This document summarizes a research paper presented at the EC-Web 2014 conference that proposes a linked data recommender system using a neighborhood-based graph kernel approach. It represents items as graphs based on their connections to entities in linked open data and defines a graph kernel to learn user models from these representations. An evaluation on a MovieLens dataset mapped to DBpedia shows the approach outperforms baselines that use naive Bayes classification or vector space models on metrics of accuracy and novelty.