面向节点演化波动的社会网络事件检测方法

发布时间：2018-08-09 15:10

【摘要】：社会网络特征千差万别,演化规律错综复杂.合理地分析网络演化规律,及时地检测网络事件具有重大意义.基于链路预测的社会网络事件检测方法利用有限的网络拓扑信息,能够有效地发现网络演化的异常波动,准确地检测网络事件.然而,现有方法大多受到链路预测的宏观评价指标的限制,忽略了不同节点演化波动的差异,用相同的相似性计算指标去描述所有节点的演化波动,不利于提升事件检测的表现.为了进一步提升事件检测的精确性和敏感性,提出一种面向节点演化波动的社会网络事件检测方法 Node ED,由节点相似性计算指标判定算法Sim Judge和网络微观演化波动检测算法Micro Fluc组成.主要工作如下:(1)结合粒子群优化算法,提出Sim Judge定量地比较不同的相似性计算指标对节点演化波动的描述程度,确定每个节点在不同时段的最佳相似性计算指标;(2)为了量化事件对网络演化的影响,提出了Micro Fluc,充分考虑节点演化波动的差异,从节点演化波动的角度对不同时段的网络整体演化波动进行定量评估;(3)在真实社会网络VAST和ENRON中进行对比实验,其结果表明,Node ED在VAST中的事件敏感性提升了100%,在ENRON中的事件敏感性提升了50%,更有利于精确地检测社会网络中发生的事件.
[Abstract]:The characteristics of social network vary widely, and the law of evolution is complicated. It is of great significance to analyze the law of network evolution and detect network events in time. Based on link prediction, the method of social network event detection can effectively detect the abnormal fluctuations of network evolution and accurately detect network events by using limited network topology information. However, most of the existing methods are limited by the macro evaluation index of link prediction, ignoring the differences of evolution fluctuations of different nodes, and using the same similarity calculation index to describe the evolution fluctuations of all nodes. It is not good for improving the performance of event detection. To further improve the accuracy and sensitivity of event detection, An event detection method for social network based on node evolution fluctuation (Node) is proposed, which is composed of Sim Judge algorithm and Micro Fluc algorithm. The main work is as follows: (1) combining with particle swarm optimization algorithm, the paper presents a quantitative comparison of the degree of evolution and fluctuation of nodes by Sim Judge. (2) in order to quantify the influence of events on network evolution, Micro lucs are proposed, which take into account the differences of evolution fluctuations of nodes, in order to quantify the influence of events on the evolution of the network, and determine the optimal similarity calculation index of each node at different time periods. From the point of view of node evolution fluctuation, the overall evolution fluctuation of network in different periods is evaluated quantitatively. (3) A comparative experiment is carried out in real social network VAST and ENRON. The results show that the event sensitivity of Node Ed in VAST is increased by 100%, and that in ENRON is increased by 50%, which is more favorable for accurate detection of events in social networks.
【作者单位】：武汉大学计算机学院;
【基金】：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(2015010101010023)~~
【分类号】：O157.5;TP301.6

【相似文献】