###
Journal of Software:2016.27(11):2747-2762

混合指标量子群智能社会网络事件检测方法
胡文斌,王欢,严丽平,邱振宇,肖雷,杜博
(武汉大学 计算机学院, 湖北 武汉 430072)
Hybrid Quantum Swarm Intelligence Indexing for Event Detection in Social Networks
HU Wen-Bin,WANG Huan,YAN Li-Ping,QIU Zhen-Yu,XIAO Lei,DU Bo
(Computer School, Wuhan University, Wuhan 430072, China)
Abstract
Chart / table
Reference
Similar Articles
Article :Browse 904   Download 1198
Received:June 03, 2015    Revised:August 11, 2015
> 中文摘要: 社会网络错综复杂,如果能够及时发现和预测当前网络可能发生的重大事件并采取有效的处置策略,将具有重大意义.链路预测的理论框架和评价方法为社会网络事件检测提供了一条有效途径.目前,链路预测的研究工作大多针对特定网络提出相似性指标,试图取得更高的链路预测精度.这些研究存在如下问题:(1)不同的相似性指标适用于不同的网络,不具有普适性;(2)独立的相似性指标无法全面反映网络演化的多样性和复杂性;(3)链路预测时未考虑网络演化过程中可能出现波动,无法进行事件检测.基于上述问题,提出一种社会网络事件检测的混合指标群智能方法IndexEvent,由最佳权重算法OWA(optimal weight algorithm)和波动检测算法FDA(fluctuationdetection algorithm)组成,可以评价不同网络的演化波动,发现网络波动异常,进行事件检测.主要工作如下:(1)提出了混合指标,并证明了基于混合指标的链路预测算法可以取得更高的预测精度;(2)基于量子粒子群算法提出了最佳权重算法OWA,以高效地确定不同网络的最佳混合指标;(3)提出了一种网络波动检测算法FDA,定量评价不同时段网络演化的波动程度,并在考虑微观因素的基础上进行改进.对不同特征的网络进行实验,结果表明,IndexEvent方法能够准确地反映事件造成的网络演化波动,有效地检测事件.
Abstract:In complicated social networks, discovering or predicting important events is significant. The theoretical framework and evaluation methods of link prediction offer an effective solution for detecting events in social networks. Most of the current research focuses on proposing different similarity indexes to achieve higherlink prediction accuracy. However this type of approach has following problems:(1) Because different similarity indexes are designed for different networks, they are not universal; (2) The independent similarity index is difficult to reflect diversity and complexity of real network evolutions; (3) Without considering the fluctuation in the network evolution, the link prediction cannot detect events. To solve these problems, this paper proposes a swarm intelligence method based on mixed indexes (IndexEvent), which can evaluate fluctuations and detect events in social networks. The main work is as follow:(1) A proof is provided on the proposed mixed indexes that the link prediction algorithm based on mixed indexes can achieve a higher accuracy; (2) Based on the quantum-behaved particle swarm algorithm, an optimal weight algorithm (OWA) is developed to determine best mixed indexes for different networks efficiently; (3) A fluctuation detection algorithm (FDA) is designed to quantitatively estimates fluctuations in network evolutions at different periods. And micro factors are taken into account to improve FDA. The results of the experiments show that IndexEvent can effectively reflect evolution fluctuations and detect events.
文章编号:     中图分类号:    文献标志码:
基金项目:国家重点基础研究发展计划(973)(2012CB719905);国家自然科学基金(61572369,61471274);湖北省自然科学基金(2015CFB423);武汉市重大科技计划项目(2015010101010023) 国家重点基础研究发展计划(973)(2012CB719905);国家自然科学基金(61572369,61471274);湖北省自然科学基金(2015CFB423);武汉市重大科技计划项目(2015010101010023)
Foundation items:National Program on Key Basic Research Project of China (973) (2012CB719905); National Natural Science Foundation of China (61572369, 61471274); National Natural Science Foundation of Hubei Province (2015CFB423); Wuhan Major Science and Technology Program (2015010101010023)
Reference text:

胡文斌,王欢,严丽平,邱振宇,肖雷,杜博.混合指标量子群智能社会网络事件检测方法.软件学报,2016,27(11):2747-2762

HU Wen-Bin,WANG Huan,YAN Li-Ping,QIU Zhen-Yu,XIAO Lei,DU Bo.Hybrid Quantum Swarm Intelligence Indexing for Event Detection in Social Networks.Journal of Software,2016,27(11):2747-2762