Traditional recommendation systems often fail to adapt to the dynamic nature of user preferences, resulting in reduced relevance and user satisfaction. This paper proposes a novel approach that integrates Temporal Gradient Entropy (TGE) into collaborative filtering to model the evolution of user behavior over time. By analyzing the rate of change in user-item interactions and using entropy to measure changes in behaviour, our method enhances prediction accuracy and personalization. We implement the TGE-enhanced model using Singular Value Decomposition (SVD) and evaluate it against baseline methods using the Netflix dataset. Experimental results demonstrate that the proposed TGE-SVD model outperforms traditional collaborative filtering techniques in terms of RMSE and robustness to user interest drift, offering a more context-aware and temporally sensitive recommendation framework.