Financial Time Series Prediction Based On Text Analysis And LSTM

By collecting social idle funds,the financial market helps enterprises and governments obtain funds,reduces financing costs,and improves social development efficiency.However,financial market volatility also affects the development of the national economy,changes in people's living standards,and corporate financing environment.At the same time,with market volatility and economic cycle changes,it also brought about negative effects such as financial crisis and economic downturn.In order to reduce the role of negative impacts,financial market regulation and adjustment measures are needed,which requires forecasting the operating conditions and future trends of financial markets.According to the principle of historical similarity,the historical data of the securities market can reflect the market change pattern.Therefore,the financial characteristic data is organized into financial time series according to the time label,which can effectively predict the financial indicators.At the same time,the changes in the securities market are not only affected by the national economic trends and the state of business operations,but also by the willingness and psychological expectations of investors.Therefore,by increasing the method of text analysis and extracting the emotional index of investors,it can be more effective and comprehensive.Forecast the trend of the securities market.This paper establishes a financial time series prediction model based on long-term and short-term memory(LSTM)neural network,quantifies the emotional characteristics of the securities market through text sentiment analysis method,and predicts the future securities market index through emotion and financial index.The research content of this paper is as follows:(1)Research and implementation of text sentiment analysis methods.Choosing financial news as the source of emotional characteristics index,according to the characteristics of financial related texts,a dictionary-based sentiment analysis method was selected to extract the financial sentiment characteristic index.Aiming at the problem that the general text sentiment analysis dictionary can not be fully applied to the field of financial sentiment analysis,resulting in inaccurate sentiment analysis results,this paper adds the emotional feature words in the financial field,so that the results of text sentiment analysis containing financial features are more accurate and can effectively Combined with financial feature time series.(2)Analyze and improve the construction method of emotional time series.Aiming at the consistency of weights of daily text sentiment index,the text relevance analysis based on K nearest neighbor(KNN)algorithm is designed and implemented.The value of relevance between text and financial domain is used as the weight of the text,and the text sentiment index and The text relevance is combined and standardized to obtain the emotional index of the day.This method effectively improves the accuracy of the emotional time series,reduces the influence of unrelated text on the emotional time series,and improves the accuracy of the prediction model.(3)Research and implement a financial time series prediction model based on long-term and short-term memory(LSTM)neural network.This predictive model preserves long-term memory and forgets useless memory.It not only uses financial time series,but also adds emotional time series,taking into account the impact of investor sentiment fluctuations on securities market trends and forecast results.Compared with the financial index-based LSTM neural network,the cyclic neural network(RNN)and the particle swarm BP neural network(PSO-BP),the results of the prediction model are more accurate.(4)Because the securities market does not exist in isolation,it will be affected by factors such as enterprise development,related policies,and changes in domestic and foreign forms.At the same time,there are mutual exchanges between financial markets.Capital and information are freely circulated among countries' markets around the world.There is a correlation between the financial indices of the securities market.Therefore,this paper proposes an improved prediction model based on LSTM neural network,adding time series of financial indicators in other securities markets,and improving the prediction accuracy of the target financial index through optimization training.