QRNN-GRU framework for automatic argument and annotation extraction in medical drug reviews

Argument annotation in medical drug reviews is a challenging task due to noisy user-generated content, domain-specific terminology, and subjective expressions of medication efficacy and adverse effects. While argument mining has been explored in this domain, the diversity of modeling architectures investigated remains narrower compared to more extensively studied NLP tasks. In this study, we investigate a sequential QRNN–GRU framework for identifying argumentative components in the medical datasets, aiming to explore a hybrid architecture that balances effective modeling of argumentative structures with computational efficiency for this domain. The proposed model employs QRNN layers to efficiently capture local temporal patterns, followed by GRU layers to model longer-range sequential dependencies. Firefly Optimization is utilized for hyperparameter tuning to improve training stability and convergence behavior. Experimental results on drug review dataset show that the proposed approach achieves an F1-score of 89.20% and an accuracy of 91.47%, outperforming selected baseline models under the evaluated setting. To further examine the generalizability, the model additionally evaluated on Abstract RCT benchmark dataset, where it achieves competitive comparative performance against established state-of-the-art methods. Although the model demonstrates consistent performance across two benchmark datasets, the analysis is limited to medical texts and does not include cross-domain evaluation.