Spatio-temporal knowledge embedding via circular correlation: insights into functional urban area travel pattern mining

Qihong Pan, Yao Chen, Guojiang Shen, Yao Yang, Xiangjie Kong

Abstract: In recent urban studies, understanding the flow patterns of urban residents has become crucial for effective transportation planning and business district design. Traditional data-driven approaches have provided insights but often lead to random and uninterpretable results due to their sole reliance on data features, lacking a deeper contextual and semantic analysis of the underlying patterns. To overcome these limitations, our work introduces a novel framework that fuses holographic knowledge embedding with graph deep learning to predict urban population travel patterns. This dual-driven approach of data and knowledge uniquely integrates traffic geographic information, vehicle trajectory data, and Points of Interest (POI) into a comprehensive urban traffic knowledge graph. Our method not only captures the spatial-temporal dependencies of big data traffic but also models the relationships between geographic, semantic POI information, and urban travel behaviors. The knowledge graph is then processed through a graph deep learning model, enhancing the embedding features and enabling sophisticated link prediction. Compared with conventional data-driven methods, our approach demonstrates significant advancements in harnessing semantic information, leading to more accurate and interpretable predictions of travel patterns. Experimental validation on real-world datasets confirms the effectiveness of our method in capturing complex urban dynamics.

External IDs:dblp:journals/nca/PanCSYK24