Subgraph-Aware Joint Modeling for Graph Classification via Efficient Frequent Pattern Mining and Fusion

This paper proposes a subgraph-aware classification framework that integrates efficient frequent subgraph mining with graph neural networks (GNNs) to address the limitations of existing GNNs in capturing explicit local structures. Specifically,we first introduce an improved mining strategy to extract representative and low-redundancy structural patterns from large-scale graph data, significantly reducing computational complexity and pattern duplication. These patterns are then dynamically injected into a subgraph-aware GNN module via a subgraph-corrected attention mechanism and gated fusion strategy, enabling deep collaborative modeling of local and global semantics. An end-to-end joint training mechanism is also introduced to coordinate subgraph selection with representation learning. Experimental results on the IMDB-MULTI and REDDIT-MULTI-12K datasets demonstrate that our method outperforms several state-of-the-art baselines. On IMDB-MULTI, our model achieves an Accuracy of 78.6% and a Macro-F1 score of 77.4%, showing notable gains over competing approaches. Further ablation and visualization analyses confirm the importance of both the frequent subgraph mining component and the subgraph-aware fusion mechanism in enhancing representation quality and model interpretability. This work presents a unified and interpretable framework for graph classification, offering new insights into structural perception and feature fusion in GNN-based learning.