中文微博句子倾向性分类中特征抽取研究

针对中文微博句子倾向性分类问题,在充分降低由于情感词典的扩充工作带来系统开销的基础上,抽取了中文微博句子中标点符号、情感词权重、词汇级和句法级等新型平面和结构化特征,探索了有效的特征选择方法.在基准COAE和NLP&CC中文微博语料上进行双向交叉和独立实验,并研究了有效的不平衡性语料的处理方法.实验结果表明:采用该文提出的特征后,中文微博句子倾向性分类的性能得到显著提升.

The Research on Feature Extraction of Polarity Classification of Chinese Micro Blogging

According to Chinese micro blogging sentence polarity identification problem,while fully reducing due to the emotional lexicon expansion work brought on the basis of system overhead,many novel flat and structural features,e.g.punctuation,sentiment word weighting,lexical and syntactic level information,from Chinese micro blogging,together with the effective feature selection method has been extracted.In-depth bidirectional and independent experiments on both COAE and NLP&CC,along with the effective imbalance corpus handling method has been conducted.Evaluation results show that the effectiveness of our novel features.Its also show that the model significantly outperforms existing model currently in the research field.