Using Tweets for single stock price prediction

Stock price, been studied for hundreds of years, is one of the most versatile thus hardly predictable things that is deeply rooted in the modern economy. With the trading frequency reaching sub-second and beyond, more advanced real-time stock price prediction tools would be highly demanded in addition to traditional financial analysis. In this work, we applied SVM and Naïve Bayes algorithms to this field and successfully built the connection between stock price and selected keywords frequency from Twitter tweets. We firstly used TF-IDF method to filter out the candidates of keywords that is most relevant to stock price. Then based on these keywords, we systematically studied the testing error using cross validation for different number of keywords and unit time length. Finally, with the optimized keywords and unit time length, we plotted out the predicted stock price compared with the real stock price. We found that by using Gaussian kernel, SVM gives the lowest testing error, while Naïve Bayes catches pretty accurately the trend of stock price change over a longer period of time. Though limited by the available data size, the optimistic outcome opens a new approaching to real-time stock price prediction.