With the rapid development of e-commerce, the influence of user comments on potential cuslomers,merchants and manufacturers has become greater. There are a large number of new reviews emerged on Internet everyday ,so it is difficult to acquire and process all information artificially. Reviews always have strong subjectivity, this paper proposes a whole scenario that helps to solute the acquirement, processing and visualization of reviews information. Through researching on computational methods of semantic similarity between words and sentiment orientation of character, this paper implements sentiment orientation quantization of polarity words and automatic expanding of lexicon. We also propose a method of making simple semantic analysis for sentences using semantic role labeling tool, and design an approach of calculating fine-grained sentiment orientation. The result shows that these methods are more availahle to analysis of sentiment orientation than the method based on POS.%随着电子商务的飞速发展,用户评论信息对潜在顾客、商家和商品生产商的影响越来越大.由于在线的评论信息十分海量,所以很难通过人工浏览方式进行全面获取.评论句子往往具有很强的主观性,本文提出了整体方案帮助解决评论信息的获取、处理和可视化显示.通过利用词语的相似性计算方法和字的情感倾向分布概率计算方法,实现了极性词典的倾向值量化计算和极性词典的自动扩展.通过利用语义角色标注实现对评论句子的浅层语义分析,并利用统计结果设计出计算句子细粒度情感倾向值的方法.实验结果证明,基于语义角色标注方法比基于词性标注方法在句子细粒度情感倾向值计算中更有效.
展开▼
