Research On Stock Price Prediction Based On Sentiment Analysis And Deep Learning Mode

As one of the most active investments in the financial market,stocks have been favored by numerous investors.After over 30 years of development,China’s stock market has had a profound impact on promoting economic construction and social development.However,as the financial market is inherently dynamic,non-stationary,noisy,and chaotic,and stocks entail both risks and returns,there has always been great interest in academia and the financial industry on how to effectively predict stock price trends.The stock market is influenced by various factors such as politics,economy,market,news,and investor behavior.Researchers hope to leverage the power of technology to predict stock prices based on these factors affecting stock price movements,in order to minimize investment risks to the greatest extent possible.Traditional machine learning methods use linear models to predict stocks based on historical data,while stock time series are characterized by high-frequency disturbances,medium-frequency fluctuations,and long-term trends.Compared with traditional linear models,deep learning has highly non-linear operation and mapping capabilities,which can filter out time series market noise and extract deep features.Therefore,deep learning models are more suitable for predictive analysis of financial assets.Many scholars have shown through research that deep learning models have higher accuracy in prediction compared to traditional models.This article proposes a stock market prediction model based on time series neural networks,which combines historical trading data and other influencing factors to predict the trends of index and individual stock prices.The main research contents are as follows:1.Considering the impact of news on stock indices,a news-driven stock index prediction method based on Trellis Net and sentiment attention mechanism is proposed.The model uses web crawlers to obtain relevant news texts about stock market indices,and uses an LSTM-CNN network for sentiment analysis of the news texts to obtain a sentiment index.In addition,the stock trading data and news sentiment index are used as inputs to the model,and the sentiment attention mechanism is introduced to reveal the relationship between the two.The experimental objects include seven major indices such as S&P500,NYSE,DJI,NASDAQ,FTSE 100,Nikkei225,and SSE.The results show that the model proposed in this paper is superior to three other prediction models.2.In addition to the research on stock indices,the impact of investor sentiment and network search sentiment on individual stocks is also considered.Based on the adversarial learning mechanism of GAN model,a sentiment analysis and GAN-Trellis Net-based individual stock price prediction method is proposed.This method uses web crawlers to obtain stock comments from mainstream forums(stock bars,snowballs,Sina Finance)and Baidu search index,and uses word2 vec and LSTM-CNN models for sentiment analysis of comments to obtain stock sentiment data.The generator of the prediction model is constructed by a grid network(Trellis Net)to generate the predicted stock prices.Trellis Net establishes the connection between CNN and RNN in processing time series,which can better capture data features.The discriminator is constructed by a convolutional neural network(CNN)to distinguish the difference between the predicted results and the real stock prices,dynamically predicting stock prices from comment sentiment,Baidu search index,and trading data.Multiple industry representative stock trading data and stock bar comments are selected as the sample set for experiments.Compared with Conv LSTM and GAN-LSTM models,the results show that the GAN-Trellis Net model proposed in this paper has higher accuracy and stronger generalization,and can be applied to different stocks.