODUCT: precision: 67.90%; recall: 72.37%; FB1: 70.06 81
QUANTITY: precision: 69.83%; recall: 77.14%; FB1: 73.30 116
TIME: precision: 62.79%; recall: 63.68%; FB1: 63.23 215
WORK_OF_ART: precision: 62.16%; recall: 69.28%; FB1: 65.53 185