Exploitation of incomplete data in recommender systems

Description

The goal of recommender systems is to suggest items (films, books, etc.) that are specifically relevant to each user. Those personalized recommendations are built upon preferences expressed either explicitely or implicitely, using techniques such as collaborative filtering [1, 2]. However, these techniques rely on data that are often of poor quality, often incomplete, noisy or incoherent data, possibly coming from multiple sources, including social networks. For example, in film recommendations the users may tend to only rate movies that they really liked (or the opposite), which creates data that are missing, but not at random [3, 4]. The goal of this work is to study various methods to exploit such data, to extract reliable information about users preferences. . [1] D. Jannach, M. Zanker, A. Felfernig, and G. Friedrich. Recommender Systems: An Introduction. Cambridge Univ Press, 2010. . [2] J.A. Konstan and J. Riedl. Recommender systems: from algorithms to user experience. User Modeling and User-Adapted Interaction, pages 1–23, 2012. . [3] B.M. Marlin and R.S. Zemel. Collaborative prediction and ranking with non-random missing data. In Proceedings of the third ACM conference on Recommender systems, pages 5–12. ACM, 2009. . [4] B.M. Marlin, R.S. Zemel, S.T. Roweis, and M. Slaney. Recommender sys- tems: Missing data and statistical model estimation. In Twenty-Second International Joint Conference on Artificial Intelligence, 2011.