小粒度策略下基于CRFs的军事命名实体识别方法

A Military Named Entity Recognition Method Based on CRFs with Small Granularity Strategy

军事命名实体(Military Named Entities,MNEs)内部嵌套关系复杂、语法区分不明显,从而影响实体识别效果,针对这一问题,提出了一种小粒度策略下基于条件随机场(Conditional Random Fields,CRFs)的MNEs识别方法。运用小粒度策略,结合手工构建的MNEs标注语料进行建模,采用CRFs模型识别出不可再分的小粒度MNEs,再通过对小粒度MNEs进行组合得到完整的MNEs。最后,通过实验对该方法进行了验证,结果表明:在作战文书语料的开放测试中,MNEs识别的召回率达到72%以上,准确率达到85%以上。

The recognition of Military Named Entities（ MNEs） is restrained by the complex nested relation of MNEs and obscure grammatical distinction. To resolve this problem,the authors put forward MNEs recognition method based on Conditional Random Fields（ CRFs） model with small granularity strategy. The authors construct a marked corpus to train the model,and use the model to recognize small granularity MNEs which can＇t be divided,then get the complete MNEs by composing small granularity MNEs. Finally,the method is verified by the experiment,the results show that the recall rate and the precise rate of MNEs recognition is 72% and 85% respectively in the open test of operational document corpus.