Step by Step Intent Recognition With BERT

Nutan
14 min readOct 18, 2020

Use BERT for intent recognition

Contents

  1. What is BERT?

2. Model architecture

3. Data Preparation

  • Import modules
  • Declare data folder path
  • Read csv files
  • Concat train and valid dataframe
  • Get the unique intent and it’s count

4. Visualization

4.1 Plot intents

  • Countplot
  • Barplot

5. Load BERT model

  • Declare bert configuration files

6. Tokenize vocabulary

  • Load vocab.txt file
  • Tokenize vocab with FullTokenizer

7. Preprocess data

  • Create class
  • Add input parameters to the constructor
  • Sort train and test data by length of text
  • Preprocess train and test data
  • Add padding to the sequence
  • Full class code

8. Create custom model

  • Create a function
  • Read bert configuration file
  • Create bert model layer from bert configuration parameters
  • Create Keras model and add input and output layers in that

--

--

Nutan

knowledge of Machine Learning, React Native, React, Python, Java, SpringBoot, Django, Flask, Wordpress. Never stop learning because life never stops teaching.