面向产品特征的中文在线评论情感分类:以本体建模为方法

Sentiment Classification for Chinese Online Reviews at Product Feature Level through Domain Ontology Method

在线产品评论反映了用户对产品的体验,对其进行情感分类不仅有利于商家的战略发展,也有助于消费者理性购物。然而,现有研究大多采用上下文无关的情感分类方法,却无法处理褒贬混合的评论及情感词极性随上下文变化的情况,从而导致情感分类的精度不高。针对现有研究的不足,提出一种产品特征级情感分类方法。基于领域本体识别评论中的特征观点对,根据已知极性的评论判断特征观点对的极性。最后,通过加权平均的方法合计评论中各个特征观点对的极性,最终实现对评论的情感分类。为了验证方法的有效性,以手机和数码相机评论为对象设计实验,实验结果表明,本文提出的方法具有一定的领域普适性,能有效识别不同领域评论中的特征观点对,并判断其情感极性。结果还显示,在准确率、召回率和调和评价值3项性能指标上,该方法都优于文档级、句子级和词语级的基线实验方法。

The sentiment classification of product online reviews can facilitate both consumers＇ rational shopping and companies＇ business strategy development, for these reviews reflect consumers＇ feedbacks and contain valuable information. However, most of the literature adopts a context-free sentiment classification method, which is not applicable for reviews mixed with both positive and negative opinions as well as context-sensitive sentiment words, thus resulting in poor performances. We proposes a new sentiment classification method at product feature level based on Chinese online reviews. First, product features and opinions, also known as feature-opinion pairs, are identified with the help of domain ontology. Then the sentiment polarities of feature-opinion pairs are determined with the help of the reviews with given sentiment polarities. The sentiment polarity of unknown review is calculated by the weighted average of the sentiment polarities of features and opinions contained in the reviews. We further conduct several experiments on online reviews of both mobile phone and digital camera to justify theeffectiveness of the proposed approach. Experimental results indicate that the ontology-based method is applicable across heterogeneous product categories for identifying and classifying product features and opinions. The proposed approach significantly outperforms the three baseline methods at document level, sentence level and phrase level, respectively.