E-commerce has shown to be a promising platform for nowadays business. Customers are now provided a wider set of items to choose from and a lot of companies are migrating from shelf-stores to online markets and offering their products on the web. Because of this broader online catalog, Recommender Systems have been widely used to exhibit the most appropriate items to users given their past consumption preferences. Nonetheless, available data tends to be highly sparse since users only evaluate a small fraction of available items. Recent techniques such as Matrix Factorization and Deep Learning's Autoencoders have demonstrated to be effective on recommendations, yet sparsity effects on such techniques are still unclear. In order to provide the effects of sparsity changes on recommender systems, this paper compares three different algorithms, namely Non-negative Matrix Factorization, Singular Value Decomposition and Stacked Autoencoders, under specific sparsity scenarios of the MovieLens 100k dataset.