基于隐含语义索引的文本情感序列回归方法  

Review Text Sentiment Ordinal Regression Based on Latent Semantic Index

作　　者：刘贤友[1,2] 孙丙宇[1,2] 李文波[2] 汪超永[1,2] 

机构地区：[1]中国科学技术大学自动化系,合肥230036 [2]中国科学院合肥智能机械研究所,合肥230031

出　　处：《计算机系统应用》2014年第7期256-259,共4页Computer Systems & Applications

基　　金：国家科技支撑计划(2012BAH89F02)

摘　　要：传统上,文本情感分析技术仅限于情感分类,即仅局限于简单的将评论分为正面或负面两类.而在实际中,有时更需要将评论进行分级,比如把商品划分为"好"、"中"、"差"、"极差"等若干个级别,以便更准确表达评论者的情感;现有的情感分类方法无法解决评论分级问题.为此,提出了基于潜在语义索引的评论文本情感序列回归方法,首先采用潜在语义索引对评论文本进行特征变换,并在此基础上采用核判别学习序列回归方法进行序列回归,实现对评论文本的情感分级.通过在Movie Reviews数据库的实验,验证了提出方法的有效性.Traditionally, text sentiment analysis is only limited to sentiment classification. The review is simply divided into two types： positive and negative comments. In practice, we sometimes need to rank the review, which cannot be solved by traditional sentiment classification methods. To solve this problem, this paper proposes a novel review text sentiment ordinal regression based on Latent Semantic Index. Firstly latent semantic indexing is used to extract features for review texts and then an ordinal regression method is used for review text sentiment analysis. The experimental results on Movie Reviews database proved the effectiveness of the proposed method.

关 键 词：情感分析 序列回归 隐含语义索引 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]