Advancing passenger next-station prediction via collaborative knowledge graph representational learning.

Conventional passenger next station prediction models are often restricted by rigid graph structures, failing to account for the dynamic interactions between passengers and stations. This limitation results in an insufficient representation of travel patterns and associated knowledge. To address these shortcomings, this paper proposes a novel approach that integrates reinforcement learning with knowledge graphs to enable a holistic fusion of heterogeneous data. The proposed method enriches the environmental variables within reinforcement learning frameworks and introduces collaborative updating mechanisms for representing passenger-station interactions based on human travel knowledge graphs. These enhanced representations are then employed to improve the accuracy of next-station prediction. To evaluate the effectiveness of the approach, ablation experiments and comparative analyses with classical algorithms were conducted. The results demonstrate that the proposed method significantly outperforms existing models in predicting next-stations, routes, and travel distances, establishing its efficacy in capturing complex passenger travel behaviors.