Sentiment analysis of stock markets using a novel dimensional valence–arousal approach

Sentiment analysis has been used in many studies to predict stock market trends. Current sentimental analysis approaches focus only on the upward and downward movement of the price, which is not sufficient for more precise prediction of stock sentiments. Previous studies have focused on the trend (valence) regarding stocks because it represents the upward and downward trend of a stock. However, investors should be attention on trading (arousal) because the stock price increases or decreases even quickly or slowly; they may not to do trade. Therefore, this paper applies the concept of dimensional valence–arousal to define sentiments which intensities of trading and trends of the stock market for data annotation. The paper proposes a deep learning prediction model to predict the intensities of stock dimensional valence–arousal (SDVA), called the hierarchical title–keyword-based attentional hybrid network (HAHTKN). The paper modifies the hierarchical attention network (HAN) to our proposed HAHTKN model, comprising (1) a title encoder, (2) a keyword encoder, (3) a title–keyword encoder, (4) word-level encoder, (5) a sentence-level encoder and (6) stock DVA prediction layer. The experimental results show that the proposed HAHTKN prediction model for the SDVA task outperformed other baseline machine learning models and HAN-based models.