基于层叠隐马尔可夫模型的中文命名实体识别

Chinese named entity identification using cascaded hidden Markov model

提出了一种基于层叠隐马尔可夫模型的中文命名实体一体化识别方法,旨在将人名识别、地名识别以及机构名识别等命名实体识别融合到一个相对统一的理论模型中。首先在词语粗切分的结果集上采用底层隐马尔可夫模型识别出普通无嵌套的人名、地名和机构名等,然后依次采取高层隐马尔可夫模型识别出嵌套了人名、地名的复杂地名和机构名。在对大规模真实语料库的封闭测试中,人名、地名和机构识别的F-1值分别达到92.55%、94.53%、86.51%。采用该方法的系统ICTCLAS在2003年5月SIGHAN举办的第一届汉语分词大赛中名列前茅。

An approach for Chinese named entity identification using cascaded hidden Markov model, which aimed to incorporate person name, location name, organization name recognition into an integrated theoretical frame was presented. Simple named entity was recognized by lower HMM model after rough segmentation and complex named entity such as person name, location name and organization name was recognized by higher HMM model using role tagging. In the test on large realistic corpus, its F-1 measure of person name, location name and organization name was 92.55%, 94.53% and 86.51%. In the first international word segmentation hakeoff held by SIGHAN （the ACL Special Interest Group on Chinese Language Processing） in 2003. ICTCLAS, which name entity identification base on this model achieved excellent score.