Research on digital advertising precision delivery system based on reinforcement learning algorithm

This study proposes a new reinforcement learning framework for Trans-Proximal Policy (TPP), which aims to combine the advantages of Transformer network and Proximal Policy Optimization (PPO) to realize efficient and precise digital advertising. TPP uses the powerful sequence processing capability of Transformer network to capture historical patterns of user behavior, and combines the efficient policy optimization algorithm of PPO to achieve fast and stable policy learning. In the digital advertising scenario, TPP can capture the user's historical behavior sequence data to predict the future interaction probability, and optimize the advertising strategy accordingly. Finally, through comparative experiments and analysis between the model framework and traditional optimization algorithms, this paper verifies that the TPP proposed in this paper can significantly improve the click-through rate, accuracy rate and conversion rate of advertisements while maintaining the stability of policy update, especially in the processing of large-scale user behavior data. This model is proved to be feasible and effective in the digital advertising precision delivery system.