Recommendation engines leverage past user preferences to forecast their future interests. Many deep learning-based recommendation systems aim to explore the intricate dynamics between users and particular items. Typically, the most recommendation systems have a tendency to suggest items based on their popularity or relevance to the user’s previous interactions, which will lead to a narrow range of recommendations. We would like to tackle this issue by evaluating potential items based on the novelty concept, effectively broadening the spectrum of recommendations. Furthermore, we detail a framework that allows for the seamless integration of this novel feature into next-k item recommendation architectures. Through rigorous experiments, the proposed method demonstrated superior performance compared with existing methods, confirming that integrating a novelty score can significantly refine the quality of recommendations.