In this paper we present a method that allows us to use a generic full text engine as a k-nearest neighbor-based recommendation system. Experiments on two real world datasets show that accuracy of recommendations yielded by such system are comparable to existing spreading activation recommendation techniques. Furthermore, our approach maintains linear scalability relative to dataset size. We also analyze scalability and quality properties of our proposed method for different parameters on two open-source full text engines (MySQL and SphinxSearch) used as recommendation engine back ends.