Finite state machine based place name address component recognition

Abstract

In the process of Chinese address information processing, the identification accuracy of address components directly affects the accuracy of matching, and plays an indispensable role in people's life. In this paper, the finite-state machine (FSM) model is used to enter the finite-state machine with the address elements as input, and the CRF++ tool is used to train the CRF annotation model among the states of the address components annotated by the word, and the finite-state machine transformation function is constructed. After further disambiguating by state verification function, this paper compares the address recognition results of finite state machine model with those of statistical model tools such as cascade conditional random field. The results show that the finite-state machine address component recognition model with verification function has higher accuracy and better and more comprehensive understanding of address diversity.