SKOS-Based Concept Expansion for LOD-Enabled Recommender Systems

This paper presents a concept expansion strategy for Linked Open Data-enabled recommender systems (LDRS). This strategy is based on annotations from Simple Knowledge Organization System (SKOS) vocabularies. To this date, the knowledge structures of SKOS graphs have not yet been thoroughly explored for item similarity calculation in content-based recommender systems (RS). While some researchers have already performed an unweighted concept expansion on skos:broader links, the quantification of the relatedness of concepts from SKOS graphs with quality issues, such as the DBpedia category system, should be further investigated to improve recommendation results. For this purpose, we apply our approach in conjunction with a suitable concept-to-concept similarity metric and test it on three different LDRS datasets from the multimedia domain (i.e., movie, music and book RS). The results showed that our approach has a diversifying effect on result lists, while at least providing the same level of accuracy as a system running in non-expansion mode.