Recommender systems play a crucial role in guiding users through large volumes of digital content by tailoring suggestions based on individual preferences. However, many collaborative filtering methods assume that user interests remain unchanged over time, which limits their effectiveness in domains where preferences shift, such as movies. This study explores the use of TimeSVD++, an extension of matrix factorization that incorporates time-based variations in user behavior and item popularity. To improve learning efficiency, we filter the dataset to include users with 100 to 150 ratings, a range that balances data sufficiency and sparsity. We evaluate the model using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), and compare it against a standard SVD baseline. Our results indicate that TimeSVD++ improves prediction accuracy by 4.66% in MAE and 2.52% in RMSE, highlighting the benefit of capturing temporal dynamics. These findings support the need for time-aware recommendation models, particularly in systems that serve dynamic user interests, such as streaming platforms and online marketplaces.