一种基于语义距离的Web评论SVM情感分类方法  

SVM Sentiment Classifier Based on Semantic Distance for Web Comments

作　　者：肖正[1] 刘辉[1] 李兵[1] 

机构地区：[1]湖南大学信息科学与工程学院,长沙410012

出　　处：《计算机科学》2014年第9期248-252,284,共6页Computer Science

基　　金：湖南大学湖南省自然科学基金(13JJ4038);湖南大学"青年教师成长计划"资助

摘　　要：情感倾向分析本质上可以看作是一个情感极性分类问题。在海量数据处理的大背景下,为了提高文本情感判断的准确率,提出了一种结合潜在语义分析LSA(Latent Semantic Analysis)和支持向量机SVM(Supported Vector Machine)的文本褒贬情感倾向分类方法。从语义的角度利用潜在语义分析方法建立"词-文档"的语义距离向量空间模型,然后使用具有良好分类精度和泛化能力的支持向量机进行情感分类。实验结果表明,该方法在句子简短、情感倾向比较明显的Web评论中的准确率较传统的SVM方法有了一定的提高,在测试集上的分类准确率接近88%。The analysis of sentimental orientation can be regarded as a problem of classification on emotional polarity. Under the background of the mass data processing, we proposed a classification approach in terms of sentimental orientation of texts based on LSA（Latent Semantic Analysis） and SVM（Supported Vector Machine）, in order to improve the accuracy of the text emotional judgment. On the concept of semantics,we established a space model of ＂word-document＂ semantic distance vectors by the latent semantic analysis, and then on account of the privileges of accuracy and generalization of support vector machine, designed a SVM classifier with semantic distance as the input feature vectors. Experimental results validate that our method effectively improves the classification accuracy compared with the traditional SVM method. The classification accuracy rate rises to near 88% on the test set of Web comments with short sentences and explicit sentimental orientation.

关 键 词：文本处理 语义距离 情感极性分类 潜在语义分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]