The variability of user preferences over time poses a major challenge for recommender systems. Consequently, several works have focused on time-sensitive recommender systems, leading to many studies in this domain. However, existing works that adapt Graph Convolutional Networks (GCN) to temporal recommendation struggle with the item-wise cold start problem, where it’s difficult to make accurate recommendations for new items with little or no interaction data. This paper targets the temporal dimension when providing user recommendations, integrating and leveraging temporal data to enhance the recommendation process. For this purpose, we develop a temporal recommender model incorporating seasonality filtering to the famous graph-based model LightGCN by tweaking its recommendation mechanism to overcome the difficulties posed by item-wise cold start. The Season Filtering component is designed to account for periodic patterns according to the most dominant season by explicitly recognizing and modeling these seasonal effects. The empirical study on real-world datasets proves that the proposed recommender system significantly outperforms the state-of-the-art methods in terms of recommendation performance, with a mean average precision of 0.90.