Misinformation and fake news are becoming larger and larger issues for our society, with the stakes becoming greater as we rely on our sources to be trustworthy. But with so many new sources of information becoming available to us every day, how can we determine quickly how credible a given article is? Validity aims to reinforce certainty by analyzing a news article and referencing a neural network to determine its credibility. With Validity, we can further work towards a world that aims to be fair and truthful.

What it does

It all starts with the user. Validity takes a news article as input and calls on a Microsoft Azure Cloud Node API. This API works with TensorFlow, which utilizes the power of a neural network implemented on a dataset of over 37,000 news articles to determine its integrity, ultimately resulting in an 85% accuracy rate in training. This data is then presented back to the user in a website, while presenting other websites that might be worth looking at next.

How we built it

We built and trained the model using Tensorflow.js, built the website front-end with ReactJS, and built a Node.js + Express.js server to connect the two, which is hosted on Microsoft Azure Cloud. Our professional logo was built in Figma.

Challenges we ran into

Hosting the server on Azure with issues with Node versions, training the model on a huge amount of text data with a laptop (110 MB), building a ReactJS website in a little under 3 hours, and a power outage for one of our members.

Accomplishments that we're proud of

Getting a trained and prediction-ready model, hosting on Azure, and collaborating in a language that is none of our strongest languages.

What we learned

Azure services, Tensorflow character processing, ReactJS hosting.

What's next for Validity

Expanded capabilities for analysis (including source validation and metadata), browser extension. Currently, our application runs locally, but our Azure Node.js server is running and accessible, we just plan to put everything on Azure eventually.

Built With

Share this project: