A HYBRID MODEL FOR STOCK PRICE FORECASTING BASED ON MARKET SENTIMENT AND DEEP LEARNING OPTIMIZATION

Abstract

If you can accurately predict the stock price, you may increase your returns while decreasing your investment risks. Combining deep learning, swarm intelligence techniques, sentiment analysis, and multi-source data affecting stock prices, this research constructs the MS-SSA-LSTM model. The first step in developing a unique emotion dictionary and calculating the sentiment index is to peruse the posts made on the East Money forum. After that, the Sparrow Search Algorithm is used to optimise the hyperparameters of the Long and Short-Term Memory networks (LSTM). Last but not least, LSTM is used to forecast stock values in the future by combining fundamental trade data with the sentiment index. Results demonstrate that the MS-SSA-LSTM model outperforms the competition and is very generalisable. Compared to ordinary LSTM, MS-SSA-LSTM improves R2 by an average of 10.74%. We found that: 1) Adding the sentiment index to the model may increase its prediction ability. 2) SSA is used to fine-tune the LSTM's hyperparameters, which enhances the prediction's effect and offers a neutral justification for the model's parameter values. Thirdly, short-term forecasts work better in the very volatile Chinese financial market.