基于链接模型的主动半监督社区发现方法

发布时间：2018-04-04 13:34

本文选题：半监督社区发现　切入点：主动学习　出处：《计算机应用》2017年11期

【摘要】：链接模型可对网络的社区发现问题建模,相比具有相同目标的对称模型和条件模型,PPL模型处理网络类型更多、社区发现准确率更高。但PPL模型是一个无监督模型,在网络社区结构不清晰时效果不佳,且不能利用易获取的先验信息。为使用尽可能少的先验,获得社区发现链接模型性能较大的提升,提出了一个主动节点先验学习(ANPL)算法,该算法主动选择效用高、易标记的成对约束进行标记,基于标记的约束对自动生成信息量更大的标记节点集合。基于PPL模型设计了一个融合网络拓扑结构和标记节点先验的半监督社区发现(SPPL)模型,并给出模型用于半监督社区发现的参数估计算法。人工网络和实际网络上的实验结果表明,利用ANPL获得的标记节点先验和网络拓扑结构,SPPL模型的社区发现准确率高于无监督PPL模型及当前流行的基于非负矩阵分解(NMF)的半监督社区发现模型。
[Abstract]:The link model can model the community discovery problem in the network. Compared with the symmetric model and the conditional PPL model with the same objective, the PPL model can deal with more network types and the accuracy of community discovery is higher.However, the PPL model is an unsupervised model, which is not effective when the structure of the network community is not clear, and can not use the prior information which is easy to obtain.In order to improve the performance of community discovery link model with as few priori as possible, an active node priori learning algorithm is proposed. The algorithm has high active selection utility and is easy to label in pairs of constraints.Tag-based constraints automatically generate a set of tag nodes with more information.Based on the PPL model, a semi-supervised community discovery model is designed, which combines the topology structure of the network and a priori of the labeled nodes, and a parameter estimation algorithm for semi-supervised community discovery is presented.The experimental results on artificial and real networks show that,The community discovery accuracy of the ANPL model is higher than that of the unsupervised PPL model and the current semi-supervised community discovery model based on the nonnegative matrix factorization (NMF).
【作者单位】：河北地质大学信息工程学院;河北中医学院公共课教学部;
【基金】：国家自然科学基金资助项目(61503260)~~
【分类号】：O157.5;TP181

【相似文献】