Based Hybrid Method for Chinese Location Recognition

This paper proposed a hybrid method of the Chinese location recognition which combines conditional random fields model and pattern-selection. The conditional random fields model is used based on statistic method, some interesting features have been proposed, the new probabilistic feature is proposed, which are used instead of binary feature functions, however, it is one of the several differences between this model and the most of the previous CRFs-based model, we also explore several new features in our model, which includes semantic information, local features, global features, diffusion feature, related features etc. The pattern selection is used for revising experimental results. Combined the rule-based and statistic-based, we evaluate this approach on large-scale corpus with open test method using People´s Daily (January, 1998), the evaluation results show that our approach based on hybrid method significantly outperforms previous approaches.