Description
nerdl_conll_deberta_base is a Named Entity Recognition (or NER) model, meaning it annotates text to find features like the names of people, places, and organizations. It was trained on the CoNLL 2003 text corpus. This NER model does not read words directly but instead reads word embeddings, which represent words as points such that more semantically similar words are closer together. nerdl_conll_deberta_base model is trained with the deberta_v3_base word embeddings, so be sure to use the same embeddings in the pipeline.
Predicted Entities
PER, LOC, ORG, MISC
Live Demo Open in Colab Download Copy S3 URI
How to use
document_assembler = DocumentAssembler() \
.setInputCol('text') \
.setOutputCol('document')
tokenizer = Tokenizer() \
.setInputCols(['document']) \
.setOutputCol('token')
embeddings = DeBertaEmbeddings.pretrained("deberta_v3_base", "en")\
.setInputCols(["token", "document"])\
.setOutputCol("embeddings")\
.setCaseSensitive(True)\
.setMaxSentenceLength(512)
ner_model = NerDLModel.pretrained('nerdl_conll_deberta_base', 'en') \
.setInputCols(['document', 'token', 'embeddings']) \
.setOutputCol('ner')
ner_converter = NerConverter() \
.setInputCols(['document', 'token', 'ner']) \
.setOutputCol('entities')
pipeline = Pipeline(stages=[
document_assembler,
tokenizer,
embeddings,
ner_model,
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 embeddings = DeBertaEmbeddings.pretrained("deberta_v3_base", "en")
.setInputCols("document", "token")
.setOutputCol("embeddings")
.setCaseSensitive(true)
.setMaxSentenceLength(512)
val ner_model = NerDLModel.pretrained("nerdl_conll_deberta_base", "en")
.setInputCols("document"', "token", "embeddings")
.setOutputCol("ner")
val ner_converter = NerConverter()
.setInputCols("document", "token", "ner")
.setOutputCol("entities")
val pipeline = new Pipeline().setStages(Array(document_assembler, tokenizer, embeddings, ner_model, ner_converter))
val example = Seq.empty["My name is John!"].toDS.toDF("text")
val result = pipeline.fit(example).transform(example)
import nlu
text = ["My name is John!"]
ner_df = nlu.load('en.ner.nerdl_conll_deberta_base').predict(text, output_level='token')
Model Information
| Model Name: | nerdl_conll_deberta_base |
| Type: | ner |
| Compatibility: | Spark NLP 4.0.0+ |
| License: | Open Source |
| Edition: | Official |
| Input Labels: | [sentence, token, embeddings] |
| Output Labels: | [ner] |
| Language: | en |
| Size: | 16.2 MB |
References
https://www.clips.uantwerpen.be/conll2003/ner/
Benchmarking
Test:
label precision recall f1-score support
B-LOC 0.92 0.93 0.93 1668
I-ORG 0.87 0.94 0.90 835
I-MISC 0.62 0.71 0.66 216
I-LOC 0.82 0.91 0.87 257
I-PER 0.99 0.99 0.99 1156
B-MISC 0.83 0.81 0.82 702
B-ORG 0.89 0.93 0.91 1661
B-PER 0.97 0.97 0.97 1617
micro-avg 0.91 0.93 0.92 8112
macro-avg 0.86 0.90 0.88 8112
weighted-avg 0.91 0.93 0.92 8112
processed 46435 tokens with 5648 phrases; found: 5719 phrases; correct: 5194.
accuracy: 93.13%; (non-O)
accuracy: 98.18%; precision: 90.82%; recall: 91.96%; FB1: 91.39
LOC: precision: 92.05%; recall: 92.99%; FB1: 92.51 1685
MISC: precision: 80.73%; recall: 78.77%; FB1: 79.74 685
ORG: precision: 87.97%; recall: 91.57%; FB1: 89.73 1729
PER: precision: 96.85%; recall: 97.03%; FB1: 96.94 1620
Dev:
label precision recall f1-score support
B-LOC 0.96 0.96 0.96 1837
I-ORG 0.88 0.94 0.91 751
I-MISC 0.90 0.76 0.83 346
I-LOC 0.91 0.92 0.92 257
I-PER 0.99 0.98 0.98 1307
B-MISC 0.93 0.87 0.90 922
B-ORG 0.90 0.95 0.92 1341
B-PER 0.97 0.99 0.98 1842
micro-avg 0.94 0.95 0.94 8603
macro-avg 0.93 0.92 0.92 8603
weighted-avg 0.94 0.95 0.94 8603
processed 51362 tokens with 5942 phrases; found: 5985 phrases; correct: 5622.
accuracy: 94.65%; (non-O)
accuracy: 98.89%; precision: 93.93%; recall: 94.61%; FB1: 94.27
LOC: precision: 95.88%; recall: 96.24%; FB1: 96.06 1844
MISC: precision: 91.65%; recall: 85.68%; FB1: 88.57 862
ORG: precision: 88.56%; recall: 93.51%; FB1: 90.97 1416
PER: precision: 97.16%; recall: 98.26%; FB1: 97.71 1863