We were inspired by the amount of subtle (or not so subtle) biases in news articles and wanted to build a website that reflects a more objective perspective. Social media giants (including Google, Facebook and Twitter) use algorithms that are ever-changing and top secret to show us content and keep us hooked. Almost all platforms now constantly feed us content that aligns with our own interests, friends and belief systems. This reinforcement of our beliefs without any contrasting viewpoints can lead people down into extremism, as it decreases empathy and stifles conversation on controversial topics that need progress.
Our project analyses articles using AI to tell you their biases and rewrites them using less inflammatory vocabulary to make you consider why you have such an emotional reaction to certain content. It classifies articles based on their biases and creates a less inflammatory version of the article by finding most emotionally charged words in the article and replacing them with synonyms using wordnet. Additionally, it highlights the most biased sentences from the article.
We built it by merging python and html through utilizing flask. We used co:here for natural language processing. We fine tuned a model on co:here with a dataset of political statements and their sentiment and used it to classify the articles' bias and find the most emotionally charged words to replace. We deployed our app on Heroku.
Some challenges we went through were implementing co:here as it was our first time using it, troubleshooting flask and heroku, and matching the inflammatory key words to replace them with neutral-sounding ones.
Several accomplishments that we are proud of are being able to use flask and jinja to connect html and python, using machine learning to accurately portray the bias, and building a easy to use and visually appeasing UX design in html and css.
We learned a lot more about deploying on Heroku and using co:here. We also refined our Jinja skills.
What's next for Bias Begone! is to improve our replacement of inflammatory words by ensuring no grammatical errors are caused by the replacement. We also hope to find articles multiple news outlets with different political orientations based on a headline search so that users can read articles from all across the aisle.
We would like to be considered for Best Use of NLP with Cohere and Best Hack for Social Good by BlackRock
Log in or sign up for Devpost to join the conversation.