Bing-CSF-IDF+: A Semantics-Driven Recommender System for News

Lies Hooft van Huijsduijnen, Thom Hoogmoed, Geertje Keulers, Edmar Langendoen, Sanne Langendoen, Tim Vos, Frederik Hogenboom, Flavius Frasincar, Tarmo Robal

Abstract: This work proposes the Bing-CSF-IDF+ recommender – a content-based recommender that makes use of semantic relationships, and combines the best features of our earlier introduced Bing-SF-IDF+ and CF-IDF+ systems. First, we make use of concepts and concept relationships from a domain ontology. Next, Bing-CSF-IDF+ employs the synsets and synset relationships from a semantic lexicon that have not been previously captured by the domain ontology. Last, named entities and their frequencies as provided by Bing – not present in the semantic lexicon and domain ontology – are utilized. Our experiments show that Bing-CSF-IDF+ significantly outperforms Bing-SF-IDF+ and CF-IDF+ on \(F_1\)-scores and Kappa statistics based on a news data set.