Preventing unraveling in social networks gets harder

Abstract

The behavior of users in social networks is often observed to be affected by the actions of their friends. Bhawalkar et al. (Bhawalkar et al. 2012) introduced a formal mathematical model for user engagement in social networks where each individual derives a benefit proportional to the number of its friends which are engaged. Given a threshold degree κ the equilibrium for this model is a maximal subgraph whose minimum degree is ≥ κ. However the dropping out of individuals with degrees less than κ might lead to a cascading effect of iterated withdrawals such that the size of equilibrium subgraph becomes very small. To overcome this some special vertices called "anchors" are introduced: these vertices need not have large degree. Bhawalkar et al. (Bhawalkar et al. 2012) considered the ANCHORED κ-CORE problem: Given a graph G and integers b, κ and p do there exist a set of vertices B ⊆ H ⊆ V (G) such that |B| ≤ b, |H| ≥ p and every vertex ν ∈ H \ B has degree at least κ is the induced subgraph G[H]. They showed that the problem is NP-hard for κ ≥ 2 and gave some inapproximability and fixed-parameter intractability results. In this paper we give improved hardness results for this problem. In particular we show that the ANCHORED κ-CORE problem is W[1]-hard parameterized by p, even for κ = 3. This improves the result of Bhawalkar et al. (Bhawalkar et al. 2012) (who show W[2]-hardness parameterized by b) as our parameter is always bigger since p ≥ b. Then we answer a question of Bhawalkar et al. (Bhawalkar et al. 2012) by showing that the ANCHORED κ-CORE problem remains NP-hard on planar graphs for all κ ≥ 3, even if the maximum degree of the graph is κ+ 2. Finally we show that the problem is FPT on planar graphs parameterized by b for all κ ≥ 7. Copyright © 2013, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved.