Chinese word segmentation model using bootstrapping
Baobao Chang, Mairgup Mansur · 2010
We participate in the CIPS-SIGHAN-2010 bake-off task of Chinese word segmentation. Unlike the previous bakeoff series, the purpose of the bakeoff 2010 is to test the crossdomain performance of Chinese segmentation model. This paper summarizes our approach and our bakeoff results. We mainly propose to use χ 2 statistics to increase the OOV recall and use bootstrapping strategy to increase the overall F score. As the results shows, the approach proposed in the paper does help, both of the OOV recall and the overall F score are improved. 1