基于扩展情感词典及特征加权的情感挖掘方法  

The sentiment mining method based on extended sentiment dictionary and integrated features

作　　者：徐晓丹[1] 段正杰[1] 陈中育[1] 

机构地区：[1]浙江师范大学数理与信息工程学院,浙江金华321004

出　　处：《山东大学学报（工学版）》2014年第6期15-18,69,共5页Journal of Shandong University（Engineering Science）

基　　金：浙江省教育厅科研资助项目(Y201328291);浙江省语委十二五科研规划资助项目(ZY2011C77);国家自然科学基金资助项目(61272007)

摘　　要：针对情感分类中采用单一特征分类精度不高的问题,提出多特征加权的分类算法:根据扩展的情感词典计算每个词的情感倾向度,经CHI特征选择后,根据情感词的极性强度调整贝叶斯分类模型中该词的正负后验概率,在原值的基础上加上极性强度影响值。实验将该方法和其他3种单特征选择方法在酒店、影视等语料上的分类精度进行了对比,分类精度得到提升。实验结果表明,将词语的情感倾向度的特征融入到分类器中方法,在有效提高情感倾向性分类精度的同时降低了特征维数。In the traditional classification method,only one feature was considered,that was not good enough for the precision.In order to improve the precision,a classification method based on integrated features was provided.First, the emotional tendency value of one word was calculated according to an extended sentiment dictionary;then after the CHI selection,the weights of the positive and negative emotion word posterior probability in the Bayesian model were adjusted acrodding to its tendency value.In the experiments,four kinds of corpus such as hotel and movie reviews were used,compared with other three methods,the integrated features method was better.The results showed the precision of classification was improved and the dimension of the feature was reduced.

关 键 词：情感挖掘 分类 倾向性分析 特征选择 情感词典 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]