Enhanced Stock Movement Prediction with Event Graph and Dynamic Sentiment Analysis

Stock movement prediction is of paramount importance in financial markets, informing sound investment strategies. However, traditional numerical analysis often falls short due to market volatility and the inherent challenge of modeling dynamic change. We propose a Dynamic Event and Sentiment Interaction Graph Attention Network (DESIGN) method integrating numerical characteristics, news and comment text, with the added dimension of an event graph tracking the chain of events from news text. The method follows a multi-step approach for stock movement prediction, it starts by classifying the news text and then performs reasoning through the event graph to obtain event-inference-based indicators. These indicators are then fused with the ones derived from the raw numerical data, resulting in evolution features. Then, the method classifies the sentiment of comment text and dynamically constructs a stock relationship graph based on this result. Finally, a graph neural network is employed for prediction. Extensive experiments verify DESIGN's superior performance in stock movement prediction.