A Hybrid Deep Learning Framework Integrating RoBERTa, Gated Recurrent Units, and Quantum Crested Porcupine Optimizer for Social Media Engagement Prediction

Social media engagement prediction has become a significant research area in computational social science, enabling content optimization and providing users with personalized experiences across digital platforms. Despite significant advances in machine learning and deep learning, current models still face challenges such as class imbalance, limited ability to capture long-term temporal dependencies, and unstable training convergence in high-dimensional semantic spaces. To address these challenges, this study proposes a hybrid RoBERTa-GRU-QCPO framework that combines contextual language representations generated by RoBERTa with the sequential modeling capability of Gated Recurrent Units (GRU), while the Quantum-Inspired Crested Porcupine Optimizer (QCPO) automatically identifies high-quality hyperparameter configurations to improve training convergence and predictive performance. The contextual embeddings are then fed into a Gated Recurrent Unit (GRU) layer that models sequential dependencies among the contextual token representations generated by RoBERTa, enabling more effective learning of semantic patterns relevant to engagement prediction. The framework addresses the optimization challenges associated with training complex deep architectures by using the Quantum-Inspired Crested Porcupine Optimizer (QCPO), a metaheuristic optimization algorithm inspired by defensive mechanisms in crested porcupines and quantum-inspired concepts such as superposition and tunneling. Extensive experiments conducted on a large-scale social media dataset containing approximately 1.6 million samples demonstrate the effectiveness of the proposed framework. The proposed model achieved an accuracy of 97.95%, precision of 97.76%, recall of 98.12%, and F1-score of 97.94%, consistently outperforming conventional machine learning methods, deep learning architectures, and state-of-the-art transformer-based models. These results demonstrate the robustness and practical applicability of the proposed framework for accurate social media engagement prediction.