位置社交网络中地理社交影响力的衡量

发布时间：2019-02-15 20:44

【摘要】：位置社交网络是一种日益兴起的新型社交网络,在位置社交网络中,每个用户不仅可以与其它用户建立好友关系,还可以通过签到分享自己在不同地点所参与的事件。通过挖掘用户的社交关系和地理位置,我们可以衡量事件参与者之间的相互影响力。这种用户影响力的衡量在现实生活中有着广泛的应用场景,比如精准广告投放和病毒式营销等。本文深入分析了位置社交网络用户之间的社交关系邻近性和地理位置邻近性,并提出了一种新颖的地理社交影响力度量。这种度量融合了用户之间的社交影响力和地理影响力：在社交影响力方面,我们采用随机游走模型对网络中信息的传播进行了建模,设计了一种称为处罚命中时间的邻近性权重用于计算用户的社交影响力；在地理影响力方面,我们根据位置社交网络用户的签到规律,采用了幂律分布模型计算空间距离对用户参与事件的影响。基于上述地理社交影响力度量,本文解决了位置社交网络中的两个问题：(1)用户地理社交影响力计算：我们提出了两种近似算法,即全局迭代算法和动态邻域扩张算法,这两种算法具有紧致的理论误差界限,并能够高效衡量单个用户在位置社交网络中的影响力：(2)Top-K影响力事件挖掘：我们提出了一种基于抽样和阈值界定的算法,可以高效地挖掘出位置社交网络中最有影响力的K个事件。在真实位置社交网络数据集及合成数据集上的大量实验表明,我们所提出的算法是有效、高效和可扩展的。
[Abstract]:Local social network (LSN) is a new type of social network, in which each user can not only establish friends with other users, but also share the events they participate in different places by signing in. By mining users' social relationships and geographical locations, we can measure the influence of event participants. This measure of user influence has a wide range of applications in real life, such as precision advertising and viral marketing. In this paper, the proximity and geographical proximity of the users of local social networks are analyzed in depth, and a novel measure of geographical social influence is proposed. This measure combines social and geographic influence between users: in terms of social influence, we use random walk models to model the spread of information in the network. A proximity weight called penalty hit time is designed to calculate the social influence of users. In terms of geographical influence, we use the power-law distribution model to calculate the influence of the space distance on the user's participation in the event according to the check-in rule of the users in the social network. Based on the above measurement of geographical social influence, this paper solves two problems in location social network: (1) user geo-social influence calculation: we propose two approximate algorithms, namely, global iterative algorithm and dynamic neighborhood expansion algorithm. These two algorithms have tight theoretical error limits and can efficiently measure the influence of a single user in a local social network. (2) Top-K impact event mining: we propose an algorithm based on sampling and threshold definition. It can efficiently mine the most influential K events in local social networks. A large number of experiments on real-world social network datasets and composite datasets show that the proposed algorithm is effective, efficient and scalable.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP393.09

【相似文献】