多维相似度特征的社交网络链接分类

Multidimensional Similarity Feature Link Classification in Social Networks

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.

Mining social network feature information plays an important role on link classification. However,existing features are based on single-dimensional social network, ignoring their multi-dimensional complexity and redundant information between features. To solve this problem,this paper presents a RReliefF feature selection based multi-relational link classification algorithm. Firstly, in the framework of similarity methods in single-dimensional model, we constructed a multidimensional similarity metric by incorporating the dimensional correlation factor and weighted cascading similarity index, then introduced RReliefF feature selection algorithm to effec- tively deal with the redundant information and noise information between such features. The experimental results on YouTube datasets show that the introduced dimensional correlation factor indeed improve the characteristics of the multi-dimensional attributes. And the feature selection process is helpful to determine the discrimitive multi-dimensional characteristics, thereby enhancing the quality of the link classification.