Abstract:
Modeling real-world systems using complex network analysis has become a popular approach in the last two decades. A complex network is loosely divided into four types of networks, namely (i) Social Network, (ii) Information Network, (iii) Technological Network, and (iv) Biological Network. However, most real-world networks can be represented as Information Network. Majority of the previous literature over information networks consider homogeneous network representation (singular types of nodes and relations) e.g., Citation network, World Wide Web, etc. However, it has recently been realized that Heterogeneous Information Network (HIN) that consists of multiple types of nodes and relations is a better representation for real-world physical systems. For example, a HIN representing a Citation network by considering node types, such as Author, Paper, Venue, etc. and their corresponding relations, captures rich semantics in comparison to the homogeneous Citation network (considering only the papers as node and citation as relation). Motivated with this, our objective is to leverage Heterogeneous Information Network representation in modeling evolution of a given system by solving link prediction problem. In particular, the major contributions of this thesis towards link prediction can be divided into three types of approaches; (i) topology-based, (ii) graph kernel-based and (iii) network embedding-based. For topology-based methods, we adapt the state-of-the-art common neighbor-based local similarity measures to heterogeneous information network.
