Named entity recognition with multiple segment representations

摘要: Named entity recognition (NER) is mostly formalized as a sequence labeling problem in which segments of named entities are represented by label sequences. Although considerable effort has been made to investigate sophisticated features that encode textual characteristics (e.g. PEOPLE, LOCATION, etc.), little attention paid segment representations (SRs) for multi-token the IOB2 notation). In this paper, we effects different SRs on NER tasks, and propose feature generation method using multiple SRs. The proposed allows model exploit not only highly discriminative complex but also robust simple against data sparseness problem. Since it incorporates functions Conditional Random Fields (CRFs), can use well-established procedure training. addition, tagging speed integrating be accelerated equivalent most SR integrated model. Experimental results demonstrate incorporating into single improves performance stability NER. We provide detailed analysis results.