Community deception: from undirected to directed networks

Abstract

Community deception is about hiding a target community that wants to remain below the radar of community detection algorithms. The goal is to devise algorithms that, given a maximum number of updates (e.g., edge additions and removal), strive to find the best way to perform such updates in order to hide the target community inside the community structure found by a detection algorithm. So far, community deception has only been studied for undirected networks, although many real-world networks (e.g., Twitter) are directed. One way to overcome this problem would be to treat the network as undirected. However, this approach discards potentially helpful information in the edge directions (e.g., A follows B does not imply that B follows A). The aim of this paper is threefold. First, to give an account of the state-of-the-art community deception techniques in undirected networks underlying their peculiarities. Second, to investigate the community deception problem in directed networks and to show how deception techniques proposed for undirected networks should be modified and adapted to work on directed networks. Third, to evaluate deception techniques both in undirected and directed networks. Our experimental evaluation on a variety of (large) directed networks shows that techniques that work well for undirected networks fail short when directly applied to directed networks, thus underlying the need for specific approaches.

Community Deception in Directed Influence Networks

“…One hypothesis to explain this behaviour, is that mod, which partly relies on deleting intra-edges from the target community, consumes more time in finding a non-bridge edge for deletion, especially that it does not consider directions. Similar behaviour of mod in this particular network have been also mentioned by [7,10].…”

“…• Permanence-based deception [24] (neu): another approach based on maximizing permanence loss, which is computed over individual nodes. • We considered the variants of saf, mod, and neu for directed networks introduce in Fionda et al [7] called dmod, dsaf, and dper, respectively. • Random node/edge updates (rnd): randomly selecting nodes/edges for deletion or addition.…”

“…Figure 8 plots changes in numbers of communities ∆ on the vertical axis, against several small-size networks on the horizontal axis. Since both IDec and INDec accommodate directed influence for the first time, we considered several state-of-the-art deceptors [8] that work only with directed networks (although they ignore influence). In the vast majority of cases, both of our novel deceptors IDec and INDec resulted in an increase in the number of communities, and this agrees with their relatively higher deception scores shown previously in Figure 4.…”

“…[24] introduced the neural deception approach based on permanence minimization. dsaf, dmod, and dper have been proposed as counterparts of saf, mod and neural in directed networks, respectively [9]. Q-Attack [1], rem [19], cgn [17] , and prohico [18] are approaches that hide the entire community structure.…”