Recommender Systems Over Data Streams

András A. Benczúr, Levente Kocsis, Róbert Pálovics

Abstract: Recommender systems predict preferences of users to offer them relevant items in case the selection is too large. Recommender systems have to serve in online environments that can be highly nonstationary. Traditional recommender algorithms may periodically rebuild their models, but they cannot adjust to quick changes in trends caused, for example, by timely information. In this article, we investigate online learning based recommender algorithms that can efficiently handle nonstationary datasets. We show that online learning for recommendation is rather the usual than the exceptional task: For example, if no user history is available, we have to build a user model on the fly, based on the interactions in the live user session. To the best of our knowledge, this is the first survey with a comprehensive overview of the ideas for recommendation over streaming data and their implementation in various distributed data stream processing systems.