A recent direction on improving the performance of recommender systems is by exploiting data semantics. However, previous research has given little attention to the selection of the most relevant data for recommendations. In this paper we present a schema-driven approach for top-N recommendations. We identify the most promising path types in a structured information network, i.e., Linked Open Data (LOD), based on variable importance scores. In contrast to previous work, we focus on which information is most important and remove path types that appear unimportant. The methodology is tested on the MovieLens 1M dataset, semantically enhanced with data from the LOD cloud. The results are threefold. First, we find that the LOD cloud is useful especially for small user profiles. Secondly, we find that the LOD cloud is useful for large user profiles only when the most popular items are not considered. Lastly, we find that selecting the most relevant data from the LOD cloud improves performance compared to using all extracted LOD data.

Additional Metadata
Keywords Information network schema, Linked Open Data, Random forest, Top-N recommendations
Persistent URL dx.doi.org/10.1145/3019612.3019843, hdl.handle.net/1765/100595
Conference 32nd Annual ACM Symposium on Applied Computing, SAC 2017
Wever, T. (Thomas), & Frasincar, F. (2017). A Linked Open Data schema-driven approach for top-N recommendations. In Proceedings of the ACM Symposium on Applied Computing (pp. 656–663). doi:10.1145/3019612.3019843