摘要： 为利用情感文本不同侧面的信息，提出一种基于状态和行为描述的情感分类方法。将情感文本的描述分为情感的状态和行为2个视图，并利用2个视图的融合进行情感分类。为自动获得2个视图，人工标注了相应的语料，通过二元分类器构建状态和行为检测系统。在此基础上采用组合分类器方法融合2个不同的视图。实验结果证明，该方法在3个领域中文情感分类任务上的分类效果均有所提高。

关键词: 中文信息处, 情感分析, 监督学习, 最大熵, 分类器融合

Abstract: In order to make use of different information of sentiment text. This paper proposes sentiment classification method based on status and action description. It aims to classify the sentimental text into two views called status views and action views and employs them in supervised sentiment classification. In order to get the two views, it manually annotates corpus and trains a binary classifier with the corpus to detect the two views. Each view is trained to be a single classifier and an ensemble of the individual classifiers is employed to improve the effect of the classification. Experimental results show the classification effectiveness of the proposed method has improvement on three domains of Chinese sentiment classification tasks.

Key words: Chinese information processing, sentiment analysis, supervised learning, maximum entropy, classifier fusion

中图分类号: 

• TP18