Edit on GitHub
  • John Snow Labs
  • Sep 20, 2022

English BertForSequenceClassification Cased model (from unitary)

  • bert
  • sequence_classification
  • classification
  • open_source
  • en

Description

Pretrained BertForSequenceClassification model, adapted from Hugging Face and curated to provide scalability and production-readiness using Spark NLP. toxic-bert is a English model originally trained by unitary.

Predicted Entities

obscene, insult, severe_toxic, identity_hate, threat, toxic

Download Copy S3 URI

How to use

documentAssembler = DocumentAssembler() \
        .setInputCol("text") \
        .setOutputCol("document")

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

sequenceClassifier_loaded = BertForSequenceClassification.pretrained("bert_classifier_toxic","en") \
    .setInputCols(["document", "token"]) \
    .setOutputCol("class")

pipeline = Pipeline(stages=[documentAssembler, tokenizer,sequenceClassifier_loaded])

data = spark.createDataFrame([["PUT YOUR STRING HERE"]]).toDF("text")

result = pipeline.fit(data).transform(data)

val documentAssembler = new DocumentAssembler() 
          .setInputCol("text") 
          .setOutputCol("document")

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

val sequenceClassifier_loaded = BertForSequenceClassification.pretrained("bert_classifier_toxic","en") 
    .setInputCols(Array("document", "token")) 
    .setOutputCol("class")

val pipeline = new Pipeline().setStages(Array(documentAssembler, tokenizer,sequenceClassifier_loaded))

val data = Seq("PUT YOUR STRING HERE").toDF("text")

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

import nlu
nlu.load("en.classify.bert.by_unitary").predict("""PUT YOUR STRING HERE""")

Model Information

Model Name: bert_classifier_toxic
Compatibility: Spark NLP 4.2.0+
License: Open Source
Edition: Official
Input Labels: [document, token]
Output Labels: [class]
Language: en
Size: 412.7 MB
Case sensitive: true
Max sentence length: 256

References

  • https://huggingface.co/unitary/toxic-bert
  • https://github.com/unitaryai/detoxify/issues/15
  • https://github.com/unitaryai/detoxify
  • https://laurahanu.github.io/
  • https://www.unitary.ai/
  • https://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge
  • https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification
  • https://www.kaggle.com/c/jigsaw-multilingual-toxic-comment-classification
  • https://homes.cs.washington.edu/~msap/pdfs/sap2019risk.pdf
  • https://arxiv.org/pdf/1703.04009.pdf%201.pdf
  • https://arxiv.org/pdf/1905.12516.pdf
  • https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/data
  • https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/data
  • https://www.kaggle.com/miklgr500/jigsaw-train-multilingual-coments-google-api
  • https://www.kaggle.com/c/jigsaw-unintended-bias-in-toxicity-classification/overview/evaluation