社会网络影响力最大化算法及其传播模型研究

发布时间：2018-08-09 16:55

【摘要】：近年来,随着软件与硬件的飞速发展以及个人电脑和互联网的普及,基于熟人关系的网络如微信、基于同学关系的网络如人人网和基于关注关系的网络如微博等各类在线社交平台深受人们的喜爱并占据着人们几乎所有的业余时间,这些平台可以产生海量的数据,给社会网络分析带来了前所未有的机会,因此吸引了大批科研工作者对社会网络空间结构、传播规律等课题的研究和分析。其中,如何选择社会网络里影响力最大化的TOP-K节点及如何挑选社会网络传播模型这两个方向,成为了学术界研究的热门选择。本文首先在前人研究的基础上,对社会网络影响力最大化算法里现有的算法进行了改进;其次,详细分析了独立级联模型和线性阈值模型,并引入人们在第一次接收信息和以后再次接收信息时会有不同反应这一现象以及遗忘规律,提出了一种新型的社会网络传播模型。具体研究内容如下:(1)基于三度影响力原则的线性衰减度中心性算法。根据三度影响力原则,影响力主要在三度分隔以内有效,超过三度分隔,影响力几乎趋近于0。因此线性衰减度中心性以节点在三度分隔以内的潜在影响力来衡量节点的实际影响力,且这种潜在影响力从源节点向外传播到距离为2时影响力衰减到原来的α倍,传播到距离为3时再次衰减β倍,其中0α,β1。计算出线性衰减度中心性之后,本文从3种不同的角度分别在4个公共数据集上验证了算法的有效性。(2)混合式传播模型。真实的人际关系网络里存在着如下的事实:人们在第一次接触某些信息时,是否接受常常取决于信息本身;而在第一次拒绝之后,以后的每一次是否接受取决于以往所拒绝的人和现在推荐的人对其影响力的累积是否大于其自身的阈值,且累积的影响力遵循着遗忘规律会随着时间的推进而不断衰减。混合式传播模型尝试基于这些事实,吸收独立级联模型和线性阈值模型的精华,新提出一种更加符合社会网络影响力传播规律的传播模型,并以两种不同的验证方法在维基百科投票数据集上验证了混合式传播模型的有效性。
[Abstract]:In recent years, with the rapid development of software and hardware and the popularity of personal computers and the Internet, networks based on acquaintance relationships such as WeChat, Various online social platforms, such as Renren based on classmate relationship and Weibo based on concern, are popular and occupy almost all of our spare time. These platforms can generate huge amounts of data. It brings an unprecedented opportunity to the analysis of social network, so it attracts a large number of researchers to study and analyze the spatial structure of social network, the law of communication and so on. Among them, how to choose the TOP-K node with maximum influence in social network and how to select the communication model of social network have become the hot choice of academic research. In this paper, based on the previous studies, the existing algorithms of maximizing the influence of social networks are improved. Secondly, the independent cascade model and the linear threshold model are analyzed in detail. By introducing the phenomenon that people will react differently when they receive information for the first time and then receive the information again, and the law of forgetting, a new social network communication model is proposed. The main contents are as follows: (1) A linear attenuation centrality algorithm based on the three-degree influence principle. According to the principle of three degrees of influence, influence is mainly effective within three degrees of separation, more than three degrees of separation, and the influence is almost close to 0. Therefore, linear attenuation centrality measures the actual influence of nodes by the potential influence of nodes within three degrees, and this potential influence propagates from the source node outward to the distance of 2. The propagation to the distance of 3 again attenuates 尾 times, in which 0 伪, 尾 1. After calculating the centrality of linear attenuation, this paper verifies the validity of the algorithm on four common data sets from three different angles. (2) Hybrid propagation model. In a real network of relationships, there is the fact that when people first come into contact with certain information, it often depends on the information itself, and after the first rejection, The acceptance of each time in the future depends on whether the cumulative influence of the person who has been rejected or recommended now is greater than its own threshold and that the cumulative influence follows the law of forgetting will continue to decline with the advance of time. Based on these facts and absorbing the essence of the independent cascade model and the linear threshold model, a new communication model is proposed, which is more consistent with the law of social network influence transmission. Two different verification methods are used to verify the validity of the hybrid propagation model on the Wikipedia voting data set.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：G206;TP301.6

【参考文献】