摘要
针对短文本长度短、描述信号弱的特点,提出了一种利用上下位关系的中文短文本分类框架。该框架首先利用"知网"确定训练文本中概念对的上下位关系,进而确定词语对的上下位关系,再将其用于扩展测试文本的特征向量,从而实现对测试文本的分类。实验表明:利用上下位关系能够改善短文本的分类性能。
Concerning the short length and weak signal to describe the characteristics of short text,a framework of Chinese short-text classification was put forward by using hyponymy.In order to achieve the classification of the test text,the framework first used "Hownet" to determine the hyponymy between concept pairs in training text,thus determining the hyponymy between word pairs,and then the feature vectors of test text were extended.The experimental results show that short-text classification performance can be improved by using the hyponymy.
出处
《计算机应用》
CSCD
北大核心
2010年第3期603-606,611,共5页
journal of Computer Applications
基金
国家自然科学基金资助项目(60703010)
教育部回国留学人员启动基金资助项目(教外司留[2007]1108号)
重庆市自然科学基金资助项目(2009BB2079)
关键词
短文本
知网
上下位关系
特征扩展
short-text
Hownet
hyponymy relation
feature extension