Application Of Short Term And Long Term Memory Neural Network In Stock Trend Prediction

Stock trend prediction,as a research hotspot in the field of data mining,has a wide range of application prospects in the financial market.In recent years,with the rapid development of machine learning and deep learning,deep learning methods are widely used to solve some problems in the financial industry.In this paper,through the study of the application of long and short-term memory neural networks in stock trend prediction,and in view of the shortcomings and deficiencies of related algorithms in the field of stock trend prediction at the present stage,new solutions and improvement measures are proposed through the methods of deep learning and feature learning.The main research contents and findings of this paper are summarized as follows.First,a CNN-BiLSTM-Attention model is proposed to address the problem of low accuracy and stability of stock model prediction.The model consists of an input layer,a CNN layer,a BiLSTM layer,an Attention layer,a fully connected layer,and an output layer,with the CNN layer consisting of a stack of convolutional and maxpooling layers,and a Dropout layer with randomly discarded nodes at the end of the CNN and BiLSTM layers,respectively,to prevent overfitting.Sequence information is combined,and the Attention layer learns the importance of each feature from the sequence.Stock data can be analyzed more circumspectly in terms of spatiotemporal characteristics.Then,to address the problem of predicting only basic stock information such as opening price,closing price,high price,low price and volume as features,we propose to fuse various improved technical indicators and predict them by the aforementioned CNN-BiLSTM-Attention model.Stock technical indicators are the neutralizing feedback of many factors in the financial market and are interpretations of stock trends.Single technical indicators are easily falsified by the main forces,but they cannot be falsified for all of them,so multiple technical indicators should be integrated for forecasting.Finally,to address the problem of using only financial time series as features for prediction,we propose to fuse news sentiment features and predict them by the aforementioned CNN-BiLSTM-Attention model.The crawler crawls the news text of the website,and after de-duplication and de-noise processing,the text analysis algorithm is applied to quantitatively analyze the sentiment tendency of the news text,and the features after fusing news sentiment are used as input to predict the stock trend,and the diversity of input features improves the accuracy of the model.