基于异构特征的符号社交网络关系分类  

RELATIONS CLASSIFICATION IN SIGNED SOCIAL NETWORKS BASED ON HETEROGENEOUS FEATURE

作　　者：伍杰华[1,2] 朱岸青[1,3] 

机构地区：[1]广东工贸职业技术学院计算机工程系,广东广州510510 [2]华南理工大学信息科学与技术学院,广东广州510641 [3]暨南大学信息科学与技术学院,广东广州510632

出　　处：《计算机应用与软件》2015年第12期42-45,58,共5页Computer Applications and Software

基　　金：广东省科技计划项目(2011B080701082);广东省教育部产学研结合项目(2012B091100043)

摘　　要：符号社交网络关系分类是研究社交关系挖掘领域中一个崭新的研究方向。传统基于同构社交网络的关系分类模型在进行特征提取时,并未考虑符号社交网络中存在异构边(正、负边),提取特征需要代表网络的异构属性这一问题,同时也忽略了异构特征中所蕴含的社交平衡理论。针对以上不足,提出一种新颖的基于异构网络特征的关系分类模型,在特征提取方面主要通过引入朴素贝叶斯模型度量相邻异构关系的影响和结合社会化平衡理论形成的三角关系构建获得,并采用SVM等三类经典的有监督模型进行分类,验证特征的有效性。实验结果表明,改进后异构特征选择算法优化了特征的提取,显著提高了分类效果,从而证明了异构特征提取算法的有效性,为符号社会网络关系特征提取及关系分类提供一种新的思路。Relations classification in signed social networks is a brand new research direction in the field of social relationship mining.When the traditional relationship classification model based on isomorphic social networks extracting the features,it does not consider the issue of the presence of heterogeneous signed social network tie（ positive and negative edges）,and that the feature extraction needs to represent the heterogeneous properties of network,meanwhile it also ignores the social balance theory inherent in heterogeneous feature. For the above shortcomings,we proposed a novel heterogeneous network-based relationship classification model. In feature extraction aspect,it realises mainly by measuring the influence of neighbouring heterogeneous relations through the introduction of naive Bayesian model and integrating the triangular relationship construction formed by the socialised balance theory; it uses three classic supervised models such as SVM for classification,and verifies the effectiveness of features. Experimental results show that the improved heterogeneous feature selection algorithm optimises the feature extraction,and significantly improves the classification results,therefore proves the effectiveness of the heterogeneous feature extraction algorithm,this provides a new way of thinking for the feature extraction of signed social networks and relations classification.

关 键 词：符号网络 社交网络 异构特征 链接分类 关系分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]