江汉学术 ›› 2020, Vol. 39 ›› Issue (1): 53-59.doi: 10.16388/j.cnki.cn42-1843/c.2020.01.006

• 语言学专栏_领域语言研究 • 上一篇    下一篇

基于短语结构文法的评价关系识别研究

何伟1,王玉玲2   

  1. 1中国传媒大学 人文学院,北京 100024;2清华大学 人文学院,北京 100084
  • 收稿日期:2019-08-19 出版日期:2020-02-15 发布日期:2019-12-18
  • 作者简介:何 伟,男,安徽合肥人,中国传媒大学人文学院副研究员,博士;王玉玲,女,山东聊城人,清华大学人文学院博士生。

Research on Evaluation Relationship Identification-based PSG

HE Wei1,WANG Yuling2   

  1. 1School of Humanities,Communication University of China,Beijing 100024;2School of Humanities,Tsinghua University,Beijing 100084
  • Received:2019-08-19 Online:2020-02-15 Published:2019-12-18

摘要: 基于短语结构文法的评价关系识别方法,需首先对评价句进行短语结构句法分析,然后提取出评价词语和候选评价对象,判断评价词语和候选评价对象之间的句法路径是否合法,并将评价词语和评价对象之间的对应关系分为三种情况:第一,评价对象为主语,评价词语为谓语;第二,评价对象为宾语,评价词语为后指动词;第三,评价词语为修饰语,评价对象为中心语,最终识别出评价对象和评价词语之间的评价关系。在京东手机商品的用户评论分析实验中,这种方法的准确率为88.02%,召回率为84.32%,F1值为85.82%。实验结果表明,这种方法是有效的,在评价关系识别中引入相对成熟的短语结构文法能有效提升系统在容错率、评价对象边界识别、避免规则冲突、远距离搭配识别这些方面的性能和优势。

关键词: 评价关系, 短语结构文法, 评价分析, 中文信息处理, 计算语言学

Abstract: The evaluation relationship based on phrase structure grammar(PSG) is identified in three steps:first syntactically analyze the sentence to be evaluated;then extract evaluation expression and the evaluation targeted candidate,and judge whether the syntactic path between the expression and the candidate is grammatical;finally divide the corresponding relationship into three groups:1. candidatesubject,expression-predicate;2. candidate-object,expression-posterior verb;3. expression-modifier,candidate-central statement. In the test analyzing customers’comment on JD’s cell phones,the accuracy rate of this method is 88.02%,recall rate 84.32 % ,and the F1 value 55.82%. The result proves this method is effective. The introduction of the comparatively mature phrase structure grammar in the identification of evaluation relationship can effectively improve the system’s performance and enhance its advantage in fault tolerance,boundary identification of candidate,avoiding rule conflict,and recognizing distant pairing.

Key words: evaluation relationship, PSG, evaluation analysis, Chinese information processing, computational linguistics

中图分类号: