摘要
股票价格的变动是投资者在股票市场关注的焦点,所以股价趋势预测一直是量化投资研究的热门话题。传统的机器学习预测模型难以处理非线性、高频率、高噪声的股价时间序列,使得股票价格趋势的预测精度低。为了提高预测精度,针对股票价格数据的时序性特征,提出用结合经验模态分解(EMD)、投资者情绪和注意力机制的双向长短期记忆神经网络来对股票价格进行涨跌预测。首先使用经验模态分解算法提取股票价格时间序列在不同时间尺度上的特征,并通过构建金融情感词典来提取上一个股票交易日收盘后至下一个交易日开盘前文本的投资者情绪指标,最后使用注意力机制优化的BiLSTM模型对下一个股票交易日进行涨跌预测。在股票价格序列的数据集上进行实验,结果表明,改进后的BiLSTM模型较改进前的BiLSTM模型,准确率从58.50%提升至71.26%;预测为涨的精确率从58.20%提升至70.06%,预测为跌的精确率从59.34%提升至72.36%;预测为涨的召回率从59.85%提升至73.41%,预测为跌的召回率从57.73%提升至69.11%;预测为涨的F1值从58.60%提升至71.61%,预测为跌的F1值从58.08%提升至70.53%;最终通过与长短期记忆(LSTM)网络、基于Attention机制的LSTM(Attention-LSTM)、支持向量机(SVM)、极端梯度提升(XGBoost)等4种典型的股价涨跌预测模型结果对比,验证了所提模型的准确有效性。
Stock price change is the focus of investors in the stock market,so stock price trend prediction has been a hot topic in quantitative investment research.The traditional machine learning prediction model is difficult to deal with nonlinear,high frequency and high noise stock price time series,which makes the prediction accuracy of stock price trend low.In order to improve the prediction accuracy,a Bidirectional Long Short-Term Memory(BiLSTM)neural network stock price trend prediction model based on Empirical Mode Decomposition(EMD),investor sentiment and attention mechanism was proposed in accordance with the temporal characteristics of stock price data.Firstly,the empirical mode decomposition algorithm was used to extract the characteristics of stock price time series on time scale,and then the emotional characteristics of the text from the closing of the previous stock trading day to the opening of the next trading day were extracted by constructing a financial emotional dictionary.Finally,the BiLSTM model combined with attention mechanism was used to predict the rise and fall of the next stock trading day.The experimental results on the data set of stock price series show that the accuracy of the improved BiLSTM model increases from 58.50%to 71.26%.The prediction precision of rise increases from 58.20%to 70.06%,and the prediction precision of fall increases from 59.34%to 72.36%.The recall predicted to rise increases from 59.85%to 73.41%and the recall predicted to fall increases from 57.73%to 69.11%.The F1 value predicted to rise increases from 58.60%to 71.61%,and the F1 value predicted to fall increases from 58.08%to 70.53%.Finally,the accuracy and effectiveness of the proposed model were verified by comparing with the results of four typical stock price rise and fall prediction models,namely,Long Short-Term Memory network(LSTM),Attention-Long Short-Term Memory(Attention-LSTM),Support Vector Machine(SVM)and eXtreme Gradient Boosting(XGBoost).
作者
赵帅斌
林旭东
翁晓健
ZHAO Shuaibin;LIN Xudong;WENG Xiaojian(College of Mathematics and Information,South China Agricultural University,Guangzhou Guangdong 510642,China)
出处
《计算机应用》
CSCD
北大核心
2023年第S01期112-118,共7页
journal of Computer Applications
关键词
双向长短期记忆神经网络
注意力机制
经验模态分解
投资者情绪
股票涨跌预测
Bidirectional Long Short-Term Memory(BiLSTM)neural network
attention mechanism
Emprical Mode Decomposition(EMD)
investor sentiment
stock price trend prediction
