基于财经新闻V-A情感分歧的股票价格预测方法研究

Stock Price Forecasting Based on V-A Sentiment Divergence of Financial News

摘要: 互联网财经新闻已然成为股市投资者获取股票相关信息的首要来源，其引发的投资者情感波动必然会对股市造成影响，充分挖掘财经新闻潜在的情感信息，可以更好地洞察股票市场趋势。而现有的财经新闻情感分析，常忽略新闻情感信息的多样性，只考虑单维度情感，无法量化复杂情感隐含状态，造成情感信息缺失。因此，本文旨在利用财经新闻板块V-A多维度情感分歧对股票市场进行研究，提高股票板块价格预测精度。利用CNN-LSTM组合模型提取文本局部特征与语义特征，构建财经新闻连续维度V-A情感计算模型，从连续多维度量化复杂情感，更全面地表示情感信息，进而精确地计算财经新闻情感分歧。后结合情感分歧与股票价格构建股票板块价格预测模型，采用GridSearchCV对SVR预测模型进行参数寻优。模型预测准确率进一步提升，其中，制造业板块平均绝对误差为0.2030，证明本文V-A连续计算情感模型测度新闻情感，可以有效地提高股票预测准确率。

Abstract: Internet financial news has become the primary source for stock market investors to obtain stock-related information. Investors’ sentiment fluctuations triggered by financial news will inevi-tably affect the stock market. Fully mining the potential sentiment information of financial news can provide better insights into stock market trends. However, prior financial news sentiment analysis methods often consider single-dimensional sentiment, ignore the diversity of news sentiment in-formation, and cannot quantify the complex sentiment, resulting in a lack of sentiment information. Therefore, this article aims to use the V-A multi-dimensional sentiment divergence of the financial news sector to study the stock market and improve the accuracy of stock sector price forecasts. The CNN-LSTM combined model is used to extract the local and semantic features of the text, and the continuous-dimensional V-A sentiment computing model of financial news is constructed, which quantifies complex sentiment from continuous multi-dimensional and expresses sentiment infor-mation more comprehensively, so as to accurately compute the sentiment divergence of financial news. After combining sentiment divergence and stock prices, the stock sector price prediction model is constructed, and GridSearchCV is used to optimize the parameters of the SVR prediction model. The accuracy of model prediction is further improved. Among them, the average absolute error of the manufacturing sector is 0.2030, which proves that the V-A continuous computing sen-timent model to measure news sentiment can effectively improve the accuracy of stock prediction.