Urban Vehicle Path Recommendation Method Based on the Improved Adversarial Inverse Reinforcement Learning

Adversarial inverse reinforcement learning (AIRL) algorithms have been used to learn path selection patterns from GPS trajectories. However, their performance can be limited in traffic environments with complex road networks and sparse data. This paper proposes a path recommendation model based on improved adversarial inverse reinforcement learning. By integrating an attention mechanism into the AIRL framework, the model enhances learning efficiency and adaptability. The learned model outputs are utilized as path costs for the Dijkstra algorithm to determine optimal routes. Experimental evaluations on a real-world taxi GPS dataset from Wuhan demonstrate that the proposed model effectively balances path preference learning and travel cost optimization, offering a more intelligent and efficient solution for path recommendation systems.