Stock Composite Prediction Model Based On 1DCNN-LSTM And Text Mining

In the context of big data,the demand for financial data forecasting continues to increase,and deep learning has become a research focus in the field of financial forecasting.From the perspective of data fusion and model combination,this paper proposes a research paradigm of stock forecasting based on deep learning and text mining,and verifies the applicability of deep learning cutting-edge algorithms in the field of financial forecasting.In practice,it can not only provide investors with investment strategies,but also have great significance for the government’s macro-control and market supervision.The procedure of this paper are as follows:(1)A calculation model of investor sentiment index is constructed based on the text information of stock forums to capture the impact of investors’ irrational factors on stock price fluctuations;(2)An integrated empirical mode decomposition algorithm based on principal component analysis(EEMD-PCA)decomposes and reconstructs the stock price sequence to extract the effective information of the stock price sequence;(3)A hybrid neural network prediction model 1DCNN-LSTM is constructed.It learns the long-term dependence information of the stock price sequence,and constructs joint features by combining the feature learning of the two networks for different types of data at the same time,so as to predict the stock price fluctuation trend.The conclusions of this paper are as follows:(1)The investor sentiment index constructed in this paper has a significant correlation with stock price series,and the measurement of investor sentiment can improve the performance of a single price data set;(2)The information extraction of stock price series can effectively improve the signal-to-noise ratio of the data,avoid over-fitting of the prediction model during training,and improve the performance and generalization ability of the algorithm;(3)The hybrid neural network can combine the advantages of a single neural network to make up for the single model in the stock prediction task.Compared with the selected reference model,the composite prediction model constructed in this paper has significant preponderance in the task of predicting stock price fluctuations,and the improvement in various indicators such as classification accuracy is up to 16%.The innovations of this paper include:(1)The hybrid neural network model 1DCNN-LSTM is built,and the sub-network structure is improved for the financial time series forecasting,which fully combines the feature extraction capabilities of the 1D-CNN network and the LSTM network in the spatial and temporal dimensions.(2)A calculation model of investor sentiment index based on financial sentiment dictionary is constructed,and the relevant metrics are introduced into the model to comprehensively reflect the emotional fluctuations of investors.(3)The EEMD-PCA algorithm is constructed,and the signal decomposition and dimensionality reduction technology is applied to the stock forecasting task to mine multi-scale information of financial time series and reduce noise.