Inspiration
With voting season approaching, accurate information is crucial for making an informed decision. Therefore we found it important for people to be able to verify news.
What it does
This project is taking an input of politically related events' text and determining whether it is true or or fake news. This is done using a BERT pretrained model that was fine tuned over a fake and real news dataset from kaggle. Please note that due to time shortage our training data was trimmed from 45,000 values to 10,000 data values.
How we built it
Frontend done in React + Tailwind Backend Done in Python with FastAPI AI Model used was a pretrained BERT
Challenges we ran into
- Data was too large to train the full dataset in time
- Needed to fine tune model instead of being able to directly call model from API
- Connecting frontend to model
Accomplishments that we're proud of
- None of us have much experience at all with React but were able to make a decent frontend web app
What we learned
- How to train language models over data and how they tokenize data.
What's next for Fake News Detection
At the moment, the response is binary. In the future, we would like the model score the text and to provide an explanation annotating the text with what is real and what is fake.
Log in or sign up for Devpost to join the conversation.