Co-presented by: Stephen Mussman & John Moore
Online social networks have become increasingly prevalent in today’s world. Much research has been devoted to the study of the data created by these social networks. However, for privacy reasons, companies can’t release their datasets to many researchers. We also want to somehow evaluate various properties in our data, but have no way of doing so since we may not have very much data. Moreover models help to explain the process by which networks are naturally generated. Research in the area of generative network models attempts to alleviate these problems by generating synthetic data with similar properties to the corresponding real-world data.
A few generative models known as Erdos-Renyi, Chung Lu, and TCL are generative graph models that are efficient samplers and linear in number of edges rather than quadratic in the number of nodes. In our research, we consider the problem of matching the assortativity of the original graph by implementing a windowing technique in TCL. However, since assortativity is a one dimensional measure, it is hard to capture other kinds of relationships between nodes. For this reason, we looked at the full distribution of the degrees of nodes in edges and modeling it using a binning method with accept-reject sampling. Our initial results seem to indicate that our method captures degree distribution, clustering coefficients, and the concept of assortativity fairly well.
You can view the recording of the presentation at the "Event Link" below.
Event Link: https://csoi.adobeconnect.com/p55s7h040nq/