News Balancer

Inspiration

Today, America is more misinformed and divided than ever. With the increasing prevalence of pseudo-news websites such as BuzzFeed, incorrect news sources and politically biased articles stand in the way of a clear understanding of contemporary issues. Furthermore, the absence of differing viewpoints in today's media has led to echochambers in which peoples prejudices and beliefs are repeatedly confirmed. This leads to a lack of understanding between different groups that can breed discrimination and intolerance.

What it does

News Balancer takes any story or article and provides numerous credible articles on that story sorted by their political bias. The homepage will randomly generate a story from its archives, and display different articles along with their bias and credibility so the user can easily scan multiple viewpoints at once. Additionally, a user can type in a query (and a time range) to get stories relating to any query they might have along with their credibility / political bias.

How I built it

News Balancer uses scikit learn to find similar articles based on a machine learning algorithm known as Multinomial Naive Bayes Classification. It was trained on a recently released dataset (News Data Aggregator, UCI 2016).
The biases and credibility of publications within this dataset were evaluated using well-known websites (ex. allsides.com).
A Django application was built to visualize the biases in the studied articles, and also to extend it to new queries.

Personal accomplishments

Our group struggled both with inexperience in using machine learning packages, understanding a new dataset, and especially with development using Django. It is a miracle to us that we were able to piece this together in the amount of time we did.

What's next for News Balancer

News Balancer would best work as an extension for news sites that people already go to. Thus News Balancer should take the form of a Chrome extension that could in real-time, check the credibility and bias of sources on news websites. This is especially important for sites such as Facebook that don't always fact-check sources.