Spatial-temporal attention-based convolutional network with text and numerical information for stock price prediction

Abstract In the financial market, stock price prediction is a challenging task which influenced by many factors. These factors include economic change, politics and global events that are usually recorded in text format, such as daily news. Therefore, we assume real-world information can be used to forecast market activity. However, only few works considered both numerical predict or analyse trends. preprocessed features model inputs; therefore, latent may lost because relationships between not considered. this paper, propose fusion network, i.e. spatial-temporal attention-based convolutional network (STACN) leverage advantages of an attention mechanism, neural long short-term memory extract for prediction. Benefiting from utilisation reliable highly relevant value extracted, improves overall performance. The experimental results on data demonstrate our STACN training scheme handle achieve high accuracy regression tasks. compared with CNNs LSTMs different settings, e.g. CNN data, news titles data. considering have mean squared errors 28.3935 0.1814, respectively. 0.0763. 0.0304, outperforming