The aim of this master thesis is to investigate a set of context-aware recommendation approaches that rely on matrix factorization. In particular, we are interested in comparing not only the approaches as a whole, but, more importantly, their single components. For instance, many of those approaches in principle rely on Single Value Decomposition for computing a lower-dimensional representation of the users, items and ratings given within a recommender system. However, the SVD is e.g., varied slightly or the utilized modules (e.g., for computing the features of items to be recommended or for ranking the items) vary. Therefore, we are interested in decomposing the different recommendation approaches, analyzing and comparing them in detail and aim to develop novel combinations of modules that may contribute to improving recommender performance.