Title: Studying Homophily through Network Randomization

Speaker: Prof. Kostas Pelechrinis

Date: Tuesday, April 5, 2016

Time: 1:30 to 3:30 PM

Location: 10th Floor Conference Room, 177 Huntington Ave.

Abstract: Homophily is the tendency of individuals to connect with others with similar characteristics. Sociologists and network scientists have extensively studied homophily and have identified peer influence and social selection as its main causes. While well-established metrics for quantifying homophily with respect to single dimensional characteristics exist, these provide a biased view when applied directly on multi-dimensional attributes. The latter are able to capture more complex characteristics of nodes in a network such as mobility.

In the first part of my talk, I will present VA-index, a metric for quantifying the homophily of a network with respect to a vector attribute. VA-index is based on network randomization and empirical hypothesis testing. In brief, VA-index quantifies the pairwise similarity of connected nodes in the network with the one that should have been expected if connections were made at random. Our experiments on synthetic network data show that the VA-index is able to accurately quantify the homophily patterns in the network.

In the second part of my talk, I will present our study of spatial homophily, i.e., homophily with respect to the mobility patterns. We begin by quantifying the phenomenon using the VA-index. Then we decompose the impact of peer influence and social selection on the observed homophily by utilizing appropriate randomized null models. Our main finding indicates that up to 40% of the spatial similarity between friends - at a city-scale level - can be attributed to peer influence!

Readings:

  • Ke Zhang and Konstantinos Pelechrinis. Understanding Spatial Homophily: The Case of Peer Influence and Social Selection. In WWW 2014: 271-282.
  • Konstantinos Pelechrinis. Matching Patterns in Networks with Multi-dimensional Attributes: A Machine Learning Approach. In Social Network Analysis and Mining, 4:188, December 2014.
  • Konstantinos Pelechrinis and Dong Wei. VA-index: Quantifying Assortativity Patterns in Networks with Multidimensional Nodal Attributes. PLOS ONE, 11(1): e0146188, January 2016.

    • Short bio: Konstantinos "Kostas" Pelechrinis is an Assistant Professor of Information Sciences at the University of Pittsburgh, where he leads the Network Data Science Lab. He received his doctorate from the Computer Science Department at the University of California at Riverside (UCR), under the supervision of Professor Srikanth V. Krishnamurthy. Before UCR, he obtained his Diploma degree from the Electrical and Computer Engineering department of the National Techincal University of Athens, where he worked with Professor Vasileios Maglaris at the Network Management and Optimal Design Laboratory. For more details, visit his website.