This paper introduces a method that can be used for trading Bitcoin. It is based on specific word usage within Twitter messages (also known as 'Tweets') by Twitter users. The goal of this research is to show if predictions regarding the Bitcoin price can be made by utilizing textual relationships of Tweets for specific users and if user-specific information carries importance when making such predictions. We instantiate matrices that are specific to a user's Tweets over a set amount of time. These matrices are used to train Convolutional Neural Nets (CNNs), which predict whether or not a profitable trade can be made. Keywords are found by using TF-IDF, which is a general technique that is used to find important keywords in a set of documents (i.e. Tweets in our case). The user-specific results are aggregated by day, resulting in a single prediction per day by utilizing a Logistic Regression (LR) and arithmetic mean based method. Success is measured by calculating the Precision and Recall scores. These scores are related to the, assumed to be goal of the trader, which is to increase its overall capital. The results show that it is preliminary to say whether or not the method is of significance to a trader, but the work aims to serve as a baseline for future research.
The remainder of our paper can be found here.