Vairavan Murugappan's Dissertation Defense

Over the past decade, the availability of large and dynamic data sources from diverse domains including business, healthcare, and social media has significantly increased. Most of these data sources consist of rich information about actors’ social roles, behavioral traits, and interactions which are often represented as complex social networks. Many current works focus on designing efficient computational tools (data structures and software libraries) to deal with the large-scale nature of these networks. These social network analysis (SNA) methodologies are essential for extracting meaningful insights from the complex web of relationships and interactions. However, the landscape of existing SNA methodologies is highly fragmented, with solutions often developed in isolated silos making it difficult to adapt, integrate and analyze. Moreover, they frequently lack a systematic way to identify and incorporate the rich contextual information that governs most real-world network dynamics, highlighting the lack of focus on unifying approaches to guide the design of effective and scalable solutions.

To address this foundational gap, in this dissertation we present a novel Context-aware social network analysis framework. This framework provides a holistic approach by focusing on the primary drivers—the varied forms of information available in the data, the different contextual elements shaping the network dynamics, and the critical trade-offs necessary for practical scalability—that connect all phases of SNA. By focusing on these drivers and their interplay, the framework guides the design of computational methodologies that are not only efficient and scalable but also adaptable to the specific characteristics of the data and the network dynamics. This is essential for analyzing and understanding critical real-world phenomena, from information diffusion to public opinion formation. One of the central tasks within these domains is the identification and analysis of key actors, individuals or entities who hold important positions. This problem highlights the importance of context, as the definition of a “key actor” and the methods to find them vary significantly based on the available data and the underlying network dynamics. Therefore, this dissertation focuses on this complex challenge to apply and evaluate the proposed general framework.

Guided by our framework, this dissertation presents three complementary methodologies, each tailored towards different aspects of the key actor analysis problem: an iterative-refinement approach for handling structural dynamism; a modeling and simulation approach for capturing rich socio-cultural processes; and a learning-based approach for predictive analysis. Collectively, this dissertation with the holistic approach and the three methodologies demonstrates that a framework-guided approach yields more efficient, scalable and insightful solutions across a broad range of computational paradigms. This dissertation delivers both a holistic framework and a practical toolkit, which provides a path towards more integrated computational methodologies for large-scale SNA and, in particular, for the critical task of key actor analysis.

Question? Contact Vairavan Murugappan.