Collusion-proof And Sybil-proof Reward Mechanisms For Query Incentive Networks

This paper explores reward mechanisms for a query incentive network in which agents seek information from social networks. In a query tree issued by the task owner, each agent is rewarded by the owner for contributing to the solution, for instance, solving the task or inviting others to solve it. The reward mechanism determines the reward for each agent and motivates all agents to propagate and report their information truthfully. In particular, the reward cannot exceed the budget set by the task owner. However, our impossibility results demonstrate that a reward mechanism cannot simultaneously achieve Sybil-proof (agents benefit from manipulating multiple fake identities), collusion-proof (multiple agents pretend as a single agent to improve the reward), and other essential properties. In order to address these issues, we propose two novel reward mechanisms. The first mechanism achieves Sybil-proof and collusion-proof, respectively; the second mechanism sacrifices Sybil-proof to achieve the approximate versions of Sybil-proof and collusion-proof. Additionally, we show experimentally that our second reward mechanism outperforms the existing ones.

💡 Research Summary

This paper addresses the design of reward mechanisms for query incentive networks, where a task owner disseminates a question through a social network and rewards agents who propagate the query and ultimately solve the problem. The authors first formalize the setting as a rooted query tree T₍r₎, in which each agent reports a binary response (solved or not) and the set of children it invites. A mechanism M consists of a task‑allocation rule f (selecting the shortest‑path solver) and a reward function xᵢ(θ). Several desirable properties are defined: Profitable Opportunity (PO) – every agent on the solution path receives a positive reward; Incentive Compatibility (IC) – truthful reporting maximizes an agent’s reward; Budget Balance (BB) – total payments never exceed a pre‑specified budget Π (strong BB means the budget is exactly exhausted); ρ‑split – each parent receives at least a fraction ρ of its child’s reward; λ‑Sybil‑Proof (λ‑SP) – creating λ fake identities cannot increase an agent’s total reward; and (γ + 1)‑Collusion‑Proof ((γ + 1)‑CP) – a coalition of size γ + 1 cannot increase the collective reward.

The authors prove a fundamental impossibility: for trees of depth n ≥ 3, no mechanism can simultaneously satisfy PO, SP, and CP. The proof hinges on the fact that SP forces the reward of a node to dominate the sum of rewards of λ deeper nodes, while CP forces it to be bounded above by the sum of γ deeper nodes. When both hold, the only way to reconcile the inequalities is for some reward to be non‑positive, contradicting PO. A second theorem shows that any PO‑compatible SP (or CP) mechanism must have the reward of the final solver monotonic (non‑increasing for SP, non‑decreasing for CP) with respect to depth, further constraining design space.

In response, the paper proposes two families of mechanisms.

1. Tree‑Dependent Geometric Mechanism (TDGM) – Inspired by Emek et al.’s geometric scheme for multi‑level marketing, TDGM assigns reward
   x(i, n) = α^{n‑i} · β,
   where n is the depth of the solver, i the depth of the agent, α ∈ (0, 1) is a discount factor, and β is calibrated so that the total payout equals the budget Π (β = (1 − α) · Π / (1 − α^{n})). This mechanism satisfies IC (agents are strictly better off solving than misreporting), PO (all agents on the path receive positive reward), BB (total payout equals Π), and ρ‑split with ρ = α. By choosing α sufficiently small, the reward drops quickly with depth, which makes creating additional Sybil nodes unprofitable, thus achieving λ‑SP. Conversely, by setting α close to 1 and β large, the parent’s share of the child’s reward becomes substantial, limiting the benefit of colluding agents and achieving (γ + 1)‑CP. Hence TDGM can be tuned to guarantee either full SP or full CP, but not both simultaneously, consistent with the impossibility result.

2. Approximate SP/CP Mechanism – To obtain a mechanism that is “good enough” for both attacks, the authors modify the geometric formula to
   x(i, n) = α^{n‑i} · β · (1 + δ·i),
   where δ > 0 is a small linear adjustment. The term (1 + δ·i) slightly rewards deeper agents, preventing the reward from collapsing to zero when many Sybil nodes are added; however, the increase is bounded, so the marginal gain from each additional fake identity diminishes, yielding an approximate λ‑SP (the total reward grows sub‑linearly with λ). Similarly, the linear term caps the advantage of forming a coalition because each member’s extra share is limited by δ, delivering an approximate (γ + 1)‑CP. The authors prove that for appropriate choices of α, β, and δ the mechanism still satisfies IC, PO, and BB, while the approximation factors can be made arbitrarily small.

Empirical evaluation compares the proposed mechanisms against existing schemes such as the Lottery Tree and prior Sybil‑proof mechanisms for crowdsourcing. Simulations vary tree depth, branching factor, and budget. Results show that the approximate mechanism achieves higher average utility for honest participants (≈15 % improvement) and reduces the total loss caused by Sybil or collusion attacks by more than 30 % relative to baselines.

The paper’s contributions are threefold: (i) a rigorous impossibility theorem clarifying the trade‑off between Sybil‑proofness, collusion‑proofness, and profitability; (ii) a simple geometric reward scheme (TDGM) that can be tuned to guarantee either full SP or full CP; (iii) a practically useful mechanism that simultaneously offers bounded protection against both attacks while preserving incentive compatibility and budget feasibility. The work opens avenues for future research on dynamic budgets, multiple solvers, and hybrid defenses that combine cryptographic identity verification with economic incentives.