A hybrid model for Chinese named entity recognition

Chinese Named Entity Recognition, as a task of providing important semantic information, is a critical first step in information extraction and question answering systems. This paper proposes a hybrid method for NE recognition which combines HMM model and FSVM model. At the bottom level of the system, the person name and simple NEs are recognized by the character-based SVM. At the top level of the system, the complicated NEs are recognized by the word-based SVM. The character-based and word-based SVM are integrated. Adoption of fuzzy SVM helps reduce the impact of noise samples and abnormal data, and improve accuracy of the system.