Systematic Review of Machine Learning in Recommendation Systems Over the Last Decade

This study presents a comprehensive overview of the approaches employed in recommendation systems over the last decade. The review primarily draws from two categories of filtering techniques: content-based filtering and collaborative filtering methods. We have reviewed and tabulated approximately forty articles that have been published. Major findings include: (1) collaborative filtering is more often used than content-based filtering, 70% to 23%, the rest is hybrid methods of these two; (2) more than half of the machine learning approaches adopted are supervised learning; however, (3) algorithm-wise, K-means the unsupervised learning algorithm emerged as the most frequently adopted approach in recommendation systems. Also notably, cosine similarity stands out as the prevalent measurement technique.