摘要： 在中文产品评论中利用无监督的识别评价对象，准确率和召回率较低。为此，提出一种中文产品评论中的评价对象识别方法。对特殊词、评价对象非完整性、评价对象非稳定性等情况过滤噪声，利用评价对象在评论文本中与评价短语规则出现频率较高的特征，进行置信度排序。实验结果表明，对于14 799篇数码类评论文章，该方法的准确率、召回率和F值分别为0.605、0.780、0.681。
关键词:
无监督,
评价对象,
完整性,
稳定性,
产品评论
Abstract: Aiming at the ineffectiveness of the precision and recall of the unsupervised identification of comment target in the Chinese product reviews, a new method of the comment target recognition is proposed. The integrality and stability of the comment target and their calculation methods are put forward to filter the uncommented words, and the confidence level is scheduled with the consideration of the characteristics of the comment target in comment text, namely, its comment phrase rule and high occurrence frequency. Experimental results show that for the 14 799 digital comments, the precision, recall and F value reach 0.605, 0.780 and 0.681.
Key words:
unsupervised,
evaluation object,
integrality,
stability,
product review
中图分类号:
徐叶强, 朱艳辉, 王文华, 杜锐, 鲁琳, 邓程, 刘洪婧. 中文产品评论中评价对象的识别研究[J]. 计算机工程, 2012, 38(20): 140-143.
XU Xie-Jiang, SHU Yan-Hui, WANG Wen-Hua, DU Dui, LU Lin, DENG Cheng, LIU Hong-Jing. Research on Recognition of Evaluation Object in Chinese Product Review[J]. Computer Engineering, 2012, 38(20): 140-143.