Adaptive Royalty Mechanism for Smart Contract Platforms Based on Deep Reinforcement Learning and Dirichlet Process Mixture: Serv

    In the digital era, intellectual property infringement cases occur frequently, and traditional rights registration and royalty distribution models struggle to adapt to the complex scenarios brought by massive digital works and dynamic contributors. Existing blockchain-based smart contracts can achieve ownership registration and automatic settlement, but most assume fixed contribution relationships, unable to handle the randomness of generative artificial intelligence outputs or perform adaptive modeling of fairness among different contributors. To address this issue, this paper proposes an adaptive royalty mechanism that combines the Dirichlet process mixture model with deep reinforcement learning. It dynamically clusters intellectual property tasks using nonparametric Bayesian methods and, on this basis, utilizes the soft policy gradient algorithm to optimize royalty policies in real time. We construct a simulation dataset and compare it with seven reinforcement learning algorithms (SAC, DDPG, DQN, PPO, TRPO, A2C, TD3). The results show that the proposed DPM-SAC achieves significant advantages in terms of returns, fairness, energy consumption, and convergence speed. This study provides theoretical and practical references for building a transparent, secure, and scalable digital copyright ecosystem.