SMPDF: stock movement prediction based on stock prices and text

Abstract

Stock movement prediction is a challenging task due to its dynamic and nonlinear characteristics. Different information sources provide a rich variety of perspectives and dimensions, which can be integrated to understand market dynamics more comprehensively. However, different information may have complex interrelationships and nonlinear features, the deep fusion method can make the model have a powerful representation learning ability, and automatically learn complex features and relationships. Therefore, how to deeply fuse numerical data and text to achieve interaction between different types of information to complete more effective prediction tasks is a difficult problem. In this paper, we propose a stock movement prediction model based on deep fusion (SMPDF), including text and stock price feature extraction, feature fusion and feature processing, which can successfully fuse text and stock price. Experiments show that the proposed method has better modeling performance under the SMPDF framework and brings greater improvement in prediction performance.