【作者】 焦超；

【导师】 刘功申；

【作者基本信息】 上海交通大学 ， 通信与信息系统， 2012， 硕士

【摘要】 近年来,网络突发事件数量急剧增加,如果能够在突发事件爆发的初期就能对事件的规模进行预测,并且对事件形成的网络推动力量有一个明确的认识,掌握其是否有人幕后推动,将非常有利于更快速对事件做出更合理应对。本文的主要研究工作有:(1)提出了检测突发事件幕后推手的方法。综合事件潜伏期热度、ID集中度、简单文章比例、新注册ID比例和作者地域集中度等方面的特征,计算生成一个综合指标,表示事件幕后推手存在的可能性以及推动力量的大小。(2)总结了网络突发事件的热度分布规律。提出了浏览-回复模型,并根据模型指出自然爆发的网络突发事件满足多次泊松分布的叠加。人为推动事件如果推动力量很小,去除推动因素后仍近似满足泊松分布叠加规律。(3)提出预测模型和指标。在分析突发事件分布规律的基础上,提出了基于曲线拟合的热度预测模型,可以在事件爆发初期,大致预测事件的总热度。并定义了针对不同需求的预测指标。本文各部分内容依赖性比较强,因此将实验与理论部分结合在一起,以便理解。本文对比“贾君鹏事件”和“李刚事件”,验证了幕后推手检测方法的有效性;选取多个网络突发事件进行分析,验证了泊松分布规律。对于绝大部分事件,预测算法均能有良好的预测效果。更多还原

【Abstract】 In recent years, network emergencies increased rapidly. Earlier prediction of scale and mastery of driving forces behind network emergency would largely benefit our response. Our major research focused on the following aspects.(1) Proposed a driving force detection method. We synthesized the following features: the latency heat, ID concentration, proportion of simple articles, ratios of newly registered IDs and geographical concentration of authors and calculated a composite index indicating the possibility and intensity of driving force.(2) Summarized the distribution of network emergency. Set up a“Browse-Reply”model to describe the network emergency, and found that heat distribution of natural emergency fitted combination of Poisson distribution functions. Network emergency with weak driving force still fitted Poisson distribution after removing the driving parts.(3) Proposed a prediction model and indexes. We proposed a prediction method based on heat distribution information. The method can predict the entire heat of the network emergency on early outbreak stage. Then we defined different indexes according to different requirements of network users and network regulators.For dependency of each part, we didn’t put all the experiments at the end of the thesis. Instead, we put them separately on the end of each part to make it more comprehensible. We compared“Jia Junpeng event”and“Li Gang event”to verify the effectiveness of driving force detection method. We analyzed lots of network emergencies, verified that most of the emergencies fitted Poisson distribution and the prediction result was effective. Most of measurement errors were less than 50%.更多还原