社交活动网络中的影响力最大化方法研究

发布时间：2017-12-31 11:29

本文关键词：社交活动网络中的影响力最大化方法研究　出处：《中国科学技术大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着社交网络的流行,找到一组最具影响力的用户(或节点)以便触发最大的影响范围是很有意义的。例如,公司可以通过提供免费样品/折扣给这些影响力最大的用户,从而借助"口碑"效应来触发更大范围的宣传和购买,最终使该产品在网络中得以大规模普及。这类任务通常被建模为影响力最大化问题,并在过去十多年得到了广泛的研究。但是,考虑到社交网络中的用户可以参加各种各样的在线活动,例如,给产品做评价,加入讨论组等,因此影响力通过在线活动传播的情况变得更加有研究意义。本文主要研究将用户活动考虑在内的影响力最大化问题。包含用户活动的社交网络可以被称为社交活动网络,本文首先将传统的影响力传播模型扩展至其中,从而引出社交活动网络中的影响力最大化问题。基于随机游走理论,本文定义了影响力中心性指标来近似节点集合的影响力,并提出了基于蒙特卡洛的近似算法,实现了对影响力中心性的快速计算。根据子模函数的性质,本文进一步提出了贪心算法和若干优化技术,实现了对影响力最大化问题的快速求解。最后我们利用实际的社交评分数据验证了算法的有效性、效率和通用性。可以看到本文所提出的算法在运行速度上明显优于当前已有的最优算法,且在不同的影响力传播模型与异构网络中都有很好的效果。
[Abstract]:With the popularity of social networks, find a group of the most influential users (or nodes) to trigger the biggest impact is of great significance. For example, companies can offer free samples / discount to these influential users, which means "reputation" effect to trigger a wider range of publicity and purchase, the final to make the product to mass popularity in the network. This task is usually modeled as the influence maximization problem, and has been widely studied in the past more than 10 years. However, considering the social network users can participate in a variety of online activities, for example, to do product evaluation, join the discussion group, thus influence through online communication activities more meaningful research. This paper mainly studies the influence maximization problem of user activities into account. Social networks contain user activity can be called For the social network, the traditional influence propagation model is extended to them, which leads to the influence maximization problem of social activities in the network. Based on the theory of random walk, this paper defines the index nodes to approximate the center of influence of influence, and puts forward the approximate algorithm based on Monte Carlo, realizes the fast calculation of the center of influence the. According to the nature of submodular functions, this paper proposes a greedy algorithm and some optimization methods, to achieve a rapid solution to the influence maximization problem. Finally we use the social score data are verified the effectiveness of the algorithm, efficiency and versatility. You can see the optimal algorithm of the proposed algorithm is better than the current running speed, and has a good effect in the influence of propagation model and different heterogeneous networks.

【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP301.6;TP393.09

【相似文献】