Enhancing Collaborative Filtering Using Implicit Relations in Data

This work presents a Recommender System (RS) that relies on distributed recommendation techniques and implicit relations in data. In order to simplify the experience of users, recommender systems pre-select and filter information in which they may be interested in. Users express their interests in items by giving their opinion (explicit data) and navigating through the web-page (implicit data). The Matrix Factorization (MF) recommendation technique analyze this feedback, but it does not take more heterogeneous data into account. In order to improve recommendations, the description of items can be used to increase the relations among data. Our proposal extends MF techniques by adding implicit relations in an independent layer. Indeed, using past preferences, we deeply analyze the implicit interest of users in the attributes of items. By using this, we transform ratings and predictions into “semantic values”, where the term semantic indicates the expansion in the meaning of ratings. The experimentation phase uses MovieLens and IMDb database. We compare our work against a simple Matrix Factorization technique. Results show accurate personalized recommendations. At least but not at last, both recommendation analysis and semantic analysis can be parallelized, alleviating time processing in large amount of data.