bias-ml

A political bias detector, a full stack, machine learning web application.

Comment 2

Inspiration

In today's political climate being a citizen means being informed on the politics of our time. How do you invite people to a meaningful conversation concerning politics? Well, you start by giving them the facts...web articles, the news, tweets, etc... so they can form their own opinions, right? Oftentimes, these “facts” are swayed by political bias, which stands in the way of the truth and does not let civilian voices speak for themselves. If you want people to be engaged with civic affairs, then you give them the power to see how their information is biased toward a particular party. You give people a political bias detector. You give people BIAS ML because people deserve to form their own opinions.

What it does

Our application is a website and google firebase program that takes another web article's URL and parses its sentences for political bias.

How we built it

The frontend of our app was built using ReactJS. When a URL is submitted to the site, it sends that URL to a Google Firebase function that uses the Puppeteer Node JS module to extract the raw HTML content. The raw content is then parsed into sentences and sent to our Google Cloud AutoML model, which scores the text on conservative and liberal bias. This data is sent back to our site, and the results are displayed using the React-compatible version of ChartJS.

Challenges we ran into

Our biggest frustrations came from trying to learn the Google Firebase and Google Cloud APIs. We had never worked with either of these APIs so there was a lot to learn in just a few days. Training the NLP model also took much longer than we had anticipated. In addition, finding the right data to train the bot was difficult, and parsing the dataset for bad entries and duplicates was also a grueling process. Using Google Firebase HTTP functions caused issues since we had to find a workaround for CORS restrictions. Parsing raw HTML into strictly the article content also posed a significant challenge.

Accomplishments that we're proud of

We are proud of designing a sleek yet interactive interface. We are proud of building a firebase backend that works in tandem with our frontend We are proud of training our first neural network

What we learned

Group Programming Frontend Backend ReactJS ChartsJS Google-Firebase Google-Cloud-APIs Google-AutoML Node.js UX/UI Design

What's next for bias-ml

We plan to retrain our neural network on a more varied dataset that is better parsed and edited for the task at hand. We would like to deploy the website to a hosting service without error. In addition, we would like to create and use a sentiment detection model that can help decide whether the subjects in the sentence are viewed positively or negatively. We would also like to create a google chrome extension that can be downloaded from our webpage.

Built With

Share this project:

Updates