How Sensitive is Recommendation Systemʼs Offline Evaluation to Popularity? (Poster) 全itemに対する推薦精度はPairwiseの⽅が強い (⼀般にそう⾔われているはず) ただしtestをrare itemに絞っていくと 徐々にMFが優勢に
Reference • REVEAL Workshop 2019: https://sites.google.com/view/reveal2019/home • RecoGeym Challenge: https://sites.google.com/view/recogymchallenge/home • Metrics, Engagement & “Recommenders”. Mounia Lalmas. : https://www.slideshare.net/mounialalmas/engagement- metrics-and-recommenders • Marginal Posterior Sampling for the Slate Bandits. Maria Dimakopoulou, Nikos Vlassis, and Tony Jebara. In Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence (IJCAI), 2019. • Deriving User- and Content- specific Rewards for Contextual Bandits. Paolo Dragone, Rishabh Mehrotra, and Mounia Lalmas. In Proceedings of the International World Wide Web Conference (WWW), 2019. • How Sensitive is Recommendation Systemʼs Offline Evaluation to Popularity? Amir H Jadidinejad, Craig Macdonald, and Iadh Ounis. ACM RecSys Workshop on Reinforcement and Robust Estimators for Recommendation (REVEAL), 2019. • Counterfactual Cross-Validation. Yuta Saito and Shota Yasui. ACM RecSys Workshop on Reinforcement and Robust Estimators for Recommendation (REVEAL), 2019.