多维相似度特征的社交网络链接分类

发布时间：2018-03-01 06:17

本文关键词： 社交网络多维网络关系分类链接分类特征选择 RReliefF算法　出处：《小型微型计算机系统》2017年06期 　论文类型：期刊论文

【摘要】：深入挖掘社交网络的特征信息对链接的分类具有重要作用.然而已有的特征均基于单维网络构建,忽视了社交网络的多维复杂性及特征之间的冗余信息问题.为解决此问题,提出一种基于RReliefF特征选择模型的多维社交网络链接分类算法.该方法首先在单维相似度特征模型的框架下给出维度关联因子的定义,然后通过改进基于加权传播的的相似度指标,构造多维特征相似度,最后引入RReliefF算法对特征进行选择,有效处理了特征间的冗余信息和噪音信息.在YouTube数据集的实验结果表明,引入的维度关联因子提高了特征的多维度属性,对多维特征信息分析和选择有助于筛选判别性的特征,提高链接分类的质量,验证该方法的有效性.
[Abstract]:Mining the feature information of social network is very important to the classification of links. However, the existing features are based on one-dimensional network, ignoring the multidimensional complexity of social network and the redundant information between features. A multi-dimensional link classification algorithm based on RReliefF feature selection model is proposed. Firstly, the definition of dimension association factor is given under the framework of one-dimensional similarity feature model. Then, by improving the similarity index based on weighted propagation, the multi-dimensional feature similarity is constructed. Finally, RReliefF algorithm is introduced to select the features, which effectively deals with redundant information and noise information among features. The experimental results in the YouTube dataset show that, The introduction of dimension correlation factor improves the multi-dimensional attributes of features. The analysis and selection of multi-dimensional feature information can help to screen discriminant features improve the quality of link classification and verify the effectiveness of the method.
【作者单位】：广东工贸职业技术学院计算机工程系;华南理工大学信息科学与技术学院;
【基金】：广东省优秀青年教师项目(YQ2015177)资助广东省科技计划项目(2011B080701082)资助
【分类号】：O157.5

【相似文献】