论文部分内容阅读
符号社会网络正负关系分类是社会网络分析与挖掘领域的重要研究分支,在朋友关系预测,广告推荐和社团发现等方向具有重要的理论和应用价值。但是现有的分类模型所提取的特征均基于单一的节点属性和同质的链接结构,且依赖于同构网络,具有较大的局限性。针对以上问题,提出了一种新颖的基于异构网络特征的关系分类模型,特征提取主要通过引入隐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基于链接获得,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。对2个大规模符号社会网络的实验表明,本文提出的模型在Precision,Recall,F1-Measure等指标均有较优的分类效果,同时也为异构社会网络关系的特征发现提供一种新的思路。
The classification of positive and negative sign social networks is an important research branch in the field of social network analysis and mining. It has important theoretical and applied value in the fields of friend relationship prediction, advertisement recommendation and community discovery. However, the features extracted from the existing classification models are based on a single node attribute and a homogeneous link structure, and depend on the isomorphism network, which has a great limitation. Aiming at the above problems, a novel relational classification model based on heterogeneous network features is proposed. Feature extraction is mainly based on the introduction of the implicit Bayesian model to measure the influence of adjacent heterogeneous relations and the triangular relationship formed by combining social equilibrium theory Based on the links, three classic supervised models, such as SVM, were used to classify them to verify the validity of the features. Experiments on two large-scale symbolic social networks show that the model proposed in this paper has better classification results on such indicators as Precision, Recall, F1-Measure and the like, and at the same time provides a new type of feature discovery for the relationship of heterogeneous social networks Ideas.