Research On Time Series Prediction Method Based On News Events And Deep Learning

Time series research objects are phenomena that change over time and are widely found in life.For example,through the prediction of the box office of the movie,the theater can better arrange the film;support the company and the country by predicting the economic situation of the region.Therefore,studying this issue is of great significance to the national economy and the people's livelihood.ARIMA requires that the time series data is stable.SVM is limited by the rare types of kernel functions and only has good effects in some specific situations.RNN and its improved models can not only remember the historical law information to the present,but also can fit very complex nonlinear problems.However,these methods cannot memorize long-term historical information.In addition,historical data cannot express unexpected events,which also affects the accuracy of prediction.In order to solve the above problems,this paper proposes a time series prediction method based on news events and deep learning--NS_TCL.First of all,NS_TCL uses realtime Internet-related news to solve the historical law lacks the expression of emergencies.Using the word embedding,Self-Attention and LSTM to construct a news analysis model to express the news text and support the subsequent prediction process.Then,in order to fit time series data,firstly use CNN(Convolutional Neural Networks)to convolve along the time axis to extract historical long-term and short-term dependent features from time series data;then build prediction model based on LSTM suitable for time series prediction.The combination of the two not only makes use of the characteristics of LSTM to deal with time series data,but also considers the influence of different length history laws on the prediction results,making the prediction effect better and more stable and avoiding the fraud of complex sequence changes.It makes predictions more stable and accurate,avoiding the spoofing of complex sequence changes.Using NS_TCL,TCL which is a NS_TCL model without news sentiment,TC_LSTM,and ARIMA to perform experiments on three stock data sets respectively,it is found that NS_TCL has the highest prediction accuracy in each data set.