Enhancement of unsupervised feature selection for conditional random fields learning in Chinese word segmentation

This work proposed a unified view of several unsupervised feature selection based on frequent strings that improve conditional random fields (CRF) model for Chinese word segmentation (CWS). These features include character-based n-gram (CNG), accessor variety based string (AVS), term-contributed frequency (TCF), and term-contributed boundary (TCB), with a specific manner of boundary overlapping. For the experiment, the baseline is the 6-tag, a state-of-the-art labeling scheme of CRF-based CWS; and the data set is acquired from SIGHAN CWS bakeoff 2005 and SIGHAN CWS 2010. The experiment results show that all of those features improve the performance of the baseline system in terms of recall, precision, and their harmonic average as F₁ measure score, on both accuracy (F) and out-of-vocabulary recognition (F_OOV). In particular, this work presents a novel feature selection approach of the compound feature “AVS+TCB” that outperforms other types of features for CRF-based CSW in terms of F and F_OOV.