A dual-layer CRFs based method for Chinese nested named entity recognition

While substantial studies have been performed on named entity recognition to date, nested named entity recognition as a research issue has not been well studied, especially for Chinese. In this paper, we take Chinese nested named entity recognition as a cascaded chunking problem on a sequence of words. To approach this problem, we first make a corpus-based investigation of nested structures for Chinese entities and thus propose a dual-layer conditional random fields (CRFs) based solution. To exploit more informative clues for nested named entity recognition, we employ a hybrid chunking scheme to represent the nest structures in Chinese named entities. Moreover, we have also examined the performance of different dual-layer models. Experimental results on different data sets show that the dual-layer CRFs with a hybrid chunk scheme achieve the best performance.