Applications of Matrix Factorization in Recommender Systems

Due to the huge amount of information available online, it seems necessary to have a recommender system that automatically and intelligently suggests to users. Recommendation systems constitute a specific type of information filtering technique that attempt to present items according to the interest expressed by a user. Collaborative Filtering (CF) is the process of evaluating information using the opinion of other people and computes recommendations based on the information about similar items or users. One of the challenging problems is the sparsity problem of the user-item matrix. Collaborative recommenders try to capture relationships among user-user or item-item pairs by reducing the dimensionality of the user or rather item space. Matrix factorization techniques are employed to reduce the dimension of the observed dataset. In our work we want to investigate the well-known Singular Value Decomposition (SVD) and Semi Discrete Decomposition (SDD), which factorize a matrix into three low-dimensional matrices. The comparison of the methods shows that although SVD has the better results than SDD, SDD is efficient in terms of time and especially memory by producing an error close to SVD.