Project · March 11, 2020
Stock-market signals in Twitter data
A team research project testing whether company and CEO tweets could help predict short-term stock direction.
This project asked whether posts from a company and its CEO carried enough signal to help predict whether the stock would move up or down.
Our team collected tweets, removed duplicate and noisy text, applied latent semantic analysis, and compared classifiers including Naive Bayes with cross-validation. The project was an early lesson in the gap between an interesting correlation and a useful prediction—and in how much of data science is really careful data preparation.