Stock prediction under COVID-19 based on LSTM

Stock prediction has been considered as a significant problem in both economics and computer fields. Since stock historical price is extraordinary crucial that it can affect investors' attitude towards whole market, and people's emotions will influence the stock price in the future, sequence-based predict models have received much attention these years. In this paper, we propose using LSTM to predict accurate single stock price, with whose history index as experience. To take public mood into account, numerous tweets were collected and quantified by deep learning method. To revise the LSTM model, we applied attention mechanism to extract vital factors and realign weights between history index and public sentiment. Experiments demonstrate that this model performs well, for the forecast price curve almost coincides with the real price curve. But in some extreme circumstance, models without sentiment as input performed better, which indicates that population sentiment can be a noise of prediction.