MatchIA-GCL: A Hybrid Graph Contrastive Learning Approach for Semantic Schema Matching

Schema matching remains a fundamental challenge for achieving data interoperability across heterogeneous information systems. Existing deep learning-based approaches often suffer from semantic drift, overlook the structural topology of schemas, and operate as black-box models with limited interpretability. This study introduces MatchIA-GCL, an explainable hybrid framework for semantic schema matching that combines graph contrastive learning (GCL) and Large Language Models (LLMs). The proposed framework leverages Graph Attention Networks (GATs) optimized through a contrastive learning objective to generate robust structure-aware schema embeddings. Unlike conventional approaches that employ LLMs solely as semantic encoders, MatchIA-GCL integrates an LLM as an Autonomous Validation Agent responsible for resolving only complex and ambiguous matching candidates, thereby reducing computational overhead while improving decision reliability. Furthermore, a post-hoc explainability layer provides transparent and human-interpretable justifications for alignment decisions. Experimental evaluations conducted on benchmark schema matching datasets demonstrate that MatchIA-GCL consistently outperforms state-of-the-art methods, achieving up to 9.4% improvement in F1-score while enhancing explainability and robustness. These results highlight the potential of combining graph representation learning, contrastive optimization, and LLM-assisted validation for next-generation semantic schema matching systems.

@article{Raoui2026,
  title     = {MatchIA-GCL: A Hybrid Graph Contrastive Learning Approach for Semantic Schema Matching},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {7},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Mohamed Raoui and Moulay Hafid El Yazidi and Ahmed Zellou},
  doi       = {10.14569/IJACSA.2026.0170706},
  url       = {https://doi.org/10.14569/IJACSA.2026.0170706}
}