The Feature Selection Based On Crfs Model For Chinese Named Entity Recognition In Micro-Blog

Utilizing Natural Language Processing (NLP) for recognizing named entity from micro-blog was a new challenging, emerging research area. However, micro-blog texts lacked of good qualities, and instead a colloquial and mixed language, which bring great difficulties for NER on this kind of material. In this paper, we combined with the characteristics of Chinese character and micro-blog text; comparatively explored the effects of features selection on Conditional Random Fields (CRFs) model. Two categories of features were involved in our system: inner feature and external feature. The experiment results showed that appropriate features could effectively improve F-scores. Especially, the dictionary feature and POS feature made a bigger difference for NER performance.