LMM-Incentive: Large Multimodal Model-based Incentive Design for User-Generated Content in Web 3.0

Jinbo Wen, Jiawen Kang, Linfeng Zhang, Xiaoying Tang, Jianhang Tang, Yan Zhang, Zhaohui Yang, Dusit NiyatoarXiv 2025 · DOI · OpenReview

Abstract

Web 3.0 ecosystems rely on user-generated content (UGC), yet they suffer from significant information asymmetry, leading to adverse selection and moral hazard. This paper proposes LMM-Incentive, a framework that leverages Large Multimodal Models (LMMs) as objective quality evaluators within a contract-theoretic incentive design. By employing Few-Shot and Chain-of-Thought prompting, LMMs provide reliable quality ratings for multimodal content. To optimize reward structures in dynamic decentralized environments, we introduce a Mixture of Experts (MoE)-based Proximal Policy Optimization (PPO) algorithm. Experimental results demonstrate that LMM-Incentive achieves superior reward convergence and robustness compared to standard reinforcement learning baselines, effectively bridging the gap between automated content evaluation and decentralized economic incentives.

1. Introduction: The Web 3.0 Reward Gap

The transition to Web 3.0 promises a decentralized ecosystem where users maintain ownership and monetization rights over their content. However, these platforms face significant "Information Asymmetry" challenges. Specifically, platforms often cannot observe a user’s true effort (Adverse Selection), and users may provide low-effort contributions to maximize rewards (Moral Hazard). Traditional solutions, such as community voting, are often slow, unscalable, and susceptible to collusion.

Overview of the LMM-Incentive framework.

To address these issues, we propose an automated, fair, and dynamic incentive mechanism that utilizes LMMs to evaluate multimodal content and adapts to the volatile nature of decentralized networks.

2. Methodology

The LMM-Incentive framework consists of three core components:

2.1 LMM-based Quality Evaluation

Instead of relying on subjective human voters, the system employs LMM agents (e.g., GPT-5) to serve as objective arbitrators. To ensure high-fidelity assessments, we utilize Few-Shot Prompting to provide the model with context and Chain-of-Thought (CoT) Prompting to force the AI to articulate its reasoning step-by-step. This process yields a precise "Quality Rating" based on metrics like clarity and aesthetics.

Prompting strategy for LMM agents.

2.2 Contract-Theoretic Modeling

The interaction between the platform (Principal) and users (Agents) is modeled as a mathematical contract. This design ensures Individual Rationality (IR), guaranteeing user participation, and Incentive Compatibility (IC), which aligns user self-interest with honest, high-effort contributions.

2.3 MoE-based PPO Algorithm

To handle the non-stationary and dynamic environments of Web 3.0, we developed an enhanced Proximal Policy Optimization (PPO) algorithm incorporating a Mixture of Experts (MoE) architecture. A gating network dynamically selects the most suitable expert sub-networks to design the contract based on the current network state, allowing for high model capacity and efficient computation.

3. Experimental Evaluation and Results

The framework was tested using simulated environments where user reputation followed a Beta distribution and UGC (images) was generated via Stable Diffusion.

The proposed MoE-PPO algorithm achieved a test reward of 21.75, significantly outperforming standard PPO (20.63), Soft Actor-Critic (SAC) (-0.95), and Generative Diffusion Models (4.64). Furthermore, the MoE-PPO architecture demonstrated faster convergence, proving its ability to handle complex, high-dimensional state spaces effectively. The resulting optimal contracts were successfully deployed as Ethereum Smart Contracts using Solidity, demonstrating the practical feasibility of on-chain AI-driven incentives.

4. Conclusion and Future Work

LMM-Incentive provides a robust solution for automating rewards in decentralized ecosystems. While effective, current limitations include the latency and cost associated with high-end LMM APIs. Future research will explore the use of smaller, fine-tuned local models and more advanced Transformer-based MoE architectures to capture sequential dependencies in network states. Testing the framework within a live Decentralized Autonomous Organization (DAO) remains a critical next step for real-world validation.

Summary updated

Abstract

Web 3.0 represents the next generation of the Internet, which is widely recognized as a decentralized ecosystem that focuses on value expression and data ownership. By leveraging blockchain and artificial intelligence technologies, Web 3.0 offers unprecedented opportunities for users to create, own, and monetize their content, thereby enabling User-Generated Content (UGC) to an entirely new level. However, some self-interested users may exploit the limitations of content curation mechanisms and generate low-quality content with less effort, obtaining platform rewards under information asymmetry. Such behavior can undermine Web 3.0 performance. To this end, we propose \textit{LMM-Incentive}, a novel Large Multimodal Model (LMM)-based incentive mechanism for UGC in Web 3.0. Specifically, we propose an LMM-based contract-theoretic model to motivate users to generate high-quality UGC, thereby mitigating the adverse selection problem from information asymmetry. To alleviate potential moral hazards after contract selection, we leverage LMM agents to evaluate UGC quality, which is the primary component of the contract, utilizing prompt engineering techniques to improve the evaluation performance of LMM agents. Recognizing that traditional contract design methods cannot effectively adapt to the dynamic environment of Web 3.0, we develop an improved Mixture of Experts (MoE)-based Proximal Policy Optimization (PPO) algorithm for optimal contract design. Simulation results demonstrate the superiority of the proposed MoE-based PPO algorithm over representative benchmarks in the context of contract design. Finally, we deploy the designed contract within an Ethereum smart contract framework, further validating the effectiveness of the proposed scheme.
BibTeX
@misc{wen2025lmm,
  author = {Jinbo Wen and Jiawen Kang and Linfeng Zhang and Xiaoying Tang and Jianhang Tang and Yan Zhang and Zhaohui Yang and Dusit Niyato},
  title = {{LMM-Incentive: Large Multimodal Model-based Incentive Design for User-Generated Content in Web 3.0}},
  year = {2025},
  eprint = {2510.04765},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2510.04765},
}
Figure