基于情感词汇与机器学习的方面级情感分类

Aspect level sentiment classification based on sentiment words and machine learning

为综合利用基于情感词典和基于机器学习的两类情感分类方法的优点,提出一种基于情感词汇与机器学习的方面级情感分类方法。通过选取少量情感倾向与评价对象无关的情感词汇对评价搭配进行情感分类;通过构建机器学习分类器,以评价短语对各类别的互信息占比作为分类器的分类概率权重,进行加权计算,选择加权后分类概率最大的类别作为评价搭配的情感倾向类别。在中文评论数据集上的实验结果表明,该方法能有效提高情感分类性能。

To comprehensively utilize the advantages of sentiment classification methods based on sentiment lexicon and that based on machine learning,an aspect-level sentiment classification method based on sentiment words and machine learning was proposed.The evaluation collocation was classified by selecting a small number of words whose sentiment was unrelated to the target.A machine learning classifier was constructed and the proportion of mutual information of evaluation phrases was used as the weight of classification probability of the classifier.The category with the highest classification probability after weighting calculation was selected as the category of the evaluation collocation.Experimental results on the Chinese review dataset show that the proposed method can effectively improve the sentiment classification performance.