A Hybrid Model Based on CRFs for Chinese Named Entity Recognition

This paper presents a hybrid model and the corresponding algorithm combining conditional random fields (CRFs) with statistical methods to improve the performance of CRFs for the task of Chinese named entity recognition (NER). CRFs has a good performance in the task of sequence labeling. In the experiment of recognizing Chinese named entity with CRFs, it can be found that the wrong tags labeled by CRFs are mostly the ones which have lower marginal probabilities. A statistical model is introduced to compliment it. In the hybrid model, marginal probability of every label in CRFs is used to separate CRFs method and statistical method. If the probability is greater than the given threshold, the test sample is recognized by CRFs; otherwise, the statistical model is used. By integrating the advantages of two methods, the hybrid model achieves 93.61% F-measure for Chinese person names and 91.75% F-measure for Chinese location names on MSRA dataset.