Introduction | Web Application | Mobile Application UI | Documentation
Text summarization is one of the central challenges in the fields of Machine Learning and Natural Language Processing (NLP). Bidirectional Encoder Representations from Transformers (BERT), a new contextual pre-training method for language representations, has been heralded as the state-of-the-art neural network architecture that can outperform any others in over 11 complex NLP tasks at the time of its creation. In this paper, we explore the potential of utilizing BERT as the basis for a document level encoder that can capture and generate a representation for text sentences and meanings, ultimately providing a reliable and accurate automated summarization process of news articles from different international outlets.
With the app, the users will be able to copy news articles or any piece of writings and generate summaries based on three different models: extractive (summary taken directly from text) , abstractive (summary generated by encoders) or a mixture of both.
To launch the app, please download or clone the project into your local machine, change directory into the src folder of the Web Application folder, and download the following three files into the models folder. You will also need to install the Flask framework.
BERT Extractive Model
BERT Abstractive Model
BERT Mixed Model
$ pip install flask # install Flask
$ cd src # change directory to src folder
$ python app.py # launch the app
The app will be launch on one of your local ports. Please copy and paste the url to your browser and test out the app.
Quick News with BERT is a proof of concept for how these automated summary models can be utilized in real life setting. The user will be able to choose news sites that are of their interest and see the latest updated articles and generate their summaries quickly and efficiently. The functionalities, at the moment, are hard-coded in, however, and will be updated in the future.
Make sure you have Flutter installed on your local machine. For more instructions on how to install flutter, please take a look here.
To test the app UI (available on both iOS and Android platforms), please run the following commands in your terminal, after installing Flutter:
$ git clone https://github.com/huydang90/Flutter-News-Summary-App.git
$ cd Flutter-News-Summary-App
$ flutter run
To access the documentation of the models, please download the Documentation folder. The launch page for the documentation can be found under the following folder headings: Documentation/docs/_build/html/index.html
