An investigation on the serendipity problem in recommender systems

摘要: We design a Knowledge Infusion (KI) process for providing systems with background knowledge.We KI-based recommendation algorithm serendipitous recommendations.An in vitro evaluation shows the effectiveness of proposed approach.We collected implicit emotional feedback on recommendations.Results show that serendipity is moderately correlated surprise and happiness. Recommender are filters which suggest items or information might be interesting to users. These analyze past behavior user, build her profile stores about interests, exploit find potentially items. The main limitation this approach it may provide accurate but likely obvious suggestions, since recommended similar those user already knows. In paper we investigate issue, known as overspecialization problem, by proposing strategy fosters suggestion surprisingly not have otherwise discovered.The enriches graph-based knowledge allows system deeply understand deals with. hypothesis infused could help discover hidden correlations among go beyond simple feature similarity therefore promote non-obvious suggestions. Two evaluations performed validate hypothesis: an experiment subset hetrec2011-movielens-2k dataset, preliminary study. Those actually promotes narrowing accuracy loss.