基于HMM的中国组织机构名自动识别

RECOGNITION OF HMM-BASED CHINESE INSTITUTION TERMS

以 2 0 0 0年 11～ 12月《人民日报》网页中涉及社会、经济、法律等内容的 5 0万汉字语料为基础 ,利用基于隐马尔可夫模型并结合概率估值公式来评价在真实文本中构成组织机构名的能力 ,实现了中国组织机构名自动识别系统。实验研究表明准确率达 89%,召回率为 94 .5 %。

An automatic identification system of Chinese institution terms is designed and implemented. This system uses Hidden Markov Model and probability estimation equations to evaluate the ability of forming Chinese institution terms in real text. The experiments on 500000 Chinese characters from the frontpages of People′s Daily from Nov. to Dec. 2000 involving society, economy, law and other contents, show its recall rate 94.5% and precision rate 89%.