【摘要】 本文主要对股票预测的方式进行改进，使预测结果更接近真实数据。以往的股票价格预测研究大多简单地将股票价格作为序列数据，通过模型进行训练预测，或者只是通过分析新闻文本、股民评论的情感倾向预测股票价格的涨跌，这都不能全面地对股票价格进行考量。本文通过参考影响股票实际价格的多种因素，对股票价格预测结果展开研究。本研究提出新的模型M,该模型结合了时间序列预测以及文本情感分析两种方法，采用的均为LSTM的拓展模型，分别是DA-RNN、BiLSTM-Attention。一方面，通过分析对比XGBoost、LSTM与DA-RNN的实验结果验证了DA-RNN在时序预测实验中的有效性；另一方面，采用BiLSTM-Attention模型作为调整股票预测值的主要方法，使预测结果更加具备可解释性。本研究通过改进的股票价格预测模型M来进行股票价格的预测。实验结果表明，在海康威视股票数据集中，当文本情感倾向的权重值设置为0.01时，MAE、RMSE值均达到最小，即预测效果最好。更多还原

【Abstract】 This study mainly optimizes the stock forecasting method to make the stock price forecast closer to the real data.Most of the previous researches on stock price prediction simply take stock price as serial data and make training prediction through the model, or just predict the rise and fall of stock price by analyzing the emotional tendency of news text and shareholders′ comments, which cannot comprehensively consider stock price.By referring to the factors that affect real stock prices, the paper investigates the results of stock price predictions.This study proposes a new model M, combines two methods of time series prediction and text sentiment analysis, and uses the extended model of LSTM, namely DA-RNN and BiLSTM-Attention. On the one hand, the experimental results of XGBoost, LSTM, and DA-RNN are analyzed and compared to verify the effectiveness of DA-RNN in the time series prediction experiment; on the other hand, BiLSTM-Attention is used as the main method to adjust the stock forecast value, and makes the results more interpretable.This study uses an improved stock price forecasting method M to predict stock prices. The results demonstrate that in the Hikvision stock data set, when the weight value of text sentimentality is set to 0.01, the MAE and RMSE values are both minimized. This means the prediction is the best.更多还原