<p>Recommender systems typically optimize for accuracy, but an exclusive focus on accuracy risks overspecialization and reduced user satisfaction. Serendipity-oriented methods address this by introducing unexpected yet relevant items, though existing approaches such as the Serendipity-Oriented Greedy (SOG) algorithm suffer from high computational cost (O(n³)), limiting scalability. This study proposes the Improved Serendipity-Oriented Greedy (ISOG) algorithm, which removes the diversity parameter to reduce complexity to O (n log n) while preserving accuracy and serendipity. Experiments on the MovieLens dataset show that ISOG achieves significantly higher accuracy at top cutoffs, maintains comparable serendipity to SOG, and reduces runtime from several minutes to under two seconds. Statistical tests confirm the robustness of these improvements, while ablation and sensitivity analyses demonstrate that accuracy drives overall performance, with profile dissimilarity and unpopularity contributing to serendipity. Overall, ISOG provides an efficient and serendipity-aware framework that balances accuracy, surprise, and scalability, offering practical value for next-generation recommender systems.</p>