Recommender Systems aim at selecting and presenting first the information that users could be interested in. This work presents a Recommender System that relies on two concepts: semantic relations in data and a distributed collaborative filtering technique based on Matrix Factorization (MF). On the one hand, semantic technologies may increase relations among data, and thus, recommendation accuracy. On the other hand, MF grants highly accurate predictions in a parallelizable algorithm. Our proposal extends this technique by adding semantic relations to the process. Indeed, we deeply analyze the implicit user interests in the attributes of items. The experimentation phase uses MovieLens dataset and IMDb database. We compare our work against a semantic-less MF technique. Results show high accuracy in recommendations while preserving a high level of domain abstraction. Besides, we alleviate system workload by parallelizing the algorithms process.