Abstract: To address the high computational cost of pairwise graph similarity computation and the large memory overhead caused by storing intermediate query results in multi-subgraph matching, an efficient symbolic algorithm is proposed. Graph neural network(GNN) techniques are employed to pre-cluster query graphs and data graphs, thereby reducing the computational complexity of pairwise graph similarity computation. Furthermore, algebraic decision diagrams(ADDs) are adopted to compactly represent query graph sets and data graphs, thereby reducing memory consumption. To further improve shared subgraph detection efficiency and optimize query ordering, a novel maximum common connected subgraph(MCCSG) algorithm is developed. Finally, an improved subgraph matching algorithm, termed SSMGNN-MQO, is designed to solve the multi-subgraph matching problem. Experimental results demonstrate that the proposed method achieves significantly higher matching efficiency than the existing MQO-based subgraph isomorphism search method.