SVM和基于转换的错误驱动学习相结合的汉语组块识别  

SVM-Based Chinese Chunk Recognition and Transformation-Based Error-Driven Learning

作　　者：邹宏梅[1] 王挺[1] 

机构地区：[1]国防科技大学计算机学院,湖南长沙410073

出　　处：《计算机工程与科学》2007年第4期91-94,123,共5页Computer Engineering & Science

基　　金：国家自然科学基金资助项目(60403050)

摘　　要：本文研究了一种支持向量机(SVM)和基于转换的错误驱动学习相结合的汉语组块识别方法。SVM在选取特征方面有突出的优点,并且在高维特征空间也具有较高的泛化性能,通过核函数的原则,SVM能够在独立于训练数据维数的小计算范围内进行训练。利用基于转换的错误驱动学习方法对SVM的标注结果进行校正,转换规则较好地处理了语言现象中的特殊情况,进一步提高了SVM的识别结果。实验结果表明,该方法具有较好的效果。The paper presents a method of Chinese chunk recognition based on Support Vector Machines（SVM） and transformation-based error-driven learning. It is well known that SVM is good at selecting useful features and achieving high generalization of very high dimensional feature space. Furthermore, by introducing the kernel function, SVM can be trained in a high dimensional space with smaller computational cost independent of their dimensionalities. The transformation-based learning approach with supervision is further applied to improving the analysis results of SVM. Transformation rules effectively deal with the specificity of Chinese and improve the performance of SVM. The experiments show encouraging results.

关 键 词：支持向量机 基于转换的错误驱动学习 汉语组块识别 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]