流形学习算法在中文问题分类中的应用研究  

ON APPLYING MANIFOLD LEARNING ALGORITHM IN CHINESE QUESTION CLASSIFICATION

作　　者：张巍[1,2] 张绚[3] 陈俊杰[2] 

机构地区：[1]山西职工医学院信息中心,山西太原030012 [2]太原理工大学计算机科学与技术学院,山西太原030024 [3]山西省疾病预防控制中心,山西太原030012

出　　处：《计算机应用与软件》2014年第8期269-272,287,共5页Computer Applications and Software

基　　金：国家自然科学基金项目(60970059);山西省科技攻关项目(20110313019);山西省卫生厅科技攻关计划项目(2011073)

摘　　要：针对中文问题分类方法中特征向量维数过高导致处理速度过慢的问题,提出一种基于局部鉴别索引和支持向量聚类的中文问题分类方法。首先利用局部鉴别索引算法对原始高维问句数据集进行降维,将其映射到一个低维空间中,然后通过支持向量聚类算法对问句进行分类。在哈工大社会计算与信息检索研究中心的中文问题集上进行实验,实验结果证明了该方法的有效性,大类准确率87.6%,小类准确率72.5%,取得了较好的效果。Aiming at the problem of jogging speed resulted from too high dimensions of the eigenvector in Chinese question classification,we put forward a Chinese question classification method which is based on locality discriminating indexing（LDI）and support vector clustering（SVC）. First,the LDI algorithm is used to reduce the dimensions of the original high dimensional question dataset and the question dataset is then mapped onto a low dimensional space,subsequently the questions are classified by SVC algorithm. The experiment has been made on the Chinese question set of the research centre of social computing and information retrieval at HIT. Experimental results prove the effectiveness of the method,the accuracy of the coarse classes and the fine classes achieve 87. 6% and 72. 5% respectively. The experiment achieves pretty good results.

关 键 词：问题分类 局部鉴别索引 支持向量聚类 流形学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]