Description
nerdl_conll_elmo 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_elmo model is trained with elmo 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 = ElmoEmbeddings\
.pretrained('elmo', 'en')\
.setInputCols(["token", "document"])\
.setOutputCol("embeddings")\
.setPoolingLayer("elmo")
ner_model = NerDLModel.pretrained('nerdl_conll_elmo', '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 = ElmoEmbeddings.pretrained("elmo", "en")
.setInputCols("document", "token")
.setOutputCol("embeddings")
.setPoolingLayer("elmo")
val ner_model = NerDLModel.pretrained("nerdl_conll_elmo", "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_elmo').predict(text, output_level='token')
Model Information
| Model Name: | nerdl_conll_elmo |
| Type: | ner |
| Compatibility: | Spark NLP 4.0.0+ |
| License: | Open Source |
| Edition: | Official |
| Input Labels: | [sentence, token, embeddings] |
| Output Labels: | [ner] |
| Language: | en |
| Size: | 17.0 MB |
References
https://www.clips.uantwerpen.be/conll2003/ner/
Benchmarking
Test:
label precision recall f1-score support
B-LOC 0.941 0.932 0.936 1668
I-ORG 0.886 0.932 0.908 835
I-MISC 0.709 0.745 0.727 216
I-LOC 0.863 0.879 0.871 257
I-PER 0.974 0.993 0.983 1156
B-MISC 0.843 0.825 0.834 702
B-ORG 0.907 0.931 0.919 1661
B-PER 0.961 0.972 0.966 1617
micro-avg 0.920 0.932 0.926 8112
macro-avg 0.885 0.901 0.893 8112
weighted-avg 0.920 0.932 0.926 8112
processed 46435 tokens with 5648 phrases; found: 5685 phrases; correct: 5209.
accuracy: 93.24%; (non-O)
accuracy: 98.30%; precision: 91.63%; recall: 92.23%; FB1: 91.93
LOC: precision: 93.58%; recall: 92.69%; FB1: 93.13 1652
MISC: precision: 82.03%; recall: 80.63%; FB1: 81.32 690
ORG: precision: 89.46%; recall: 91.93%; FB1: 90.68 1707
PER: precision: 95.97%; recall: 97.09%; FB1: 96.53 1636
Dev:
label precision recall f1-score support
B-LOC 0.974 0.968 0.971 1837
I-ORG 0.923 0.948 0.936 751
I-MISC 0.933 0.841 0.884 346
I-LOC 0.948 0.930 0.939 257
I-PER 0.981 0.984 0.982 1307
B-MISC 0.939 0.903 0.921 922
B-ORG 0.923 0.950 0.936 1341
B-PER 0.972 0.981 0.976 1842
micro-avg 0.956 0.955 0.956 8603
macro-avg 0.949 0.938 0.943 8603
weighted-avg 0.956 0.955 0.956 8603
processed 51362 tokens with 5942 phrases; found: 5961 phrases; correct: 5651.
accuracy: 95.55%; (non-O)
accuracy: 99.07%; precision: 94.80%; recall: 95.10%; FB1: 94.95
LOC: precision: 96.99%; recall: 96.46%; FB1: 96.72 1827
MISC: precision: 92.13%; recall: 88.94%; FB1: 90.51 890
ORG: precision: 91.05%; recall: 94.03%; FB1: 92.52 1385
PER: precision: 96.72%; recall: 97.61%; FB1: 97.16 1859