Heterogeneous Network Mining and Analysis

Abstract

Nowadays, large amounts of data are being created daily through ngertips with the emergence of abundant social media. With the exponential growth of the Internet over the past decades, there has been a surge of interest in the capability to extract useful data, trends and structures on these social platforms as they act as a gateway for online commercialization and information propagation. Heterogeneous networks model di erent types of objects and relationships among them. Compared to homogeneous networks, heterogeneous networks can fuse information from multiple data sources and social platforms. Therefore, it is natural to model complex objects and their relationships in big social media data with heterogeneous networks. Despite decades of technique development for various data mining tasks, few of them target heterogeneous networks. Heterogeneity is a key element in contemporary social networks which provides diversi ed perception of networks. Therefore, heterogeneous network analysis has become an important topic in data mining in recent years that has been attracting increasing attention from both industry and academia, as they provide more comprehensive and interesting analysis results than their projected homogeneous networks. Motivated by these considerations, this thesis presents a series of new techniques for knowledge discovery in heterogeneous networks. In particular, the methods proposed in this thesis have been applied to a wide range of applications including community discovery, ranking and information retrieval. For dynamic heterogeneous networks, our research presents a more e ective network embedding technique when compared to the existing state-of-the-art methods. Throughout this thesis, we highlight how our methodologies were able to identify more tightly coupled communities in heterogeneous networks, more accurately rank top performing social actors and having the capability to view heterogeneous networks in a dynamic construct.