Abstract: Aimed at the problem that the existing trajectory-based route recommendation algorithms overly rely on user travel preferences while ignoring the structural characteristics of the road network, a structure-aware and preference-based route recommendation algorithm (SPR2) is proposed. Based on the A* algorithm, SPR2 used a multi-head self-attention network and Node2vec to embed routing preference and network structural characteristics into the representation of each node, and used multi-layer perceptron to model the mapping from two node representations to the heuristic cost h between them. According to the heuristic search of h, the optimal path that met the user's preferences and had the lowest cost was searched. The test results show that SPR2 outperforms other route recommendation algorithms in terms of accuracy and robustness.