Enhancing Recommender Systems Using Graph Neural Networks, Long Short-Term Memory and Textual Embeddings

Recommender systems struggle with cold-start and data sparsity, as new users and items lack sufficient interaction history. Capturing recent user behavior and leveraging side information like reviews and metadata are also crucial for accurate recommendations. To address these challenges, we propose Enhancing Recommender Systems Using Graph Neural Networks, Long Short-Term Memory and Textual Embeddings (ERSUGLT), which constructs a unified user-item graph with edges encoding ratings, review sentiment, and temporal order. Cluster-GCN and GAT layers capture high-order relationships, while a query-aware attention LSTM models sequential user patterns. A fusion module integrates graph and sequence embeddings. ERSUGLT outperforms strong baselines on the Amazon Review and MovieLens datasets, showing that combining relational, temporal, and textual signals yields more accurate, context-aware recommendations.