Research On Stock Trading Based On Investor Sentiment And Deep Reinforcement Learnin

In stock trading,investors process large amounts of trading data and make trading decisions based on past experience,and are easily influenced by their own emotions and find it difficult to find potentially beneficial trading strategies.Deep reinforcement learning can not only give a way to converge investor sentiment,but also find favorable stock trading strategies.This paper focuses on a multimodal deep reinforcement learning strategy for stock trading,including convolutional neural networks(CNN)and bidirectional long short-term memory neural networks(Bi LSTM).First of all,the relevant theories introduced include deep learning,reinforcement learning and deep reinforcement learning,which provide a basis for the formulation of subsequent trading strategies.Second,various pictures are generated from stock trading data and used as inputs to the CNN layer.The features extracted through the CNN layer are divided into columns and fed into the Bi LSTM layer.This paper changes the actions,neural network framework,and reward of reinforcement learning agents,and takes the probability of selling,buying,and holding as the final output,which is model one.On the basis of model 1,the emotional factors of shareholders are added,the preprocessed stock bar posts and stock history data are stitched,and the dynamic time series laws are learned through the long short-term memory neural network(LSTM)to extract features.The learned features and the features of the previous model in the Bi LSTM layer are stitched in columns as an environment for reinforcement learning,so as to further improve the performance of the model,so as to obtain Model 2.Using real stock data,experimental comparison of multiple models shows that the deep reinforcement learning model 2 with emotional factors has achieved better results.Therefore,the ablation experiment was carried out,and it was further concluded that adding emotional factors could improve the performance of the model.To verify the performance of Model 2,we backtested the data of 4 stocks in Shanghai in 2021 and 2022.When using 2021 data,the account made a profit of 10.03%,and the Shanghai Composite Index rose by 4.80% over the same period,and our account profit margin is higher than the increase of the Shanghai Composite Index.When using the 2022 data,the account lost 5.94%,which was lower than the 15.13% decline in the Shanghai Composite Index during the same period.Therefore,the model proposed in this paper has good robustness.