<p>Community discovery from trajectory data is crucial for understanding urban spatial structures. Traditional methods typically model cities as node-centric graphs, which simplifies rich trajectory flows into weighted edges, thereby losing valuable information about movement patterns. While link community detection offers a flow-centric perspective by focusing on graph edges, existing approaches often treat flows as homogeneous and overlook the unique spatial and directional attributes inherent in urban trajectories. To address this specific gap, this paper proposes an novel urban community discovery method based on the line graph for trajectory data. We first design a Line Graph-based Network (LGN) that models traffic flows as nodes and integrates their rich trajectory characteristics as node attributes. Then, we propose a Traffic Flow-based Community Discovery (TFCD) method that considers both trajectory characteristics and graph topology, which can discover communities of traffic flows with similar spatial directions and topological relationships. The effectiveness of our method is demonstrated on simulated and real-world datasets, showing its ability to reveal fine-grained travel patterns that are often missed by conventional approaches.</p>