基于混合方法的中文人名识别研究.pdf

Computer Engineering and Applications 计算机工程与应用 2015 ，51（8） 211 基于混合方法的中文人名识别研究 王祖兴，吕 钊，顾君忠 WANG Zuxing, LV Zhao, GU Junzhong 华东师范大学 计算机科学技术系，上海 200241 Department of Computer Science and Technology, East China Normal University, Shanghai 200241, China WANG Zuxing, LV Zhao, GU Junzhong. Hybrid method for Chinese person name recognition. Computer Engi- neering and Applications, 2015, 51 （8 ）：211-217. Abstract ：Most of existing researches mainly focus on recognizing the names of Chinese person while seldom specializing in recognizing Japanese and transliterated person names. This paper proposes a method based on CRF and combines per- son name reliability model and contextual rules （simply, CRRM ）to recognize the person names in Chinese sentences. Partial frequency statistical algorithm is also used to revise the misrecognized boundary of names and proliferation opera- tion is used to recall those unrecognized names with the already recognized one. Experiments based on a true dataset show that this approach is efficient in recognizing the person names from Chinese texts. The F -value for recognition of Chinese person name, Japanese name and transliterated person name is higher than 90%. Key words ：Chinese person name recognition; Conditional Random Fields （CRF ）model; person name reliability model; contextual rules; marginal probability 摘 要：当前中文人名识别的研究主要针对中国人名，而对日本人名及音译人名的专门研究相对较少，识别效果也 亟待提高。提出利用CRRM 方法进行中、日及音译人名同步识别。该方法基于CRF （Conditional Random Fields ）并 结合了上下文规则及人名可信度模型。此外，利用局部统计算法对边界识别错误的人名进行修正，并利用扩散操作 召回未被识别的人名。实验结果表明，中、日、音译人名识别的F 值均高于90% ，提出的方法可以取得较好的识别效果。 关键词：中文人名识别；条件随机域（CRF ）模型；人名可信度模型；上下文规则；边缘概率 文献标志码：A 中图分类号：TP393 doi ：10.3778/j.issn. 1002-8331.1308-0180 1 引言 的效果。