Incorporating News Summaries for Stock Predictions via Graphical Learning

Financial news shows significant impacts on stock movement. Previous stock movement prediction models mainly incorporated textual information without considering text quality, resulting in irrelevant text misleading prediction. Meanwhile, the models do not provide key news about the stock market to help investors make more rational investment decisions based on textual information. In this paper, we propose a framework for incorporating news summaries into stock predictions (SumSP) via graphical learning. It uses a graph-clustering mechanism to extract financial news closely related to stock price fluctuations as summaries and then predicts stock price movement based on the impact of the summaries. The model ultimately yields meaningful, thematically diverse and economically meaningful summaries, as well as better prediction results. Experiments demonstrate the effectiveness of our model, comparing to state-of-the-art methods.