Automatic seed set expansion for trust propagation based anti-spam algorithms

  • Authors:

  • Xianchao Zhang;Wenxin Liang;Shaoping Zhu;Bo Han

  • Affiliations:

  • School of Software, Dalian University of Technology, Economy and Technology Development Area, 116620 Dalian, China;School of Software, Dalian University of Technology, Economy and Technology Development Area, 116620 Dalian, China;School of Software, Dalian University of Technology, Economy and Technology Development Area, 116620 Dalian, China;Department of Computer Science and Software Engineering, University of Melbourne, ICT 6.05, 111 Barry St. Carlton, VIC 3010, Australia

  • Venue:
  • Information Sciences: an International Journal
  • Year:
  • 2013

Quantified Score

Hi-index 0.07

Visualization

Abstract

Seed sets are of significant importance to trust propagation based anti-spam algorithms, e.g., TrustRank. Conventional approaches require manual evaluation to construct a seed set, which restricts the seed set to be small in size, since it would cost too much and may even be impossible to construct a very large seed set manually. The detrimental effect will be caused to the final ranking results by the small-sized seed sets. Thus, it is desirable to automatically expand an initial seed set to a larger one. In this paper, we propose an automatic seed set expansion algorithm (ASE) which enriches a small seed set to a much larger one. The intuition behind ASE is that if a page is recommended by a number of trustworthy pages, the page itself should be trustworthy as well. Since links on the Web can be considered as a tool for conveying recommendation, we call links recommending the same page a joint recommendation link structure. The joint recommendation link structures with large enough support degrees are employed by ASE algorithm to obtain new seeds. It can be proved that using the joint recommendation link structure with a suitable support degree, the probability of selecting a spam page as a new seed almost to zero, thus the quality of the expanded seed set can be guaranteed. Experimental results on the WEBSPAM-UK2007 dataset show that with the same manual evaluation efforts, ASE can automatically obtain a lot of reputable seeds with very high quality, and significantly improves the performance of trust propagation based algorithms such as TrustRank and CPV (Computing Page Values).