Biascope | Devpost

GIF
Browser Extension Demo
Mage Text Classification Models

Inspiration

Fake news and the spread of biased misinformation is one of the most prevalent social problems in American and global society. Pew Research Center found 68% of Americans say fake news undermines their confidence in government institutions and 56% believe the problem will only get worse over time. Pew also found that heavily biased news is accelerating polarization as the more clickbait or eye-catching a headline is, the more views it attracts. To make matters worse, according to a Stanford study, Americans fail to identify fake news over real news with a “stunning and dismaying consistency.” This can create confusion and misunderstanding. The more misinformed people are, the less society is able to solve important social and political issues. Misinformation can also be weaponized, as seen with current Russian propaganda amid their invasion of Ukraine.

What it does

Biascope empowers users to be active consumers of news. We help people break free from the harmful echo chambers of biased news by intuitively displaying the bias and accuracy of news articles without extra effort from the consumer. By color-highlighting bias in articles and displaying metrics on news sites, our extension trains users to notice biased language and how it influences their interpretation of content. Promoting thought around news rather than accepting what’s given at face value combats the spread of misinformation.

How we built it

We built a FastAPI + SQLAlchemy Python backend with a Postgres database, and a Firefox/Chrome browser extension. Originally, we wanted a website so we used this fastapi-react template: https://github.com/Buuntu/fastapi-react, but we realized after some brainstorming that we really only need a browser extension to have an effective product. We used Poetry for python dependency management and Docker (images came with the react-fastapi template) to handle running the frontend-backend stack.

Challenges we ran into

Training the AI model took far longer than we anticipated, though Mage helped with this process.

Accomplishments that we're proud of

We built a bias-detecting transformer model with 73% accuracy! And built a neat browser extension to display the data in an intuitive way :)

What we learned

Training a model takes a lot of time, and gathering data for it takes longer!

What's next for Biascope

We want to continue working on this and increase the accuracy of our model. We also want to improve the design and ease of use of the extension. Ideally, it will be on the Chrome store some day soon!