Networks provide a natural and powerful way to model complex real-world systems in various domains. Studying structure of a network can help extract functional knowledge about the corresponding system. As real-world networks exhibit non-trivial organization at many scales, this extraction can be done on different levels: from the global perspective of the whole network to the intermediate perspective of node groups (or communities) to the local perspective of individual nodes. With new technological advances, the amount of available real-world network data in different domains rapidly increases. In addition, networks are growing in size and complexity. For example, whereas traditional network data has been static, because it has become easier to record system evolution, more of dynamic network data is becoming available. For these reasons, it is critical to develop novel computational strategies for efficient extraction of functional information from the structure of such complex (e.g., dynamic) networks. And this is the main focus of this dissertation. We achieve this goal in two different ways, by: 1) answering novel research questions via established network approaches, and 2) developing novel network approaches for established research questions.
In the first context, we apply global network analysis to answer a novel question in a novel domain in which network research has not been used to date – interpreting affective physiological data. In addition, we employ local network analysis to study the interplay between individuals’ social interactions and traits from a new dynamic (rather than traditional static) network viewpoint.
In the second context, we take a well-established local analysis approach for static networks to develop a novel method for the problem of link prediction, which we use for de-noising biological networks. Moreover, we take the same static local approach and develop new theory for dynamic network analysis. We demonstrate that accounting for temporal information helps and use our method to study human aging from biological networks. Finally, we introduce a new approach for studying dynamic networks from the intermediate perspective, which deals with the problem of segment community detection. We show that our approach outperforms existing methods in terms of both accuracy and computational complexity.