基于BPNN和SVR的股票价格预测研究

摘要

Prediction models combining with sentiment analysis and machine learning method are proposed to predict the price of the stock based on the stock news and the machine learning algorithm of Back Propagation Neural Network (BPNN) and Supportive Vector Regression (SVR).20 stocks with high transaction amount are chosen and the related news data are crawled.The words with high frequency are scaled and ranked to produce a preciser sentiment dictionary with granularity [-5,+5].Moreover,considering the 9 different combination among the negative words,adverb of degree,hypothesis words and the sentiment words,semantic rules are generated with different weight to the sentiment words to modify the emotion scale.Finally,the prediction model is built on the BPNN and SVR algorithm and a comparison is conducted.The result shows that the sentiment dictionary and semantic rule performs well in predicting the price and a more accurate match between the emotion scale and the price trend is achieved.Moreover,the SVR algorithm can produce a higher accuracy and less mean squared error.%将情感分析和机器学习方法相结合,以股票新闻数据为基础,分别采用BP神经网络(BPNN)和支持向量机回归(SVR)两种方法,对股票价格进行预测分析.首先选取交易量较大的20只股票作为研究对象,抓取相关的新闻数据.然后邀请专家对高频词进行人工情感打分,得到一个针对性更强、粒度更细[-5,+5]的情感词典,同时考虑否定词、程度副词、假设疑问词和情感词间的相互作用,归纳出9种常见的语义规则,给不同的语义规则下的情感词赋予不同的权重,对情感值进行修正.最后分别采用BPNN和SVR两种方法构造股价预测模型,并对模型的预测效果进行对比分析.结果表明,文章提出的人工情感词典和语义规则在股价预测领域表现良好,情感得分正负方向与股价涨跌方向的一致程度显著提升,另外,SVR股价预测模型的均方误差更小,且股价走势方向正确率更高.