【摘要】 随着网络电影数据库和电子商务网站的流行,用户的评论彰显出越来越大的价值。因此,意见挖掘或情感分析是目前自然语言处理和机器学习领域的研究热点之一。情感分类是一个具有代表性的情感分析应用,支撑向量机(Support Victot Machine,SVM)通常被用作为该应用的基准分类方法。首次将线性核极限学习机(线性核ELM)应用于情感分类,并在常用的情感分类和主观/客观分类数据集上,比较了不同的词条加权策略情况下线性核ELM和SVM的分类性能。实验结果显示线性核ELM在大数据集(10000样本)上有着更高的分类准确率率,在较小数据集(2000样本)上和SVM相当。进一步的,我们建立了自己的亚马逊智能手机评论集(Amazon Smartplaone Review,ASR)--由产品评论构成的非平衡数据集(1731正面样本,830负面样本)。比较结果显示线性核ELM在不平衡数据集上也是一个具有竞争力的情感分类方法。更多还原

【Abstract】 With the popularity of Internet movie databases and e-commerce websites,the reviews of users show the growing value.Thus,opinion mining or sentiment analysis is one of the hot research topics in the field of natural language processing(NLP) and machine learning(ML) at present.Sentiment classification is a representative sentiment analysis application and support vector machines(SVM) is usually used as a baseline method.In this paper,linear kernel extreme learning machine(linear kernel ELM) has been applied first to the sentiment classification,and the linear kernel ELM classier is compared with SVM through different term weighting schemes using widely used sentiment and subjectivity/objective datasets.The experimental results show that the linear kernel ELM classification accuracy is higher in large dataset(10000 samples) and it is roughly the same as SVM in small dataset(2000 samples).Furthermore,we build our dataset(Amazon smartphone review,ASR) which is an unbalanced dataset of product reviews(1731 positive samples,830 negative samples).The comparison results show that the linear kernel ELM is also a competitive sentiment classification approach for unbalanced dataset.更多还原