Improved deep belief network model and its application in named entity recognition of Chinese electronic medical records

Abstract

The common named entity recognition(NER) methods of Chinese electronic medical records(EMR), are often based on conditional probability of the likelihood of estimation. It is difficult to obtain deeper characteristics between words. In order to obtain deeper level of semantics from Chinese words, this paper purposes an improved deep belief networks (DBN) model by adding the part-of-speech (POS) node, so as to identify the named entities better. This study compares three experiments: Improved DBN model method, original DBN model method and conditional random field (CRF) based method. The result shows that the F1-score of the improved DBN method surpasses both the original DBN and the CRF method's, reaching 91.749%. The experiment shows that the POS node is beneficial to the NER, and the improved DBN model is validated in NER.