<![CDATA[The accuracy of recommendation systems is vital for successful personalization in e-commerce. However, the low frequency of repeat purchases creates hight data sparsity, limiting models in capturing user preferences. This study compares two latent factor-based algorithms. Matrix Factorization (MF) and Neural Matrix Factorization (NeuMF), using the Olist transaction dataset through data preparation, k-core filtering, and leave last out splitting. Performance was evaluated using HR@10 and NDCG@10. Results show that MF outperforms NeuMF, achieving HR@10 of 0,057 and NDCG@10 of 0,133. Although NeuMD is more complex and represents a deeper learning-based approach, MF can still be more suitable in certain data conditions, especially when interaction are limited. These findings highlight that simpler models may remain more efficient under sparse data, while NeuMF requires richer interactions histories. The study emphasizes repeat purchase frequency as a key factor in designing adaptive reommendations systems.]]>
Copyrights © 2025
