Share Email Print
cover

Proceedings Paper

Preliminary results of a Chinese address segmentation algorithm based on self-organizing neural network
Author(s): Duo Gao; Qi Li
Format Member Price Non-Member Price
PDF $14.40 $18.00

Paper Abstract

In this paper we present preliminary results and lessons learned in an effort to design an automatic Chinese address segmentation algorithm using self-organizing neural network (SONN). The SONN design is composed of an input layer and an output layer, fully connected with random initial weights. Unparsed addresses are transformed into activation patterns on the input layer by activating corresponding neurons simultaneously. The training process is controlled by the Hebbian learning rule and the "k winners take all" (KWTA) competing function. After the self-organization process, each unparsed address will be represented by k (multiple) winners with each winner representing a sub-pattern of the address. The goal of the algorithm is to make these sub-patterns correspond to the word segments of the addresses. Two tests were carried out using a prototype system implementing the algorithm. The results show that the current design can achieve the purpose conditionally and we will discuss the limitations of the SONN for actual applications and the possible improvements of the model will be also discussed.

Paper Details

Date Published: 26 July 2007
PDF: 10 pages
Proc. SPIE 6753, Geoinformatics 2007: Geospatial Information Science, 67530U (26 July 2007); doi: 10.1117/12.761374
Show Author Affiliations
Duo Gao, Peking Univ. (China)
Qi Li, Peking Univ. (China)


Published in SPIE Proceedings Vol. 6753:
Geoinformatics 2007: Geospatial Information Science

© SPIE. Terms of Use
Back to Top