Improving Recommendation Performance through Ontology-Based Semantic Similarity

Abstract

AbstractMaking personalized recommendation according to preferences of users is of great importance in recommender systems. Currently most book recommender systems take advantage of relational databases for the representation of knowledge and depend on historical data for the calculation of relationships between books. This scheme, though having been widely used in existing methods based on the collaborative filtering strategy, overlooks intrinsic semantic relationships between books. To overcome this limitation, we propose a novel approach called COSEY (COllaborative filtering based on item SEmantic similaritY) to achieve personalized recommendation of books. We derive semantic similarities between books based on semantic similarities between concepts in an ontology that describes categories of books using our previously proposed method DOPCA, and we incorporate such similarities between books into the item-based collaborative filtering strategy to achieve personalized recommendation. We validate the proposed COSEY approach through comprehensive experiments and show the superior performance of this approach over existing methods in the recommendation of books.Keywordsrecommender systemsontologysemantic similaritydegrees of overlap in pathsdepth of the lowest common ancestor node