<p>Group recommender systems (GRS) are essential for collective decision-making across various domains. However, traditional approaches often struggle to balance diverse group interests and tend to overlook crucial contextual factors and multi-criteria evaluations inherent in real-world choices. To address these limitations, we propose a Context-Aware Multi-Criteria Group Recommender System (CA-MCGRS) that employs a multi-head attention (MHA) mechanism. This architecture dynamically integrates individual user preferences, multiple evaluation criteria, and diverse contextual influences, enabling adaptive and nuanced recommendations. We conducted experiments on two distinct datasets: the ITM-Rec educational dataset, which includes multi-criteria ratings (e.g., application relevance, data quality, usability) and specific academic contexts (e.g., class, semester, lockdown status), and the widely-used MovieLens-100k dataset, adapted for group scenarios with simulated multi-criteria and contextual dimensions (e.g., time of day, season). Our results demonstrate that the MHA-based CA-MCGRS significantly improves predictive accuracy compared to various baseline models, particularly when leveraging rich multi-criteria information and adapting to specific contextual situations, such as the impactful lockdown status in the educational setting or temporal variations in the entertainment domain. Although the model exhibits longer training durations, it demonstrates superior generalization capabilities. These findings underscore the substantial benefits of integrating multi-criteria analysis and contextual awareness in GRS, offering valuable insights for developing more precise and relevant recommender systems.</p>