Fusion of multiple features for Chinese named entity recognition based on CRF model

Abstract

This paper presents the ability of Conditional Random Field (CRF) combining with multiple features to perform robust and accurate Chinese Named Entity Recognition. We describe the multiple feature templates including local feature templates and global feature templates used to extract multiple features with the help of human knowledge. Besides, we show that human knowledge can reasonably smooth the model and thus the need of training data for CRF might be reduced. From the experimental results on People's Daily corpus, we can conclude that our model is an effective pattern to combine statistical model and human knowledge. And the experiments on another data set also confirm the above conclusion, which shows that our features have consistence on different testing data. © 2008 Springer-Verlag Berlin Heidelberg.