Data sparsity is one of the biggest problems faced bycollaborative filtering used in recommender systems. Data imputationalleviates the data sparsity problem by inferring missing ratings andimputing them to the original rating matrix. In this paper, we identify thelimitations of existing data imputation approaches and suggest three newclaims that all data imputation approaches should follow to achieve highrecommendation accuracy. Furthermore, we propose a deep-learning basedapproach to compute imputed values that satisfies all three claims. Based onour hypothesis that most pre-use preferences (e.g., impressions) on itemslead to their post-use preferences (e.g., ratings), our approach tries tounderstand via deep learning how pre-use preferences lead to post-usepreferences differently depending on the characteristics of users and items.Through extensive experiments on real-world datasets, we verify our threeclaims and hypothesis, and also demonstrate that our approach significantlyoutperforms existing state-of-the-art approaches.

Sang-Wook Kim received the B.S. degree in Computer Engineering fromSeoul National University, Korea at 1989, and earned the M.S. and Ph.D.degrees in Computer Science from Korea Advanced Institute of Science andTechnology (KAIST), Korea at 1991 and 1994, respectively. From 1995 to 2003,he served as an Associate Professor of the Division of Computer,Information, and Communications Engineering at Kangwon National University,Korea. In 2003, he joined Hanyang University, Seoul, Korea, where hecurrently is a Professor at the Department of Computer Science & Engineeringand the director of the Brain-Korea-21-Plus research program. He receivedthe Presidential Award of Korea in 2017 for his academic achievement. He isan associate editor of Information Sciences. His research interests includedatabases, data mining, recommendation, and social network & media analysis.