Using linked data to build recommender systems

This paper describes how Semantic Web technologies and especially the Linked Open Data (LOD) project can be used to build a new generation of recommender systems. While most of the current recommender systems use private data sets, we show how to exploit the benefits of the LOD community effort to build recommender systems. By providing public, collaboratively created and semantically structured data, it enables cross-domain recommendations by providing data which can be exploited for different recommendation tasks and which is portable between systems, without changing the implementation of the recommendation algorithm. The contributions of this paper are (i) an overview of the LOD community effort and of the data cloud generated by it, (ii) the description and evaluation of both a semanticdistance and a collaborative filtering approach based on Linked Data, (iii) a method for evaluating recommendations using existing categorical data and (iv) the description of a reference architecture for implementing cross-domain recommender systems using Linked Data.