Research map / Learning theory and algorithms / Submodular and game algorithms
Fraud-Proof Revenue Division on Subscription Platforms
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Formalizes manipulation-resistance axioms (fraud-proofness, bribery-proofness, Sybil-proofness) for subscription revenue division, shows GlobalProp is manipulable and fraud detection NP-hard, and proposes ScaledUserProp, which is fraud/bribery-proof and empirically fairer on streaming data.
Problem
Revenue division on subscription streaming platforms is vulnerable to fraud via fake users and bribery; the status-quo pro-rata rule (GlobalProp) enables profitable manipulation and makes detecting it computationally hard, while alternatives may be unfair.
Approach
Design payment rules that are manipulation-resistant by construction: user-additive rules are fraud- and bribery-proof, and ScaledUserProp caps the influence of high-engagement users (platform takes disproportionate commission from low-engagement users) so it matches UserProp's axioms while being fairer in pay-per-stream terms.
Key results
- GlobalProp is shown NP-hard to use for detecting suspicious artist sets (PSP maximization with |U|<=k).
- UserProp, UserEQ and ScaledUserProp satisfy fraud-proofness and bribery-proofness, while GlobalProp fails both.
- On real-world data ScaledUserProp is reported as the fairest fraud-proof rule, especially for alpha not close to 1; UserEQ is least fair.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use ScaledUserProp instead of the other evaluated rules | Revenue division on real-world music listening data · Music Listening Histories Dataset | fairness (PPS) | Helped |
| Use UserEQ instead of the other evaluated rules | Revenue division on real-world music listening data · Music Listening Histories Dataset | fairness (PPS) | Hurt |
| Use ScaledUserProp instead of the other evaluated rules | Revenue division on synthetic streaming data · Synthetic streaming data | fairness (PPS) | Helped |
| Replace UserProp with UserEQ | Revenue division on synthetic streaming data · Synthetic streaming data | fairness (PPS) | No clear effect |
Surprising findings
- Any fraud-proof or bribery-proof rule can have unbounded maximum envy, so manipulation resistance trades off against payment fairness.
- UserProp and UserEQ perform nearly identically on synthetic data.
Limitations
- Focuses only on bot-created fake users and fake artists; does not address AI-generated content.
- Model abstracts away stream quality and listener heterogeneity; synthetic data does not capture real artist fan-base patterns.
- ScaledUserProp fails Pigou-Dalton consistency and strong Sybil-proofness.
Experimental setup
Datasets: Music Listening Histories Dataset synthetic streaming instances (10,000 users, 1,000 artists)
Benchmarks: Music Listening Histories Dataset synthetic streaming instances
Metrics: pay-per-stream (PPS) maximum envy
Baselines: GlobalProp UserProp UserEQ
Builds on: GlobalProp (pro-rata rule) UserProp UserEQ / Shapley value portioning rules click-fraud-proofness
Closest papers by meaning
- Robust ML Auditing using Prior Knowledge (ICML 2025 · similarity 0.83)
- A Cramér–von Mises Approach to Incentivizing Truthful Data Sharing (NeurIPS 2025 · similarity 0.81)
- Learning from Streaming Data when Users Choose (ICML 2024 · similarity 0.81)
- Carrot and Stick: Eliciting Comparison Data and Beyond (NeurIPS 2024 · similarity 0.80)
- Disguised Copyright Infringement of Latent Diffusion Models (ICML 2024 · similarity 0.80)
- Truthful Aggregation of LLMs with an Application to Online Advertising (NeurIPS 2025 · similarity 0.80)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.