Design And Implementation Of Stock Forecasting Model Based On News Sentiment Quantification And LSTM Network

The stock market is an important part of the capital market,enterprises can get more funds in the market through listing financing,while investors want to analyze the information in the market,judge the up and down trend of each company's stock,and carry out the buying and selling operation to obtain income.However,investment is often accompanied by risks.If we don't make a specific analysis of the factors that affect the rise and fall of stocks,we will blindly invest,which will also bring some losses to investors.In order to make use of financial news information to predict the rise and fall of stocks more accurately,this paper studies the data collection methods of stock price and its related financial news,uses the data collected from three stocks to carry out feature engineering,and proposes the price features and emotional quantitative features.This paper establishes the stock up and down prediction model,takes the extracted features as the input to get the stock price trend results,and carries on the contrast experiment to the existing models.A kind ofIn order to collect the data used in stock forecast,the tushare interface is used to collect the detailed data of stock price,and a method of stock news collection based on selenium and beautiful soup is designed.The process of obtaining news web page by simulating browser is used to locate the news data in DOM(Document Object Model)of HTML(Hyper Text Markup Language)Location in tree,and then automatically collect the news data of the required stocks.This paper designs a kind of news sentiment quantification model based on Naive Bayes,through which we can classify the sentiment of news text,and quantify all the news of a specific stock every day to get the sentiment value of the stock news on that day.The experimental results show that the model has a high accuracy for emotional classification of news texts.Using LSTM attention network,the paper establishes a stock price up and down prediction model,takes stock price characteristics and news emotion characteristics as model inputs,and obtains the predicted up and down results.And the prediction results of SVM(Support Vector Machine),LSTM(Long Short-Term Memory)and other models were compared.The results show that LSTM attention network has better prediction effect than traditional SVM model and single LSTM model,and the model can achieve higher accuracy after adding news emotion features.