基于情感项区分极性可信度的文本情感分类

针对语义情感知识的文本情感分析的局限性,本文提出情感项区分极性可信度的文本情感分类方法.首先,基于核心谓词结构提取修饰主题的情感项.接着,利用改进的互信息方法计算情感项可信度,选取其中可信度前N的情感项.然后,利用改进的词频-逆向文件频率(TF-IDF)算法标记前N个情感项的正或负倾向符号.最后,基于基因表达式编程分类技术和谭松波博士提供的语料集,利用训练集训练分类模型,并使用测试集检验分类精度,实验结果表明本文提出的方法具有良好的效果.

Text Sentiment Classification Based on Credibility of the Emotional Items

To overcome the limitations of sentiment analysis based on emotional knowledge, text sentiment classification using emotional credibility, that is the ability of distinguishing different emotional polarities, is proposed. First, emotional items modifying subject are extracted based on the HED structure. Then improved mutual information is used to calculate each emotional item’s credibility and select Top-N credibility. And the improved TF-IDF mark emotion item’s polarity is used. Finally, Gene Expression Programming and Dr. Tan Songbo’s dataset are used to train and check the method. The results show that the method has a good effect.