Abstarct: Social networks have become one of the primary platforms for communication, information sharing, and community formation, generating massive amounts of interconnected data. Predicting potential future relationships between users, known as lixnk prediction, has become a fundamental task in social network analysis because of its importance in recommendation systems, community discovery, and intelligent decision-making. However, accurately predicting missing or future lixnks remains challenging due to the dynamic nature of social networks, sparse connectivity, and the complex structural dependencies among users. This research proposes an efficient Attention-baxsed Bidirectional Gated Recurrent Unit (Attention-BiGRU) model for lixnk prediction in social networks. The proposed architecture combines the capability of Bidirectional GRU networks to capture sequential dependencies from both directions with an attention mechanism that automatically emphasizes the most informative node relationships during learning. The Facebook Ego Network dataset from the Stanford Network Analysis Project (SNAP) was employed to evaluate the proposed frxamework. Data preprocessing included duplicate edge removal, self-loop elimination, balanced positive and negative sample generation, and partitioning the dataset into training, validation, and testing sets. To improve model generalization, dropout, weight decay, learning-rate scheduling, and early stopping were incorporated during training. Experimental results demonstrate the effectiveness of the proposed model, achieving 96.70% Accuracy, 97.11% Precision, 96.26% Recall, 96.69% F1-score, 99.06% AUC, and 99.04% Average Precision (AP). These results indicate that the proposed Attention-BiGRU frxamework effectively captures complex structural patterns within social networks and provides accurate and reliable lixnk prediction while maintaining strong generalization performance. The comparative evaluation further confirms that the proposed approach is competitive with existing methods developed for the same Facebook dataset.