BERT Sequence Classification Base - IMDB (bert_base_sequence_classifier_imdb)

Description

BERT Model with sequence classification/regression head on top (a linear layer on top of the pooled output) e.g. for multi-class document classification tasks.

bert_base_sequence_classifier_imdb is a fine-tuned BERT model that is ready to be used for Sequence Classification tasks such as sentiment analysis or multi-class text classification and it achieves state-of-the-art performance.

We used TFBertForSequenceClassification to train this model and used BertForSequenceClassification annotator in Spark NLP 🚀 for prediction at scale!

Predicted Entities

neg, pos

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')

sequenceClassifier = BertForSequenceClassification \
.pretrained('bert_base_sequence_classifier_imdb', 'en') \
.setInputCols(['token', 'document']) \
.setOutputCol('class') \
.setCaseSensitive(True) \
.setMaxSentenceLength(512)

pipeline = Pipeline(stages=[
document_assembler,
tokenizer,
sequenceClassifier
])

example = spark.createDataFrame([['I really liked that movie!']]).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 = BertForSequenceClassification.pretrained("bert_base_sequence_classifier_imdb", "en")
.setInputCols("document", "token")
.setOutputCol("class")
.setCaseSensitive(true)
.setMaxSentenceLength(512)

val pipeline = new Pipeline().setStages(Array(document_assembler, tokenizer, sequenceClassifier))

val example = Seq("I really liked that movie!").toDS.toDF("text")

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

import nlu
nlu.load("en.classify.bert_sequence.imdb").predict("""I really liked that movie!""")

Results

* +--------------------+
* |result              |
* +--------------------+
* |[neg, neg]          |
* |[pos, pos, pos, pos]|
* +--------------------+

Model Information

Data Source

https://huggingface.co/datasets/imdb

Benchmarking

precision    recall  f1-score   support

neg       0.92      0.94      0.93     12304
pos       0.94      0.93      0.93     12696

accuracy                           0.93     25000
macro avg       0.93      0.93      0.93     25000
weighted avg       0.93      0.93      0.93     25000