Domain adaptive Chinese Word Segmentation based on domain knowledge and word-formation feature

This paper describes a novel method about domain adaptive Chinese Word Segmentation. Unlike traditional methods, our system takes advantage of the domain knowledge and word-formation feature. First, we construct a general knowledge by bootstrapping, which contains domain independent information. Then a cross-domain model is generated with general knowledge and cross-domain knowledge. Furthermore, we realize the importance of word-formation in word segmentation. The segmentation results will be revised with word-formation strategy. There is scarce study about word-formation and this method indeed plays a significant role. We test our system on the corpora given by CIPS-SIGHAN 2010, and our system achieves F score of above 0.94 in all four domains. The good performance proves the effectiveness of our approach.