Description
This model is trained with Zero-Shot Named Entity Recognition (NER) approach and it can detect any kind of defined entities with no training dataset, just pretrained RoBERTa embeddings (included in the model).
Predicted Entities
How to use
documentAssembler = DocumentAssembler() \
.setInputCol("text") \
.setOutputCol("document")
sentenceDetector = SentenceDetector() \
.setInputCols(["document"]) \
.setOutputCol("sentence")
tokenizer = Tokenizer() \
.setInputCols(["sentence"]) \
.setOutputCol("token")
zero_shot_ner = ZeroShotNerModel.pretrained("zero_shot_ner_roberta", "en", "clinical/models")\
.setInputCols(["sentence", "token"])\
.setOutputCol("zero_shot_ner")\
.setEntityDefinitions(
{
"NAME": ["What is his name?", "What is my name?", "What is her name?"],
"CITY": ["Which city?", "Which is the city?"]
})
ner_converter = NerConverter()\
.setInputCols(["sentence", "token", "zero_shot_ner"])\
.setOutputCol("ner_chunk")\
pipeline = Pipeline(stages = [
documentAssembler,
sentenceDetector,
tokenizer,
zero_shot_ner,
ner_converter])
zero_shot_ner_model = pipeline.fit(spark.createDataFrame([[""]]).toDF("text"))
data = spark.createDataFrame(["Hellen works in London, Paris and Berlin. My name is Clara, I live in New York and Hellen lives in Paris.",
"John is a man who works in London, London and London."], StringType()).toDF("text")
result = zero_shot_ner_model.transform(data)
val documentAssembler = new DocumentAssembler()
.setInputCol("text")
.setOutputCol("document")
val sentenceDetector = new SentenceDetector()
.setInputCols(Array("document"))
.setOutputCol("sentence")
val tokenizer = new Tokenizer()
.setInputCols(Array("sentence"))
.setOutputCol("token")
val zero_shot_ner = ZeroShotNerModel.pretrained("zero_shot_ner_roberta", "en", "clincial/models")
.setInputCols(Array("sentence", "token"))
.setOutputCol("zero_shot_ner")
.setEntityDefinitions(Map(
"NAME"-> Array("What is his name?", "What is my name?", "What is her name?"),
"CITY"-> Array("Which city?", "Which is the city?")
))
val ner_converter = new NerConverter()
.setInputCols(Array("sentence", "token", "zero_shot_ner"))
.setOutputCol("ner_chunk")
val pipeline = new .setStages(Array(
documentAssembler,
sentenceDetector,
tokenizer,
zero_shot_ner,
ner_converter))
val data = Seq(Array("Hellen works in London, Paris and Berlin. My name is Clara, I live in New York and Hellen lives in Paris.",
"John is a man who works in London, London and London.")toDS().toDF("text")
val result = pipeline.fit(data).transform(data)
import nlu
nlu.load("en.zero_shot.ner_roberta").predict("""Hellen works in London, Paris and Berlin. My name is Clara, I live in New York and Hellen lives in Paris.""")
Results
+------+---------+--------+-----+---+----------+
| token|ner_label|sentence|begin|end|confidence|
+------+---------+--------+-----+---+----------+
|Hellen| B-NAME| 0| 0| 5|0.13306311|
| works| O| 0| 7| 11| null|
| in| O| 0| 13| 14| null|
|London| B-CITY| 0| 16| 21| 0.4064213|
| ,| O| 0| 22| 22| null|
| Paris| B-CITY| 0| 24| 28|0.04597357|
| and| O| 0| 30| 32| null|
|Berlin| B-CITY| 0| 34| 39|0.16265489|
| .| O| 0| 40| 40| null|
| My| O| 1| 42| 43| null|
| name| O| 1| 45| 48| null|
| is| O| 1| 50| 51| null|
| Clara| B-NAME| 1| 53| 57| 0.9274031|
| ,| O| 1| 58| 58| null|
| I| O| 1| 60| 60| null|
| live| O| 1| 62| 65| null|
| in| O| 1| 67| 68| null|
| New| B-CITY| 1| 70| 72|0.82799006|
| York| I-CITY| 1| 74| 77|0.82799006|
| and| O| 1| 79| 81| null|
|Hellen| B-NAME| 1| 83| 88|0.40429682|
| lives| O| 1| 90| 94| null|
| in| O| 1| 96| 97| null|
| Paris| B-CITY| 1| 99|103|0.49216735|
| .| O| 1| 104|104| null|
| John| B-NAME| 0| 0| 3|0.14063153|
| is| O| 0| 5| 6| null|
| a| O| 0| 8| 8| null|
| man| O| 0| 10| 12| null|
| who| O| 0| 14| 16| null|
| works| O| 0| 18| 22| null|
| in| O| 0| 24| 25| null|
|London| B-CITY| 0| 27| 32|0.15521188|
| ,| O| 0| 33| 33| null|
|London| B-CITY| 0| 35| 40|0.12151082|
| and| O| 0| 42| 44| null|
|London| B-CITY| 0| 46| 51| 0.2650951|
| .| O| 0| 52| 52| null|
+------+---------+--------+-----+---+----------+
Model Information
| Model Name: | zero_shot_ner_roberta |
| Compatibility: | Healthcare NLP 4.0.2+ |
| License: | Licensed |
| Edition: | Official |
| Input Labels: | [document_question, document_context] |
| Output Labels: | [answer] |
| Language: | en |
| Size: | 460.2 MB |
| Case sensitive: | true |
| Max sentence length: | 512 |
References
As it is a Zero-Shot NER, no training dataset is necessary.