Scalable computational techniques for centrality metrics on temporally detailed social network

Gunturi, Venkata M.V.; Shekhar, Shashi; Joseph, Kenneth; Carley, Kathleen M.

Abstract

Increasing proliferation of mobile and online social networking platforms have given us unprecedented opportunity to observe and study social interactions at a fine temporal scale. A collection of all such social interactions among a group of individuals (or agents) observed over an interval of time is referred to as a temporally-detailed (TD) social network. A TD social network opens up the opportunity to explore TD questions on the underlying social system, e.g., “How is the betweenness centrality of an individual changing with time?” To this end, related work has proposed temporal extensions of centrality metrics (e.g., betweenness and closeness). However, scalable computation of these metrics for long time-intervals is challenging. This is due to the non-stationary ranking of shortest paths (the underlying structure of betweenness and closeness) between a pair of nodes which violates the assumptions of classical dynamic programming based techniques. To this end, we propose a novel computational paradigm called epoch-point based techniques for addressing the non-stationarity challenge of TD social networks. Using the concept of epoch-points, we develop a novel algorithm for computing shortest path based centrality metric such as betweenness on a TD social network. We prove the correctness and completeness of our algorithm. Our experimental analysis shows that the proposed algorithm out performs the alternatives by a wide margin.

Citation

Gunturi, V. M., Shekhar, S., Joseph, K., & Carley, K. M. (2017). Scalable computational techniques for centrality metrics on temporally detailed social network. Machine Learning, 106(8), 1133-1169. https://doi.org/10.1007/s10994-016-5583-7

• Journal Article Type: Article
• Acceptance Date: Jul 5, 2016
• Online Publication Date: Sep 8, 2016
• Publication Date: Aug 1, 2017
• Deposit Date: Sep 27, 2023
• Journal: Machine Learning
• Print ISSN: 0885-6125
• Electronic ISSN: 1573-0565
• Publisher: Springer
• Peer Reviewed: Peer Reviewed
• Volume: 106
• Issue: 8
• Pages: 1133-1169
• DOI: https://doi.org/10.1007/s10994-016-5583-7
• Keywords: Dynamic programming; Time-varying social networks; Time varying networks; Graph theory; Centrality
• Public URL: https://hull-repository.worktribe.com/output/4401459