In recent years, research has focused on enhancing recommendation list diversity while maintaining accuracy. Traditional methods typically use a two-stage process: first, generating a candidate list using collaborative filtering (CF) or hybrid methods to prioritize accuracy by selecting top-N items with the highest predicted ratings, and second, refining this list by re-ranking or removing similar items to improve diversity, resulting in M items (M ≤ N). However, these approaches often estimate diversity solely based on item differences, neglecting their relevance to user profiles during refinement. This paper introduces a novel approach that incorporates diversity considerations from the outset by addressing both item differences and their alignment with user profiles. Two proposed algorithms aim to enhance diversity while preserving accuracy by leveraging the distance between item content and user profiles. Experiments on the MovieLens dataset, comparing these algorithms with three baselines, demonstrate their effectiveness. The results show that emphasizing diversity from the start yields significantly more diverse recommendations than traditional two-stage methods. .