无监督的财经新闻情感标注和情绪指数生成  

Sentiment Labeling and Sentiment Index Generation in Unsupervised Financial News

作　　者：邵元海 何洋 吕孝敬 SHAO Yuan-hai;HE Yang;LÜXiao-jing(College of Management,Hainan University,Haikou 570228,China;College of Economics,Hainan University,Haikou 570228,China)

机构地区：[1]海南大学管理学院,海南海口570228 [2]海南大学经济学院,海南海口570228

出　　处：《海南大学学报（人文社会科学版）》2023年第3期84-95,共12页Journal of Hainan University （Humanities & Social Sciences）

基　　金：国家自然科学基金委面上项目(11871183);国家自然科学基金委地区项目(61866010);海南省自然科学基金高层次人才项目(120RC449)。

摘　　要：财经新闻报道作为金融市场重要的信息来源,其情感倾向与市场走势有着密切联系。然而财经新闻具有专业性、客观性、无标注的特点,对其情感倾向进行精准量化往往十分困难。因此,本文设计了两阶段的财经新闻情绪指数提取方法,在第一阶段,针对财经新闻无标注的问题,本文通过改进的SO-PMI算法构造财经新闻领域情感词典来对新闻进行无监督标注;在第二阶段,为了提取精确的新闻情感强度值,本文构造了新闻情绪指数,先利用已标注的新闻数据训练情感分类模型从而生成类别概率,然后通过概率值计算得到情绪指数。为了进一步验证该方法的有效性,将生成的情绪指数结合股市历史价格数据来对上证股指波动趋势进行预测。结果表明,基于注意力机制的预测模型在添加情绪指数变量后,准确率提升了3%—5%,说明新闻情绪指数对于股指波动有较好的表征作用。Financial news reports are an important source of information in the financial market,and their senti‐ment tendencies are closely related to market tends.However,financial news has the characteristics of profes‐sionalism,objectivity,and non-labeling,and it is often very difficult to accurately quantify its sentiment tendencies.Therefore,this paper designs a two-stage extraction method of financial news sentiment index.In the first stage,for the unlabeled issues of financial news,a sentiment dictionary in the field of financial news is constructed through the improved SO-PMI algorithm to label news in an unsupervised way.In the second stage,the news sentiment index is built to extract the accurate news sentiment intensity value.Initially,the labeled news data is used to train the sentiment classification model so as to generate the category probability,and then the sentiment index can be obtained through the calculation of probability value.In order to further verify the effectiveness of this method,the generated sentiment index is combined with the historical price data of stock market to predict the fluctuation of the Shanghai Stock Index.The experimental results show that the accuracy of prediction model based on the attention mechanism can be improved by 3%-5%after the addition of the sentiment index variable,indicating that the news sentiment index has a better representation on the stock index fluctuation.

关 键 词：财经新闻 无监督文本标注 情绪指数 注意力机制 

分 类 号：F832.5[经济管理—金融学]