基于细粒度特征的话题句识别方法1topicclauseidentification.pdf

基于细粒度特征的话题句识别方法1 1,2 1,3 蒋玉茹 宋柔 1 2 （北京工业大学计算机学院北京100022） （北京信息科技大学计算机学院北京100101） 3 （北京语言大学信息科学学院北京 100083） E-mail:yurujiang@126.com,songrou@126.com 摘要：话题句识别中采用穷举方法生成标点句的候选话题句影响系统的执行效率和话题句识别的准确率， 本文利用标点句在篇章中的位置特征、话题的语法特征以及话题串与说明的邻接性特征，指导候选话题句 的生成过程，减少了候选话题句的个数，提高了系统效率，并使得话题句识别的准确率在现有基础上提高 了0.96 个百分点。 关键词：话题；话题句；编辑距离；语法特征；可邻接性； 中图分类号：TP391 文献标识码：A Topic Clause Identification Based On Specific Features Jiang Yuru1,2 Song Rou1,3 （Beijing University of Technology,Computer School, Beijing 100022 ）1 2 （Beijing Information and Science Technology University, Computer School, Beijing 100101 ） 3 （Beijing Language and Culture University,Information Science School, Beijing 100083 ） Abstract: When identifing the topic clause (abbrieved as TC) of punctuation clause (abbrieved as PClause), the brute-force method to generate candidate topic clause (abbrieved as CTC) cause the identification system time-consuming and low accuracy. In this paper, we improve the generating process of the CTC by using specific features such as the PClause location in the text, the grammatical features of the topic and the adjacent features of topic and its comment. The experimental result shows that the improved method can not only improve the efficiency of the system by reducing the number of CTCs, but also make the recognition accuracy of TC increased 0.96 percent over the current state. Key words: