Research on Early Warning Word Segmentation under Domain Corrector Model

The release of early warning information is an important part of China's 13th Five-year Plan emegency response system, and the correctness verification as well as information analysis of the text are extremely essential. Word segmentation is the first step of Chinese information processing, however, due to the uniqueness of early warning text, the general word segmentation method cannot meet the requirements of accuracy and quickness of segmentation of early warning information. Combined with the domain corrector framework, a weight-based two-word hash dictionary construction method is proposed in the general rough segmentation stage, which effectively reduces times of iterations in the process of the maximum match segmentation algorithm. In the stage of domain correction, the artificial segmentation corpus is constructed based on the existing legal warning text, and the bidirectional LSTM model is used to design the early warning domain corrector, so as to improve the domain adaptation. The comparative experiment proves the superiority of result as well as the speed of word segmentation.