The inspiration behind this hackathon project was mainly these two articles:

What it does

It takes a text corpus of news articles and detects using machine learning whether each article in the corpus is real or fake.

How we built it

We used BERT algorithm developed by Google to build over neural network model which would classify a given news article. We achieved a state if art accuracy of 98% in prediction along with a loss of 0.002.

Challenges we ran into

1) The main challenge we ran into was creating a valid data set. 2) Apart from creating a dataset, we also needed a data set large enough for the algorithm to classify any news article in general. 3) There were restrictions on website scrapping and API calls 4) It was first time we were using Google cloud platform to store our data and also using Google collab.

Accomplishments that we're proud of

1) We are most proud of the fine tuning we achieved on our algorithm. 2) Also, we are probably the first ones to use BERT to fight the cause of Fake News 3) The predicted model runs in under 10 mins when given a corpus of over 50000 news articles.

What we learned

We learned web scrapping, API call handling, model fine tuning and UI communication.

What's next for FAKE NEWS WEEDER

In the future we wish to provide real time analytics on the news article. So that we can detect any fake news as soon as it is published.

Share this project: