基于双粒度模型的中文情感特征词提取研究  

Chinese emotional feature extraction base on double granularity model

作　　者：翟东海[1] 杜佳[1] 崔静静[1] 聂洪玉[1] 

机构地区：[1]西南交通大学信息科学与技术学院,成都610031

出　　处：《重庆邮电大学学报（自然科学版）》2014年第3期380-384,共5页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：国家语委"十二五"科研规划项目(YB125-49);教育部科学技术研究重点项目(212167);中央高校基本科研业务费专项资金创新项目(SWJTU2CS096);国家大学生创新创业训练计划项目(201210694017)~~

摘　　要：为了能够快速准确地提取出海量文本信息中的情感特征词,提出从情感词语集中通过人工筛选得到种子词并对其情感强度赋值,同时,以这些种子词为基准计算出情感词语集中其他词语的情感强度值,从而得到各特征词在词语级及句子级的倾向性贡献度值。然后,将特征词在词语级、句子级这2种不同粒度情况下计算出的情感倾向性贡献度值有机结合起来,构造出基于双粒度模型的中文情感特征词提取模型。该提取方法考虑了特征词在词语级和句子级2个方面的情感倾向,使最终提取出的情感词的准确率得到了提高。实验表明,只要有一个全面的情感词典系统和一组准确恰当的种子词,提出的方法可以获得良好的准确率和召回率。In order to quickly and accurately extract the emotional feature words, this paper presents the algorithm. Firstly, seed words are selected from emotional words set by artificial selection and assigned their emotional intensity, and emotional intensity of the rest of emotional words set can be calculated by referencing these seed words. Thus, tendentious contribution degree of each feature words can be gotten in word granularity. Secondly, tendentious contribution degree of each fea- ture words is calculated in sentence granularity. Finally, tendentious contribution degree of feature words, which are calcu- lated in word granularity and sentence granularity, are integrated and constructed Chinese emotional feature extraction. Due to the tendency of emotion feature word in the word level and sentence level, the proposed approach improves the accuracy of the final extract emotional words. As long as there is an emotional dictionary and seed word, the experimental results show the proposed approach can achieved desirable precision and recall rate.

关 键 词：情感分析 情感特征词 倾向性贡献度 情感词语集 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391[自动化与计算机技术—控制科学与工程]