基于中文股票博客的情感分类

基于中文股票博客的情感分类 Research of Sentiment Classification Based on the Chinese Stock Blog 李亚珍 LI Yazhen 1 first-author 李晓戈 LI Xiaoge 1 于根 YU Gen 1 西安邮电大学计算机学院 西安邮电大学计算机学院 Computer College,Xi’an University of Post and Telecommunications Computer College,Xi’an University of Post and Telecommunications 根据新浪财经股票博客的特点,使用点互信息（PMI）方法构建了股票情感词典,在所构建的股票情感词典和现有的台湾大学情感词典基础上,结合经典贝叶斯方法对新浪财经博客的情感分析进行研究,并且在中文分词、自然语言处理（NLP）技术的基础上研究文本句法结构对股票博客文本情感分类结果的影响.实验结果表明:考虑文本中的句法细节以后,如词语搭配、否定词和连词等,使用PMI股票情感词典+贝叶斯方法,宏平均准确率从60.19%提高到80.50%,宏平均召回率从原来的59.35%提高到78.70%,宏平均F1值也由59.77%达到了79.60%. According to the characteristics of the financial Sina stock blog,this paper constructs a stock sentiment dictionary by using pointwise mutual information（ PMI） method and analyses the financial Sina blog sentiment by employing the classical Bayesian methods based on the constructed stock sentiment dictionary and the existing Internet National Taiwan University sentiment dictionary. Based on Chinese word segmentation and natural language processing（ NLP）,the paper discusses the influence of the text syntactic structure on the stock blog text sentiment classification. Considering the syntactic details of the text,such as collocations,negations or conjunctions,the experimental results show that the macro average accuracy is increased from 60. 19% to 80. 50%,the macro average recall is increased from the original 59. 35% to the current 78. 70%and the macro average F1 value is also improved from the previous 59. 77% to the present 79. 60% by using PMI stock sentiment dictionary and the Bayesian method. 情感词典 PMI算法 自然语言处理 emotion dictionary PMI algorithm natural language processing 10.14188/j.1671-8836.2015.02.010 TP391.12 国家自然科学基金项目（61373116）;陕西省工业攻关项目（2011K06-2）;陕西省普通高等学校重点学科专项资金建设项目资助（112-1602）;西安邮电大学研究生创新基金（112-2033） 2015-02-01 2021-03-31 2 61