User2Vec: Stock exchange prediction using deep
learning of sentiment and social relations of twitter
users
Pegah Eslamieh, Mehdi Shajari
Abstract—Recurrent neural network approaches, especially LSTM, have been the best practices recently used in the field of stock market prediction. Most of these methods use historical market data and in some cases, the dominant direction of users and news every day. In most of the work done in the literature, the impact of users was considered equal, while users can have different effects. In this study, the idea is to transform each user into a vector (User2Vec) and then use these vectors to train the Recurrent Neural Network (RNN) and ultimately model the behavior of the users in the market. The proposed model, User2Vec is trained with both the extracted Tweeter features and market information simultaneously. This approach has been able to increase the size of training samples and effective modeling using a variety of factors including: the characteristics of each individual user, their impact on each other, and their impact on the market significant improvement in the predicting accuracy of stock market trends for the Dow Jones 30 index. In this study, Twitter sentiment, social analysis were used to analyze users, historical data of 30 Dow Jones Index stocks representing market information , the LSTM network to model user behavior, and ultimately market change trends. The accuracy obtained for predicting daily stock changes of Apple according to different models is over 93% and for the other stocks the accuracy is significant. The study results demonstrate the effectiveness of User2Vec in predicting stock market direction.
Index Terms—Stock market prediction, social network analysis, deep learning, user behavior modeling, financial market emotion analysis