|Institution:||The Ohio State University|
|Full text PDF:||http://rave.ohiolink.edu/etdc/view?acc_num=osu1461242247|
This dissertation is concerned with the spread of information and influence over dynamic networks. Most prior research focuses primarily on studying spread over static networks; yet, the underlying dynamics of the network often shape the characteristics of the influence or information spread. In this dissertation, we aim to introduce new models and tools that incorporate network dynamics into the study of influence and information spread that arise in different scenarios. In the first scenario, we focus on the spread of content in a mobile wireless network comprising a powerful communication center and a multitude of mobile users. We investigate the propagation of latency-constrained content in the wireless network characterized by heterogeneous (time-varying and user-dependent) wireless channel conditions, heterogeneous user mobility, and hybrid communication capabilities (e.g., directly from the central controller or in a peer-to-peer manner). We show that exploiting double opportunities, i.e., both time-varying channel conditions and mobility, can result in substantial performance gains. In particular, we develop a class of double opportunistic multicast schedulers and prove their optimality in terms of both utility and fairness under heterogeneous channel conditions and user mobility. Simulation results demonstrate that these algorithms can not only substantially boost the throughput of all users, but also achieve different consideration of fairness among individual users and groups of users.The second scenario looks at the speed of information spread carried by a mobile agent. The mobility of the agent is modeled by a heavy-tailed random walk, i.e., a Levy flight. We focus on the first exit time as a measure of the speed of information spread. In particular, we describe the mean and the distribution of the first exit time of a Levy flight from a bounded region in both one- and two-dimensional spaces. We characterize accurate upper and lower bounds on the tail distribution of the first exit time, and provide exact asymptotics of the mean first exit time for a given range of step-length distribution parameters. In the third scenario, we focus on the spread of influence during opinion formation on social networking platforms. Such platforms are major enablers of discussions and formation of opinions on diverse areas including, but not limited to, political discourse, market trends, news and social movements. Often, these opinions are of a competing nature. We study the battle of such competing opinions over dynamically evolving social networks. The novelty of our model is that it captures the exposure and adoption dynamics of opinions that account for the preferential and random nature of exposure as well as the persuasion power of different opinions. We provide a complete characterization of the mean opinion dynamics over time as a function of the initial adoption as well as the particular exposure and adoption dynamics. Our analysis, supported by case studies, reveals the key metrics that govern the spread of… Advisors/Committee Members: Shroff, Ness (Advisor).