The Hybrid QGNN–QSVM (H-QGQS) Framework for the Prediction of University Publication Performance

Abstrak

Assessing university publication performance using scientometric indicators is challenging due to the relational and non-linear nature of scholarly data. This study proposes a hybrid Quantum Graph Neural Network and Quantum Support Vector Machine (QGNN–QSVM) framework for graph-based classification of university publication performance into Low, Medium, and High categories. The framework integrates graph representation learning and quantum-enhanced classification through multiple experimental scenarios, progressing from classical machine learning models to a fully quantum-enhanced learning approach. Experimental results show that the classical Support Vector Machine (SVM) baseline achieved an accuracy of 0.92 on a balanced dataset of 3,818 records, whereas the proposed QGNN–QSVM framework achieved an accuracy of 0.73. The results indicate that while the classical SVM achieves higher overall accuracy, incorporating graph structures and quantum-based embeddings provides more expressive representations of the complex scientometric relationships. In particular, the proposed framework demonstrates strong capability in identifying high-performing universities, which is the most critical category in institutional performance evaluation. These findings suggest that quantum-enhanced graph learning offers a promising alternative for modeling complex scientometric data under current near-term quantum computing constraints. The study also highlights existing limitations and suggests future research directions, including architectural optimization, feature enrichment, and scalability improvements.

Identifier

• OpenAlex: https://openalex.org/W7163745275
• DOI: https://doi.org/10.48084/etasr.17650