Inspiration
I often read news articles to catch up on the latest events around the world. However, I am not sure if these articles contain any subtle biases that would influence my thoughts or outlook on these events. Therefore, I wanted to build a tool that could help readers identify hidden influences in news reporting.
What it does
News Bias Detector is a application designed to identify subtle emotional manipulation, corporate spin, and subjective framing in text. This tool serves as a critical bridge for media literacy, helping users peel back the layers of linguistic framing to reveal the objective core of a message. Users can paste articles and receives bias analysis that highlights specific phrases that might be biased.
How we built it
We used a flask backend with two analysis system: and AI analysis using OpenAI's GPT-4o-mini model and an API analysis using Textblob. In addition, direct quotes are excluded from the system for a more reliable detection. The frontend is contained of HTML, CSS, and Javascript and it features a spilt-panel interface.
Challenges we ran into
The biggest challenge we faced was calibrating the AI's sensitivity. For example, if the news article didn't go in much depth about a subject is too in detail for the general public, or if it didn't reflect enough, the AI would flag that as biased and give a higher score.
Accomplishments that we're proud of
Our AI system can highlight specific phrases, explaining why they're problematic and therefore intentionally/unintentionally biased in nature. This tool can educate users about media literacy concepts through practical examples.
What we learned
Bias exists on multiple levels from subtleness to overt manipulation; even neutral text still have their biases. In addition, we learned to balance the AI insights for a more accurate score.
What's next for News Bias Detector
Identifying bias is critically important for media literacy and since one detector may be overly sensitive, or missing some points, one point of improvement is to fact-check and cross-reference with other databases to ensure that the bias detector gives a more accurate score.
Log in or sign up for Devpost to join the conversation.