A new e-commerce platform has 10,000 products and 500 users, with sparse interaction data (90% of user-item pairs have no interaction). A data scientist immediately builds a matrix factorization model. What is the primary problem with this approach, and what would be more appropriate?