text-summarizer

Desccription

An AI application built over Google's BERT model that can produce an extractive summary of a text. This application used a custom build tensorflow model built upon the BERT to obtain to produce predicitons on the relative importance of each sentence. These predictions are then used to produce a summary.

Table of Contents

• Setup
• Functionality
• Technology
• AI in Action
• Sources

Setup

Follow these steps to setup the application on your system -

Docker

• 1. Install Docker on your PC (https://docs.docker.com/engine/install/)
• 2. Host your choice of TensorFlow model using Docker (the preprocessing is done for the model - ExtractiveSummarizer_v1_40_words)
  • Create a variable in your terminal - in windows powershell - $ Set-Variable -Name MODEL_PATH -Value "path/to/model"
  • Pull tensorflow/serving image - $ docker pull tensorflow/tensorflow
  • Run the tensorflow/serving image - $ docker run -t --rm -d -p 8501:8501 -v "$MODEL_PATH/:/models/ExtractiveSummarizer_v1_40_words" -e MODEL_NAME=ExtractiveSummarizer_v1_40_words tensorflow/serving
  • For any further reference to Tensorflow Serving, check out the following link to Docker documentation (https://www.tensorflow.org/install/docker)

Python

Have Python installed on your system (if not installed, use this https://www.python.org/downloads/) Install the following packages globally in python

• json
• os
• numpy
• nltk
• tensorflow

Nodejs

• 1. Have nodejs installed in your system
• 2. Run $ node index.js
• 3. The application will be hosted on localhost:5000

Functionality

• The Tensorflow/serving image serves the tensorflow model (whose building will be posted in another repository) at the REST port (8501)
• The Nodejs Express application build a server which takes in the article to be summarized at '/summarize.html'
• On hitting the 'Summarize' button, the python scripts in the backend run the preprocessing steps to convert the input into a format that the A.I model can understand. This preprocessed string is then saved.
• The Express application then uses this preprocessed string to send a post request to the A.I model hosted at port 8501 and receives predictions.
• The Express application then uses these predictions to produce an extractive summarization of the given input.
• This summary is then posted onto '/summary.html' page.

Technology

• JavaScript
  • Nodejs
  • Express
  • Child process
  • Ejs
  • Server side rendering
• Python
  • numpy
  • nltk
  • TensorFlow
  • bert uncased model
  • custom A.I model building
• Docker
  • TensorFlow/serving
  • Keras A.I model hosting

AI in Action

• The Guardian article given in the Sources section has been put into the system.
• Homepage
• Summary Page
• Docker Tensorflow Model Hosting

Sources

• 1. Guardian Article considered in AI in Action section (https://www.theguardian.com/world/2021/jul/18/uk-covid-cases-could-hit-200000-a-day-says-neil-ferguson-scientist-behind-lockdown-strategy-england)
• 2. Google BERT (https://blog.google/products/search/search-language-understanding-bert/)
• 3. Repositories performing extractive summarization (https://github.com/dmmiller612/bert-extractive-summarizer)
• 4. Docker installation documentation (https://docs.docker.com/engine/install/)
• 5. Tensorflow Serving documentation (https://www.tensorflow.org/install/docker)
• 6. Python Installation (https://www.python.org/downloads/)
• 7. Tensorflow Installation (https://www.tensorflow.org/install)
• 8. TFHUB link for BERT uncased (https://tfhub.dev/google/bert_uncased_L-12_H-768_A-12/1)