微博情感倾向分类是分析微博语句带有正向、负向或者中性情感。已有的研究大多根据手工标注微博情感极性进行有监督或半监督分类。该文主要结合了稀疏自动编码器和支持向量机,自动提取情感特征,实现了无监督的微博情感分类。实验结果表明:稀疏自动编码器在微博情感倾向分类精度上基本和手工标注情感特征算法相近,但是微博文本形式多变,自动提取情感特征适应性更强。%Micro-blog sentiment classification analysis is to analyze the emotions that macro-blog statements contain, such as positive, negative or neutral emotions. Most of the existing research is based on manual annotation of micro -blog emotion to conduct supervised or semi -supervised classification. This paper automatically extracts emotional characteristics and achieves unsupervised micro-blog sentiment classification by integrating the sparse autoencoders with support vector machines. Experimental results show that sparse autoencoder is applied to micro-blog emotion tendency classification, although the accuracies are close to manual annotation emotional characteristics algorithm, since micro-blog text is changeable, the model with automatically extracting emotional characteristics is adaptable.
展开▼