<![CDATA[As digital data volumes surge, recommendation systems have emerged as a vital solution to mitigate information overload. Traditional methods such as collaborative filtering and content-based filtering serve as the foundation of this field. However, these systems—particularly collaborative filtering—face major challenges such as data sparsity and over-smoothing. Recent advancements in deep learning, specifically Graph Neural Network (GNN) models like LightGCN, have proven effective in capturing complex structural user-item relationships. The proposed model integrates Contrastive Learning as an additional supervision mechanism to address overfitting issue that usually appear when model occurs data sparsity. This mechanism enables the model to learn more robust latent features through self-supervised learning while preventing representation degradation in deep graph layers. Furthermore, temporal contexts—including day, month, hour, season, year, and weekend status—are incorporated to capture dynamic user preferences more accurately. Experiments conducted in an offline environment using the MovieLens 1M dataset yielded significant results. The model integrating LightGCN, Contrastive Learning, and temporal context successfully achieved a Hit Ratio of 70.74%. Additionally, the model recorded an NDCG of 0.1888, an MRR of 0.3383, Recall@20 of 0.0628, and a MAP of 0.0959. The proposed model also demonstrated a 4.62% improvement in Hit Ratio compared to the baseline model.]]>
Copyrights © 2026
