<p>Reciprocal recommender system seeks to provide user recommendations by taking into account a shared consensus on preferences between the involved parties. It considers the preferences of both parties involved to ensure compatibility and mutual satisfaction. The growing prevalence of online platforms for interest-based groups motivates the need for group reciprocal recommender systems, which suggest matches to a group of users based on their collective preferences. In this work, GRRS-SiameseBiGRU-UCB, a deep learning contextual bandits framework using Siamese architecture for group reciprocal recommendations is proposed. It entails the following pivotal steps: (i) Group formation, (ii) Generating reciprocal recommendations for individual group members, and (iii) Creating group reciprocal recommendations. Agglomerative hierarchical clustering is used to group similar users together. Group formation is followed by generating reciprocal recommendations for group members using Siamese bidirectional Gated Recurrent Units (bi-GRU) network. Contextual bandit policy with an upper confidence bound approach is utilized to effectively balance between exploiting and exploring user interests, resulting in higher rewards in the long term. Average without misery, a consensus-based approach is used to generate group reciprocal recommendations. This helps in finding common preferences and agreement among all members of the group. The proposed approach incorporates multiple aspects such as a user’s multi-criteria ratings, awareness of popularity, demographic information, and user availability from the explicit user profiles to produce popularity-aware group reciprocal recommendations. The effectiveness of GRRS-SiameseBiGRU-UCB is validated by conducting experimental investigations on four real-world datasets.</p>