Chinese Proper Names Recognition Based On Dynamic Bayesian Network

The recognition of proper names is one of the basic tasks on Chinese natural language processing research, being not perfectly given the end answer. The automatic recognition of proper names in the text can improve the efficiency of setting up of the large-scale corpus. In addition, proper names recognition give prop to the natural language processing in the field of information extraction, question answering system and so on.We apply the Dynamic Bayesian Networks (DBNs) to the recognizing of person name, organization name and geo-political location appearing frequently in the real text files .It offers an elegant way to integrate local and global information of context into one model. The effort in this paper includes:1. Do hard research in the classifying in proper names, location features in a single sentence, then constitute the local feature varieties and dependency among them.2. Do analysis to the co-reference information in the whole text, theproblem of nominal mentions of proper names is resolved by the reference rules;the discourse information involved into the system improves the recognition accuracy.3. Integrating the baseline system and the DBNS, make use of their merits, assuring the system efficiency.4. We construct a model for location recognition reflecting the relation between proper names and context. Experiments show that our model gives out some merits over other models or methods, because it employs the global clues in text.We tried on the 230 pieces of text files from Shanxi University Corpus for proper names test;extract 180 pieces randomly for training, others for test. We take DBNS to reinforce Nymble, improve accuracy and F value with a bit decrease in recalling.