Abstract: The EL-based （Ensemble Learning） sentiment analysis method takes three different high-quality sentiment analysis models as basic classifiers, and combines these basic classifiers into a final classifier. In order to overcome the shortcoming of the traditional combination method that only considers the different results of the components, different features are also considered. Experimental results show that the accuracy rate of the proposed method on an English movie review corpus is 88.2%.

Key words: text sentiment analysis, sentiment classification, ensemble learning

摘要： 文本情感分类是指通过挖掘和分析文本中的观点、意见和看法等主观信息，对文本的情感倾向做出类别判断。基于集成情感成员模型提出一种文本情感分析方法。把基于改进的神经网络、基于语义特征和基于条件随机场的三个情感分类模型作为成员模型集成在一起。集成后的模型能够涵盖不同的情感特征，从而克服了传统集成学习中仅关注成员模型处理结果的不足。以公开语料进行实验，集成模型融合了多个成员模型的优势，分类正确率达到了88.2%，远高于任一成员模型的效果。

关键词: 文本情感分析, 情感分类, 集成模型