A Simple and Efficient Model Pruning Method for Conditional Random Fields

Abstract. Conditional random fields (CRFs) have been quite successful in various machine learning tasks. However, as larger and larger data be-come acceptable for the current computational machines, trained CRFs Models for a real application quickly inflate. Recently, researchers often have to use models with tens of millions features. This paper consid-ers pruning an existing CRFs model for storage reduction and decoding speedup. We propose a simple but efficient rank metric for feature group rather than features that previous work usually focus on. A series of ex-periments in two typical labeling tasks, word segmentation and named entity recognition for Chinese, are carried out to check the effectiveness of the proposed method. The results are quite positive and show that CRFs models are highly redundant, even using carefully selected label set and feature templates