Project: Twitter sentiment and stock market performance
Duration: Spring 2012 – Summer 2012
Funding: None
ABSTRACT
This is a project that has continued from Marc’s excellent work in data mining. We are investigating an algorithm we developed that clusters similar Tweets in real-time and generates sentiment related to a specific company of interest from incoming Tweets. Using this information and plotting it against the performance of the company’s stock price, we are gauging the potential use of Twitter as a descriptor and predictor of the stock market.
ACHIEVEMENTS
Marc started this project toward the end of my data mining class. There was insufficient time to bring the project into a publishable / presentable form. However, Marc obtained satisfaction of learning the Twitter API, and learning first-hand that “tweets” are extremely noisy sources of information for stock prediction! We were able to uncover several instances that had minute predictive power, but not statistically significant. More work would need to be completed to filter and process tweets to discard meaningless and irrelevant data.