神经网络和匹配融合的中文分词研究  

A Hybrid Approach to Chinese Word Segmentation Using Neural Network and an Improved Match Algorithm

作　　者：李华[1] 陈硕[1] 练睿婷[1] 

机构地区：[1]厦门大学福建省仿脑智能系统重点实验室

出　　处：《心智与计算》2010年第2期117-127,共11页Mind and Computation

摘　　要：分词是中文信息处理的基础。本文提出一种使用误差反传神经网络与一种改进的匹配算法相结合的中文分词技术,此方法不需要标注或语义信息,适应性、鲁棒性好,且训练结果占用空间小、有一定冗余性,对比与单纯的神经网络分词方法和匹配的分词方法正确率有了较多的提高。Chinese word segmentation is the basis of Chinese language processing. This paper presents a hybrid approach of BP algorithm and an improved matching algorithm for Chinese word segmentation. This method does not require tagging or semantic information. The adaptability and robustness is very good. And the training results need small space, there is certain redundancy, the matching accuracy has improved significantly compared with the simple method of BP algorithm and matching algorithm.

关 键 词：神经网络 误差反传 中文分词 BP 匹配算法 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]