Stock Price Movement Prediction Using StockTwits and Topic-Sentiment Neural Network

Abstract

In this thesis, StockTwits - a social network of investors and traders that has similar structures to Twitter - is analyzed to extract topics and sentiments about a certain stock during a xed length of time using an Topic Sentiment Latent Dirichlet Allocation (TSLDA) and use the extracted topic-sentiment as features of a neural network to make a prediction of stock price movement. TSLDA is specifically studied in detail to develop a parallelization of the Collapsed Gibbs Sampling process.