基于意见目标句抽取的中文股评情感分析方法

Sentiment Classification of Stock Comments in Chinese Based on Opinion Target Sentences Extraction

针对股评文章进行情感倾向性分析时,不包含情感倾向的干扰信息过多而影响分析的正确率的问题,提出了一种股评文章中意见目标句识别及抽取的方法,利用股评文章的特点,借助主动词识别及其情感预判,识别与股评情感分析直接相关的意见目标句,并在缺少领域词典的情况下,使用基于半监督学习的分类方法进行此类句子的情感倾向分析,最终依此得到整篇文章的情感倾向。实验证明上述方法能够较大程度上改善股评文章情感分析的准确性。

This paper proposed a method of sentiment classification of stock comments in Chinese based on opin- ion target sentences extraction, to avoid noisy information without emotional tendency in stock comments makes senti- ment classification a difficult work. At first, we used this method to recognize and extract the objective sentences based on the chrematistics of stock comments and main verb recognition, and remove the sentences without emotion tendency and focus on the analysis of the emotional ones. Then we got the tendency of these opinion target sentences using a semi-supervised learning method without dictionary. At last, we obtained the emotional tendency of whole ar- ticle. The results show that the statistical method can work well without lexicon on sentiment classification of stock comments in Chinese, and this method based on opinion target sentences extraction can improve the result to a certain extent.