情感词扩展对微博情感分类性能影响的实验分析

Effect of Sentimental Word Expansion on the Performance of Microblog Sentimental Classification Task

微博情感分类是典型的情感分析任务之一,而情感词是很多情感分析方法的基础.由于手工情感词典的局限性,情感词的自动扩展经常作为情感分析的重要步骤,而情感词扩展方法的好坏也经常用情感分类等任务来间接评测.在中、英文两个语种的微博数据集上进行对比实验,详细地分析了通过典型的情感词扩展方法抽取的新情感词对微博主观性分类和倾向性分类的影响.实验中对比了中、英文两种语种、不同的情感词扩展方法、不同的情感强度计算方法、不同的微博情感分类方法、不同的候选情感词词性、不同的种子情感词典、以及不同的微博情感分类测试集,透过多个视角,观察和分析情感词扩展在微博情感分类中作用,为相关研究工作提供参照或证据.

Microblog sentimental classification is a typical task of opinion analysis, where the sentimental words play as a crux in this task. The limitation of manually constructed sentimental dictionaries is so obvious that automatically expanding sentimental words from some seeds often employed by most opinion analysis methods. Correspondingly, opinion analysis such as sentimental classification is often employed to evaluate the performance of sentimental word expanding. In this paper, we discuss the effect of sentimental word ex- panding on microblog sentimental classification based on some subtly designed experiments on bilingual microblog dataset of Chinese and English. To provide more information and evidence, we observe the effect from multiple perspectives such as subjectivity classification, polarity classification, different languages, different sentimental expanding methods, different sentimental weighting methods, different sentimental classification method, and different part of speeches of candidate sentimental words, different original seeds, and different classification test data.