A Deep Learning Model for Dimensional ValenceArousal Intensity Prediction in Stock Market

Abstract

This paper proposes a dimensional valence-arousal method to define sentiment status in the stock market. In the past, many kinds of research have focused on the valence sentiment on stock messages because it represents the stock trend such as upward and downward. In this case, if the stock price jumps or collapses (positive/negative trend) in the short term, the investor will necessarily need to immediately trade at this moment, but some case is not. Therefore, the valence-arousal method can be used to define the trend intensity and trading intensity for a stock message of the stock market. In order to obtain a powerful prediction model to learn the intensity of trend and trading of a stock message that we propose a keyword-based attention network into Hierarchical Attention Networks (HAN), namely HKAN model, to learn the relation between dimensional sentiments (trend and trading) and stock messages. The experimental results show that our proposed HKAN model for stock VA prediction has outperformed other baseline models such as HAN and Hierarchical Hybrid Attention Networks.