Network Big Data Analysis: Introduction and Goals

Network has been a widely researched area for analysis and visualization of large-scale data in our complex real world. The previous AS thematic project (ASCEND Project) successfully developed statistical tools that were useful for data analysts and computer scientists on their own needs. However, the ultimate goal of this three-years project is far from complete. The mission of this research group is to continue the development of statistical tools with the aid of new technologies, so that data scientists can efficiently analyze network (or any correlated) data under correct assumptions, while the computer scientists can develop their efficient algorithms with adequate mindsets about statistical inference.

In specific, there are four major problems our group aims to address in following topics:

1. Network Dynamics and Evolution. Instead of the traditional preferential attachment and the recently introduced popularity effects, we aim to propose a new mechanism that can better explain the creation and evolution of real networks. We further explore the generalization of this mechanism from a standard network to a hypergraph network.


2. Network Structure, Clustering and Visualization. Instead of traditional descriptions on network structure via simple statistics on degree or density, the Betti number, which is the best-known indices to characterize the topology of a manifold, and its algorithm are introduced to characterize and subdivide the network. In addition, after clustering networks, we continue to develop the visualization tool based on the U-PASS method for directed and/or weighted networks, and the representative discrepancy functions to quantify the node uniformity.


3. Network Properties. Unlike traditional network properties via all sorts of centrality, we aim to propose a variety of new node properties that describe the role of nodes in a network from different angles. For example, network diversity, node relevance, node similarity, etc. We investigate in the concepts of these new node characteristics, connect them to the existing centrality theoretically, and develop efficient algorithms for computing them from a large-scale network.


4. Applications to Large-Scale Networks. The new techniques and methods are applied to real-world large-scale networks, possibly through the collaborations with experts with respective domain knowledge. They include, but not limited to, the Web of Science, the biological neural networks, the food web, the energy consumption network, the highway traffic flow network, and many others.