Semantic relations for content-based recommendations

The main objective of the CHIP (Cultural Heritage Information Personalization) project is to demonstrate how SemanticWeb and personalization technologies can be deployed to enhance access to digital museum collections. In collaboration with the Rijksmuseum Am- sterdam1, we have developed the Art Recommender2: a content-based recommender system that recommend art concepts based on user ratings of artworks. For example, if a user gives "Night watch" a high ratings, she will get its creator "Rembrandt" recommended. The demonstrator works with the Rijksmuseum ARIA3 database,containing images and metadata descriptions of artworks. The mappings of metedata from ARIA to Icon-class4 and the three Getty vocabularies5 (AAT, TGN and ULAN) [1] allows for recommending a wide range of concepts via various semantic relations, within one (e.g. broader/narrower) or across two different vocabularies (e.g. hasStyle). Fig. 1 presents a top-level overview of the RDF Schema used in CHIP. However, for recommender systems, not all related items are useful or interesting for users. Our main challenge is to find which semantic relations are generally useful for content-based recommendations of art concepts.