350 likes | 509 Views
DSybil: Optimal Sybil-Resistance for Recommendation Systems. Ennan Zhai Peking University enzhai@pku.edu.cn. What is the Sybil Attack ?. Sybil Nodes. Honest Nodes. How about the Current Work ?. SybilGuard SybilLimit DSybil SumUp Approach using Puzzles.
E N D
DSybil: Optimal Sybil-Resistance for Recommendation Systems Ennan Zhai Peking University enzhai@pku.edu.cn
What is the Sybil Attack ? Sybil Nodes Honest Nodes
How about the Current Work ? • SybilGuard • SybilLimit • DSybil • SumUp • Approach using Puzzles
Let’s recall SybilGuard … … • SybilGuard: • Bounds the number of sybil nodes accepted; • Guarantees are with high probability; • Approach: Acceptance based on random route intersection between verifier and suspect.
SybilGuard Basic Insight • Undirected graph • Nodes = identities • Edges = strong trust E.g., colleagues, relatives
sybil nodes SybilGuard Basic Insight Sybil Nodes may collude – the Adversary honest nodes attack edges malicious user
sybil nodes SybilGuard Basic Insight But cannot search for such cut brute-force… honest nodes
Normal Random Walk f a e b d c pick random edge d pick random edge e pick random edge c
Random 1 to 1 mapping between incoming edge and outgoing edge Random Route f a e b d a d d e c randomized routing table e d b a c b f f d c Using routing table gives Convergence Property: Routes merge if crossing the same edge
Intersection Verifier Suspect Honest Nodes Sybil Nodes
Intersection Verifier Suspect same intersection Sybil Nodes Honest Nodes
Intersection Verifier for a given intersection
The Lack of SybilGuard ? Is SybilGuard so perfect ? What is the lack of the approach ? … … … …
SybilLimit SybilLimit is a novel protocol which leverages the same insight as the SybilGuard but offers dramatically improved and near-optimal guarantees. Based on the large real-world social networks, SybilLimit has been proved by the first evidence that real-world social networks are indeed fast mixing.
How about DSybil ? This work aims to answer the question: Is it possible to sufficiently diminish the influence of ( a potentially unlimited number of ) sybil identities based on historical behavior ?
DSybil’s Contributions ? • Defend against unlimited num of sybil; • The loss is limited; • Work proved the last optimal system; • Provide a growing defense.
DSybil’s Target Scenarios ? • Recommendation Systems: • Credence: recommends content items; • Digg: recommends stories. Votes: A vote claims a certain object to be good, and this is the only kind of vote DSybil uses.
Leveraging Trust The Obvious, the Subtle and the Challenge … …
DSybil • Two Key Algorithms Details: • Assumes the total number of sybil identities is M (Alg-A); • Improves the algorithm so that it can tolerate an unlimited number of sybil identities, as long as the number of sybil voters on any given object is bounded within M (Alg-B).
Design (Alg-A) At Most M Sybil Identities Total
Design (Alg-A) Overwhelming: An object is overwhelming, if the total trust of the identities voting for the object is at least some constant c, where c > S.
Discussions (Alg-A) Some Intuition about the first algorithm of DSybil … …
Design (Alg-B) How about M Sybil Voters on each object ?
Discussions (Alg-B) Some Intuition about the second algorithm of DSybil … …
A Growing Defense A salient property of DSybil is that its defense against attacks grows over time. The growing defense comes from 3 factors … …
The Rest of DSybil Some further extensions and deeper look are in the technical report. Loosely Bounding M can also be seen in the paper and report … …
My Discussions • The Sybil Attack; • Defense Mechanisms; • Future Works.