<![CDATA[The rapid growth of e-commerce has generated massive volumes of user interaction data, requiring recommendation systems capable of providing accurate and relevant product suggestions. However, conventional recommendation systems still face sparsity issues and low prediction accuracy in large-scale data environments. This study aims to improve the accuracy of big data-based product recommendation systems by implementing Matrix Factorization using the Alternating Least Squares (ALS) algorithm. The research utilized the Retailrocket E-commerce Dataset consisting of 104,287 interaction records, 18,642 users, and 7,831 products with a sparsity level of 96.8%. Implicit interactions, including product views, add-to-cart activities, and purchases, were transformed into weighted preference values to represent user behavior. The model was implemented using Apache Spark MLlib within a distributed computing environment. Model evaluation was conducted using RMSE, MAE, Precision@10, and Recall@10 metrics with a 5-fold cross-validation approach. The experimental results indicate that the optimal configuration was achieved using 50 latent factors, 0.05 regularization, and 20 iterations, producing an RMSE value of 0.836, MAE of 0.689, Precision@10 of 0.861, and Recall@10 of 0.824. These findings demonstrate that ALS-based Matrix Factorization effectively improves recommendation quality while supporting scalability for large-scale data processing in modern e-commerce environments. Keywords: Matrix Factorization, Alternating Least Squares, Recommendation System, Big Data, Collaborative Filtering.]]>
Copyrights © 2026
