Can algorithms determine whether claims made about COVID-19 are real or fake?

This project was completed as part of the Columbia University 2021 Data Science Hackathon

• Competition lasted about 14 hours to complete both code and presentation portions of submission
• Worked as part of a great team, uploaded my specific portions of the project here

Used the COVID-19 misinformation dataset to create an algorithmic way of determining fake vs. real news claims about COVID-19

Background Resources

• Data: https://github.com/cuilimeng/CoAID
• Related Paper: https://arxiv.org/abs/2006.00885

Process

• Discovered 20 recurrent topics within the COVID-19 news claims using the CorEx Topic modeling algorithm
• Used the topics discovered within the claims, along with Twitter interaction of the claims (tweets and replies), as features in a predictive model to determine fake vs. true news claims regarding COVID-19
  • Applied SMOTE oversampling technique on the fake news claims to offset the imbalance between real vs. fake news claims in the dataset

Future steps

• Apply anchoring in topic modeling process to see if topic derivation can be improved
• Apply additional algorithmic approaches to claim validity determination model to improve model AUC
• Results at time of hackathon submission:
  • Accuracy: 93.59%
  • AUC: 0.6710
• Apply hyperparamter tuning through GridSeaarch to determine best overall model
• Include additional features in predictive model such as claim source or claim date