A New Way for Chinese Place Name Recognition

Abstract

According to the characteristics of Chinese place names, we adopt N-gram model to find the Chinese place name candidates in text firstly, which is triggered by the last Chinese character of a place name according to statistical knowledge of these Chinese characters which appear in place names. Then we use maximum entropy model to process these place name candidates, and get the final recognition result, in which we introduce a kind of semantic concept features. The F value of our way is 88.49% in close-test, and 84.19% in open-test. The result also shows that semantic concept features help to improve maximum entropy model performance in Chinese place name recognition. Moreover, we also try to vary the length of training window, and the result shows that variable window can make the result better obviously.