基于条件随机场模型的中文地址分词研究

Research of Chinese Address Segmentation Based onConditional Random Field Model

地址分词是实现地理编码的重要基础。本文基于条件随机场模型对中文地址分词进行研究,实现了中文地址的快速、准确分词。文中先综合分析各种中文地址分词方法的优势和劣势,设计出一套优化的地址分词标注体系,然后制定地址分词的尾词特征和特征模板,采用半监督学习与人工标注相结合的方式获取高质量的标注训练语料,供CRF模型训练;最终根据已标注语料训练条件随机场模型,实现对中文地址信息的自动分词和识别。

Address segmentation is an important basis for geocoding.Based on conditional random field model,Chinese address word segmentation is studied in this research,and the fast and accurate segmentation of Chinese address is realized.Firstly,This paper analyzes the advantages and disadvantages of various methods for Chinese address segmentation,designs a set of optimized address segmentation tagging system,and then formulates the tail word features and feature templates of address segmentation,and obtains high-quality tagging training materials by combining semi-supervised learning with manual tagging for CRF model training.Finally,according to the training condition random field model of labeled training materials,this paper achieves the goal of automatic segmentation and recognition of Chinese address information.