The aim of this project is to run in production a deep learning model supposed to detect fake news. The model can be deployed in a simple API thanks to the Flask web-framework.

Fake news refers to misinformation, disinformation or mal-information which is spread through word of mouth and traditional media and more recently through digital forms of communication such as edited videos, memes, unverified advertisements and social media propagated rumours.Fake news spread through social media has become a serious problem, with the potential of it resulting in mob violence, suicides etc... as a result of misinformation circulated on social media.

Model

The deep learning model used was retrieved in a notebook found in kaggle : https://www.kaggle.com/madz2000/nlp-using-glove-embeddings-99-87-accuracy. This model is using word embedding thanks to a pretrained Glove model. As NLP model used in dataset containing several sentences, this model is a RNN using LSTM layers which stands for Long short-term memory.

API

After being retrained, this model had to be runned in production in a Flask API well documented thanks to the interface description langage Swaggger.

Working environment

The Flask API was coded in Python langage with a virtual environment. To use this project, it is highly recommended to create a virtual environment and get the necessary libraries :

pip install -r requirements.txt

To serve this little API, a Gunicorn application server and a Nginx web server had to be set up. The application server is used to communicate with the application and the web server is used to treat HTTP requests in our API.

This servers and the Flask API are gathered in a Docker container. You can easily use this project cloning this repository and build this container (if you have Docker, Gunicorn and Nginx previously installed) :

git clone https://github.com/VALDOM-PROJET-TRANSVERSE-2020/Mullor-Cotte-project.git
./run_docker.sh