Based Analysis (Facter)

Inspiration

The rampant spread of misinformation and its detrimental effect on our community's ability to have constructive discussions.

What it does

It takes a URL or text and applies a bias and tone detection algorithm to the individual paragraphs of an article. All paragraphs flagged as biased are then correlated with articles written by different news outlets on the same event, we provide the links and similar paragraphs as an easy way to fact-check biased material.

How we built it

We used a svelte front end deployed on Vercel with the tailwind and shadcn libraries as UI support. For our bias detection, sentiment analysis, and cross-referencing algorithms we used a fast API backend run on a Google Cloud server, with newspaper3k to extract articles from a URL, hugging face models to flag misinformation, and Beautiful Soup to find related articles.

Challenges we ran into

Our models were too large for the server we were running them on, so we reduced the amount of cross-referencing to only paragraphs flagged as biased.

Accomplishments that we're proud of

It works and it can flag misinformation very accurately while maintaining the simplicity of a single URL for our user!

What we learned

You have to optimize your algorithms for practicality sake, even if you lose some features and accuracy

What's next for Based Analysis/Facter

We need to improve the front end and add more fact-checking features, like checking other articles from the news source for bias and hopefully, with a better server we can reduce the restrictions on the cross-referencing algorithms.

Built With

• ai
• beautiful-soup
• cloud
• clustering
• fastapi
• google-cloud
• huggingface
• machine-learning
• python
• scipy
• spacy
• svelte
• tailwind
• vercel
• web