基于条件随机场与规则相结合的中文地名识别

Recognition of Chinese Location Name based on Combination of Conditional Random Fields with Multi-rules

对中文地名识别进行了研究,提出了一种结合多知识的地名识别方法,该方法首先以条件随机场模型为框架,充分利用地名的外部特征和内部颗粒特征,将局部特征、复合特征以及专家知识相融合进行中文地名识别;在此结果上,利用构建的专家规则库对实验结果进行修正。实验结果表明,本文的方法是有效的,实验语料为1998年1月的《人民日报》,开放测试准确率、召回率、和F-值分别达到了93.64%、90.36%、92.03%。

Chinese location name recognition is researched in this paper, and a new approach is proposed to recognize Chinese location name, which combing multi-knowledge. Firstly, the approach makes full use of inner features and exterior features of location name, based conditional random fields model, where, combining local features, hybrid features, related features with expert knowledge to recognize Chinese location name. Then through the analysis of experimental results, a simple rule-base is constructed, which is used to optimize the experimental results. The experimental results show that the precision is 93.64%, the recall is 90. 36% and the F-measure is 92.03% in People＇s Daily （January, 1998）, which prove the validity of this approach.