Scalable and Reliable Collaborative Spam Filters: Harnessing the Global Social Email Networks

We introduce a collaborative anti-spam system that is based on pervasive global social email networks. Essentially, we provide a solution to this open research problem: given a network of N users who are willing to share information collaboratively (e.g. the digests or ngerprints of known spams), how do we search for each user's content e ciently and reliably in a distributed manner with minimal tra c cost on the network? As a solution to this open problem, our proposed system employs the percolation search process, which makes the tra c generated due to queries for spam digests scale sublinearly as a function of N . However, in order to reap the bene ts of this novel percolation search algorithm, the node degree distribution of the underlying network must be heavy-tailed. Interestingly, latent global social email networks comprising of personal contacts possess a power-law heavy-tailed degree distribution, which renders itself an ideal natural platform to employ the percolation search algorithm. As a result, our proposed distributed spam lter requires no dedicated peer-to-peer (P2P) systems or centralized server-based systems. We have performed large-scale simulations and we nd that the system achieves a spam detection rate close to 100%, while the false positive rate is kept around zero. The bandwidth cost per user as well as the system-wide bandwidth cost are shown to be very low. Electrical Engineering Deptartment, University of California, Los Angeles, CA 90095. y Electrical and Computer Engineering Department, University of Florida, Gainesville, FL 32611