Heterogeneous graphs versus multimodal content: modeling, mining, and analysis of social network data

Complex networks arise everywhere. Online social networks are a famous example of complex networks due to: (a) revolutionizing the way people interact on the Web, and (b) permitting in practice the study of interdisciplinary theories that arise from human activities, at both micro (i.e., individual) and macro (i.e., community) level. The vast scale (Big-Data) of online human interactions impose certain challenges, such as scalable indexing and efficient retrieval of social data, which are by their nature intertwined in multiple dimensions. Understanding the rich properties and dimensional interdependencies of topology and content in complex networks is necessary to uncover hidden structures and emergent knowledge. To address these questions, we propose a formal model that abstracts the semantics of complex networks into an integrated, context-aware, time-sensitive, multi-dimensional space, enabling joint examination of their static and dynamic properties, facilitating unified mining and analysis of network structure and content, and their explicit and implicit interactions. Traditionally, network analysis methods, either ignore content and focus on the network structure, or make implicit assumptions about the complex correlation of these two components. We show that accurately modeling multiple symmetric or asymmetric, explicit and hidden interaction channels between people, integrating auxiliary networks into a unified framework, leads to significant performance improvements in a variety of prediction and recommendation tasks. We empirically verify this insight using real-world datasets from online social networks and corporate microblogging data. Our work makes several steps towards building models of complex networks, understanding their rich properties, hidden structures and dimensional interdependencies. We develop a novel model, that integrates heterogeneous networks of networks, each with rich properties and hidden dynamics, facilitating multimodal analysis of time-varying, complex social networking data. We study informal communication behavior, information sharing, and influence at the workplace, where formal structures, such as the organizational hierarchy, provide hints of the underlying, implicit social or communication network. Particularly, we develop two simple yet accurate computational models of technology adoption at the workplace at the presence of influence, accurately reproducing the adoption process at the macroscopic level. We also achieve accurate communication intention prediction based on auxiliary information. Last but not the least, we study the structure of online social bookmarking systems, where we significantly improve social tie recommendation by exploiting the dynamics of collaborative annotation.