Trusted answers to developer questions
Trusted Answers to Developer Questions

Related Tags


What is BERT?

Dian Us Suqlain

As we all know, technological advancements and new trends are introduced daily. In 2018, Google published a paper on BERTBidirectional Encoder Representations from Transformers that totally changed Machine Learning practices and bridged a major gap.

Every word in BERT has its own importance and plays a vital role in Natural Language Processing.

Traditional models vs. BERT

Unlike other models that have previously worked unidirectionallyeither left-to-right or vice versa, BERT reads the entire sequence of sentences bidirectionallyin both directions. This is the reason for its effective and accurate results.

BERT understands the context in which the sentence is spoken

Take a look at the example above, previously, trained models couldn’t actually figure out the main context of the sentence. So, a model would have struggled with the word bank since it has two meaningsthe river bank and the bank where we handle financial matters.

BERT, on the other hand, checks both sides of the highlighted word and then generates its results accordingly. This is where the concept of transformers plays a major role.

Usage of BERT

There are multiple usages for Bert like:

  • Hate speech analysis
  • Text classification (Sentiment Analysis)
  • Sentence Prediction
  • Model training etc.

Most developers use BERT as it is pretrained on a significantly large corpus of unlabelled text including the entire Wikipedia - which alone includes 2,500 million words, and Book Corpus of 800 million words).


There are currently two variants of BERT that are built on top of a transformer:

  1. 12 layers - BERT base
  2. 24 layers - BERT large
12 layers and 24 layers transformer blocks (BERT)

Sentiment analysis using BERT

To classify any statement, whether politically or in positive or negative remarks, every model has to understand the nature of the words spoken in it.

The BERT framework uses pre-training and fine-tuning to create tasks that include question answering systems, sentiment analysis, and language inference.

Fine-tuning and sentiment analysis

Sentence prediction with Bert

The primary objective of any NLP technique is to study human language and learn how it is spoken.

The simplest example for actual understanding of BERT can be Google searchWhenever you type something on the Google search bar, it automatically predicts the next sentence and shows a few suggestions in the dropdown.

Google search and Bert prediction
Google search and Bert prediction

For detailed understanding

You can learn more about BERT and its implementation from the official Google blog.




Dian Us Suqlain
Copyright ©2022 Educative, Inc. All rights reserved

View all Courses

Keep Exploring