Journal of Complex Networks (2016) Page 1 of 22 doi:10.1093/comnet/cnw007 Detection of network communities with memory-biased random walk algorithms Mesut Yucel Department of Bioengineering, Faculty of Engineering, Ege University, Izmir, Turkey Lev Muchnik Department of Internet Studies, School of Business Administration, Hebrew University, Jerusalem, Israel and Uri Hershberg † The School of Biomedical Engineering & Health Systems, Drexel University, Philadelphia, PA, USA † Corresponding author. Email: uri.hershberg@drexel.edu Edited by: Ernesto Estrada [Received on 24 April 2015; accepted on 17 February 2016] Community structure and its detection in complex networks has been the subject of many studies in the recent years. Towards this goal, we have created a novel approach based on the analysis of the motion of a memory-biased random walker, i.e. an entity that traverses the network with some tendency to follow or avoid pathways it has previously traversed. We found that the walker tends to remain inside communities, that is, subsets of the network nodes which are more connected to each other, rather than to the rest of the network. Based on this trait of the MBRW we developed a method to detect communities and tested its performance on a range of networks with different levels of community structure. In all tested cases, the method proved to be at least as effective as Girvan–Newman or Infomap while outperforming them when communities were less well defined. Keywords: memory-biased random walker (MBRW); community detection; module density; cored networks. 1. Introduction Networks are often used to describe the overall set of interactions in complex biological or social sys- tems [1,2]. Such a holistic view of these systems is quite useful to understand system-wide processes, but it conflates the local interactions necessary for understanding of emergence of dynamic behaviour in societies, cells or multi-cellular organisms. To resolve its local interactions, we must ask how a net- work is segregated. Are there common rules of action in networks that allow segregation as an emergent phenomenon of action on the network? We suggest here a novel approach based on the stochastic move- ment of an entity on the network. The movement of an entity can be seen as a metaphor of information flow from a node to one of its neighbours and modelled as a sequence of local events governed by a memory of size s which biases by a factor α the likelihood of leaving a node by the remembered path. That memory capacity defines the timeframe within which the entity tends to retrace the paths it had recently travelled. In this work, we show that a random walk with a propensity to retrace its own recent c  The authors 2016. Published by Oxford University Press. All rights reserved.