Lab
Multimodal Fake News Detection
A BERT + ResNet-50 model that reads a post's text and image together to flag fake news.
- Source codecoming soon
- accuracy / F1
- 91.7%
- fewer errors than text-only
- 20%
Problem
Detecting fake news from a post’s text and its image together, instead of from the text alone.
What I built
A multimodal classifier that combines BERT for the text with ResNet-50 for the image.
Results
- 91.7% accuracy / F1
- 20% error reduction compared with a text-only baseline
Stack
BERT, ResNet-50.