CCRF与规则相结合的中文机构名识别  

Chinese Organization Names Recognition Combined with CCRF and Rules

作　　者：杨晓东[1] 晏立[1] 尤慧丽[1] 

机构地区：[1]江苏大学计算机科学与通信工程学院,江苏镇江212013

出　　处：《计算机工程》2011年第8期169-171,174,共4页Computer Engineering

摘　　要：针对复杂机构名难以识别的问题,提出一种CCRF与规则相结合的识别方法。以CCRF为基础,利用特征融合设计特征模板,融合相应有效规则库,为复杂机构名识别提供决策。对1998年1月的《人民日报》语料库进行开放测试,实验结果显示,机构名识别的准确率为89.92%,召回率为91.41%,F1值为90.66%。Chinese organization names recognition is researched,and a new approach is proposed to recognize Chinese organization names,which combines Cascaded Conditional Random Fields（CCRF） with rules.Adding some effective rules in CCRF model and using effective feature model templates,it helps complicated organization names recognition for making decision for defect of character model.Experimental result shows the precision is 89.92%,recall is 91.41%,F-measure is 90.66% in People’s Daily（January,1998）,which proves the validity of the approach.

关 键 词：机构名 条件随机场 规则库 语料库 识别 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]