摘要
针对目前模糊支持向量机方法中,一般使用样本与类中心之间的距离关系构建隶属度函数的不足,提出一种改进的隶属度确定方法.该方法不仅考虑样本与类中心之间的关系,还考虑样本之间的关系根据样本的类中心与传统支持向量机构造的分类面构建2个超球,由样本点与超球的位置关系计算其隶属度,能够有效地区分样本点、噪音点以及孤立点.通过文本分类实验表明,与其他两种隶属度函数方法相比,基于双超球的模糊支持向量机方法可以更有效地将文本训练集中的噪音剔除,具有较好的分类性能.
Aimed at the defect in the method of fuzzy support vector machine where the membership function was constructed by means of distance relation between the sample and cluster center,an improved method to determine the fuzzy membership was proposed.In this method not only the relationship of the sample and its cluster center was considered but the relationship of all samples also considered.Hence two super-spheres were constructed according to the cluster center of the samples and the classification hyperplane constructed in traditional support vector machine; and the membership was evaluated from the position relation of the sample points and super-spheres. The sample points, noise points, and outliers could effectively be distinguished with the membership determined. The experimental result demonstrated that compared to other two membership function methods, the super-sphere based fuzzy support vector machine method could effectively eliminate the concentrated noise in text training and exhibited better classification performance.
出处
《兰州理工大学学报》
CAS
北大核心
2009年第4期89-93,共5页
Journal of Lanzhou University of Technology
基金
甘肃省高校研究生导师科研基金(0703-07)
关键词
模糊支持向量机
隶属度
文本分类
fuzzy support vector machine
membership
text classification