中文摘要：

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NodeED，由节点相似性计算指标判定算法SimJudge和网络微观演化波动检测算法MicroFluc组成．主要工作如下：（1）结合粒子群优化算法，提出SimJudge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MicroFluc，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，NodeED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．

英文摘要：

The social network is complicated with different evolution mechanisms. It is of great significance to reasonably analyze social network evolutions and effectively detect social events. The event detection methods based on link prediction make most of the limited network topological information, discover the network evolution fluctuation, and detect events. However, most of existing methods are limited by the assessment measures of link prediction, and neglect the otherness of micro node evolution mechanisms. They use the same similarity index to describe evolution fluctuations of different nodes, which is adverse to the performance of event detection. To improve the accuracy and sensitivity of event detection, this paper proposes an event detection method based on node evolution fluctuation for social networks （NodeED）. The method consists of a node similarity index judgement algorithm （SimJudge） and a micro evolution fluctuation detection algorithm （MicroFluc）. The main work is as follow： （1） Based on the particle swarm algorithm, SimJudge is proposed to compare the description performances of different similarity indexes for a node evolution fluctuation. Different nodes can find their optimal similarity indexes at different periods by SimJudge; （2） To quantify the effect of events, MicroFluc is proposed to consider the diversity of node evolution fluctuations and evaluate the entire network evolution fluctuation; （3） In real social networks VAST and ENRON, NodeED results in the event detection sensibility increase by 100% in VAST and 50% in ENRON, which shows NodeED has more advantages to detect events in social networks than other methods.