基于进化聚类的动态网络社团发现

发布时间：2018-04-04 07:41

本文选题：进化聚类　切入点：标签传播　出处：《软件学报》2017年07期

【摘要】：社团的数目和时间平滑性的平衡因子一直是基于进化聚类的动态网络社团发现算法的最大的问题.提出一种基于标签的多目标优化的动态网络社团发现算法(LDMGA).借鉴多目标遗传算法思想,将进化聚类思想转换为多目标遗传算法优化问题,在保证当前时刻的聚类质量的同时,又能使当前聚类结果与前一时刻网络结构保持一致.该算法在初始化过程中加入标签传播算法,提高了初始个体的聚类质量.提出基于标签的变异算法,增强了算法的聚类效果和算法的收敛速度.同时,多目标遗传算法和标签算法的结合使算法可扩展性更强,运行时间随着节点或者边数目的增加呈线性增长.将该算法与目前的优秀算法在仿真数据集和真实数据集上进行对比实验,结果表明,该算法既有良好的聚类效果,又有良好的扩展性.
[Abstract]:The balance factor of community number and time smoothness is the biggest problem of dynamic network community discovery algorithm based on evolutionary clustering.A dynamic community discovery algorithm based on label based multi-objective optimization is proposed.Using the idea of multi-objective genetic algorithm for reference, the idea of evolutionary clustering is transformed into an optimization problem of multi-objective genetic algorithm, which can ensure the clustering quality at the present time and at the same time make the current clustering result consistent with the network structure of the previous time.In the process of initialization, tag propagation algorithm is added to the algorithm to improve the clustering quality of the initial individuals.A label-based mutation algorithm is proposed to improve the clustering effect and convergence speed of the algorithm.At the same time, the combination of multi-objective genetic algorithm and label algorithm makes the algorithm more scalable, and the running time increases linearly with the increase of the number of nodes or edges.The results show that the algorithm not only has good clustering effect, but also has good expansibility.
【作者单位】：电子科技大学计算机科学与技术学院;上海交通大学电子信息与电气工程学院;电子科技大学信息与软件工程学院;
【基金】：国家自然科学基金(61300192) 国家科技支撑计划(2013BAH33F02) 中央高校基本科研业务费(ZYGX2014J052) 四川省科技支撑计划(2015GZ0096)~~
【分类号】：TP18;TP311.13

【参考文献】