Language Processing Network for Fake News Sites

• 

  Verification training accuracy

• 

  Verification training loss

• 

  BiGRU model diagram

• 

  Testing accuracy

Inspiration

I was inspired by the massive increase in deepfakes, spam articles, and fake news being spread through generative AI. I wanted to provide a tool to help combat it.

What it does

It uses a bi-directional GRU network to parse webpages and rank their trustworthiness.

How we built it

Using tensor flow and a dataset from Kaggle found here: https://www.kaggle.com/datasets/emineyetm/fake-news-detection-datasets?resource=download

Challenges we ran into

I have never worked with tensorflow and the documentation was rather confusing and hard to learn in such a short time period.

Accomplishments that we're proud of

I am very proud to have a 80% tested accuracy and 99% verification accuracy on my trained model.

What we learned

How to use tensor flow for natural language processing. How to tokenize and pad/concat text to format it for a neural network. Also how to change that data back into higher dimensions so we can have better comparison amongst tokens. I also learned alot about both BERT style language processing and biGRU style. Both have specific applications and I would love to explore how they work more!

What's next for Language Processing Network for Fake News Sites

A website that people can conveniently use to check their news sources on the fly.

Built With

• beautiful-soup
• pandas
• python
• tensorflow