Stock Movement Prediction via Temporal Convolutional Network and Interactive Attention Network

Abstract

Stock movement prediction is an important study that can provide investors with some reliable trading signals and long-term stable investment returns. Although many existing studies show that using text data of social platforms and historical share prices can effectively predict stock movement, the existing methods still have some limitations. For example, (i) future information is inevitably used to obtain the features of text and price; (ii) The interaction between text data of social platforms and historical share price data was not considered. In order to alleviate these restrictions, we propose a novel method to predict the stock movement based on Temporal Convolution Network (TCN) and Interactive Attention Network (IAN), in which TCN can effectively avoid using future information when obtaining the features of text and price. In addition, IAN can get the interactive feature between social texts and historical share prices. In the end, we employ a three-layer neural network classifier to predict the stock movement. The experimental results show that the proposed model can obtain a competitive results compared with other existing methods.