In many important data mining applications, the input networks may be collected from different sources, at different times, at different granularities, with partially or completely different sets of nodes, and thus create the disparity issue. The network correspondence problem, which aims to find the node or network alignment across different input networks, is a vital stepping stone behind a variety of high-impact applications. For example, in bioinformatics, network correspondence is often the very first step toward discovering which diseases are related to which proteins in order to help design new drugs or re-purpose the existing ones; in brain-informatics, it can help detect which brain wirings are correlated with certain diseases and personality traits; in management, finding the correspondence between different team networks is often the key to characterize high-performing vs. dysfunctional teams within an enterprise. The vast majority, with only very few exceptions, of the existing work on network correspondence focuses on pairwise alignment for static and homogeneous (i.e., uni-partite) graphs, although many emerging applications often produce multiple (more than two), dynamic and heterogeneous graphs. The overall goal of this project is to discover correspondence in disparate networks in order to enable collective mining of them.
This project will investigate three main research tasks, which incorporate constraints from realistic scenarios and applications: (1) Linkage of heterogeneous networks with multiple types of nodes and edges, (2) Linkage of dynamic networks, and (3) Collective network linkage, as opposed to pairwise comparison and alignment of networks. Accurate and efficient linkage of different types of networks will enable the applications of the existing graph mining tools to a collection of disparate networks, and lead to new insights in a variety of important application domains. This project will advance the state-of-the-art techniques on mining disparate networks in multiple dimensions, including its generality, applicability, effectiveness, and scalability. The algorithms developed from this project will be applicable to a wide range of high-impact domains, such as social sciences, brain-informatics, and bioinformatics. The research outcomes will be disseminated by publications, conference tutorials, open-source software, as well as potential tech transfer.
