Abstract: Methods of text orientation analysis are studied and an approach based on sentiment ontology is presented.Firstly，sentiment Ontology is constructed with the sentiment words that HowNet has tagged，and then the features are selected from an undeterminded document and their orientation is estimated，at last the orientation of the entire document is analyzed.The experimental results show that the accuracy of text orientation analysis is up to 86.7% using the features extracted according to sentiment Ontology compared with the Baseline that using all words as the features.

Key words: Ontology, text orientation analysis, HowNet, vocabulary similarity

摘要： 对文本倾向性分析方法进行了研究，并提出了一种基于情感Ontology的分析方法。首先基于《知网》构建情感Ontology，然后基于情感Ontology抽取文本倾向性分析的特征词汇并判断其情感倾向性，最后根据抽取的特征词汇对整篇文本的倾向性进行分析。实验结果表明，以实验语料中的所有词汇作为特征词汇，在Baseline的基础上，利用情感Ontology抽取特征词汇的文本倾向性分析方法可以使准确率达到86.76%。

关键词: Ontology, 文本倾向性分析, 《知网》, 词汇相似度