摘要
链接预测是社会网络分析中一个具有挑战性的问题。社会网络中的链接预测问题就是预测社会实体间未被发现的链接和即将演化产生的链接。已有的链接预测算法大多基于社会网络本身的拓扑结构,而忽视社会实体自身的个性化特征。针对以上问题,结合社会实体的个性化特征和社会网络的拓扑特征,提出一种基于概率矩阵分解模型的个性化链接预测算法。该算法整合了社会网络的拓扑特征和实体的个性化信息,建立概率矩阵分解模型,并通过基于梯度的优化算法对模型进行求解。在两个数据集上进行多组实验,一个是数据挖掘领域的合作者网络,另一个是电子商务消费者的信任网络。实验结果证明该算法较现有方法预测准确率有了较大提高。
Link prediction is a challenging task in social network analysis. The problem of link prediction in social network is tantamount to finding out the missing links and inferring the future links on evolution among social entities. Previous studies on link prediction algorithm focus more on the topological structure of social network itself but ignore the personalised features of social entity its own. In light of the problems above,this paper presents a personalised prediction algorithm,which is based on probabilistic matrix factorisation model,in combination with the personalised features of social entities and the topological features of social network. The algorithm integrates the topological features of social network and the personalised information of entities,builds probabilistic matrix factorisation model,and seeks the solution through gradient-based optimisation algorithm. We conduct groups of experiment on 2 real datasets,a co-authorship network in data mining field and a trust network of e-commerce consumers. Experimental results prove that our algorithm has big improvement than current approaches in prediction accuracy.
出处
《计算机应用与软件》
CSCD
2015年第8期243-247,314,共6页
Computer Applications and Software
基金
国家科技支撑计划项目(2012BAH13F02)
上海市科委基金项目(12511502403)
关键词
链接预测
概率模型
矩阵分解
Link prediction Probabilistic model Matrix factorisation