摘要
针对社区发现中,部分节点划分难的问题,探讨重叠社区检测的优化模型和求解的视觉进化神经网络。模型通过设计节点隶属度矩阵和节点分割规则,建立以模糊分割阈值为变量,且能评估社区划分效果的改进型模块度函数;算法设计中,以候选解构成的状态矩阵对应函数值矩阵作为输入,依据果蝇视觉系统的信息处理机制,建立以输出作为状态学习率的果蝇视觉前馈神经网络,进而借助灰狼优化的位置更新规则,设计状态更新策略,获得基于重叠社区检测的果蝇视觉进化神经网络及其算法。该神经网络的计算复杂度,由状态矩阵的大小及社区网络的节点数确定。比较性的数值实验显示,该求解重叠社区检测问题具有明显优势,有较好的应用潜力。
Aiming at the difficulty of division of some nodes in community discovery,this work probes into both the overlapping community detection optimization model and the related fly visual evolutionary neural network.In the design of the model,an improved modularity function,which takes fuzzy segmentation thresholds as variables,is proposed to evaluate the divisional effect of community division,relying upon a membership matrix and a segmentation rule.In the design of the algorithm,a fly visual evolutionary neural network(FVENN)is developed to solve the model.Therein,the input is a function-valued matrix matched with a state or candidate matrix,and meanwhile an improved fly visual feedforward neural network is designed to output the learning rate of each element in the state matrix by means of the information processing mechanism of the fly’s visual system.Hereafter,each state is updated based on its learning rate and a position update rule of grey wolf optimization.FVENN’s computational complexity is determined by its input size and the number of nodes in the community network.Comparative experiments validate that FVENN outperforms the compared approaches and is of great potential to solving the problem of overlapping community detection.
作者
罗兰
张著洪
LUO Lan;ZHANG Zhuhong(College of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China)
出处
《智能计算机与应用》
2022年第2期83-90,共8页
Intelligent Computer and Applications
基金
国家自然科学基金(62063002,61563009)
关键词
模糊聚类
重叠社区检测
果蝇视觉神经网络
灰狼优化
状态更新
fuzzy clustering
overlapping community detection
fly visual neural network
grey wolf optimization
state update
作者简介
罗兰(1996-),女,硕士研究生,主要研究方向:智能优化;通讯作者:张著洪(1966-),男,博士,教授,博士生导师,主要研究方向:数据科学与计算智能、深度学习等。Email:zhzhang@gzu.edu.cn
