Differential Trust Propagation with Community Discovery for Link-Based Web Spam Demotion

In this paper, we propose a novel differential trust propagation scheme with community discovery, which can be applied to all kinds of trust propagation algorithms. We first use a random walk-based community discovery algorithm to preselect suspicious communities in which the members are almost spam pages. We then utilize these suspicious communities to limit the across-community-boundary trust propagation. Experimental results on WEBSPAM-UK2007 and ClueWeb09 demonstrate that the proposed penalizing scheme significantly improves the performance of trust propagation algorithms such as TrustRank, LCRank, CPV.