Title: Tensor factorization for temporal recommender systems
Authors: Annie Qu - University of Illinois at Urbana-Champaign (United States) [presenting]
Abstract: Recommender systems have been widely adopted by electronic commerce and entertainment industries, as they bring insightful business intelligence for decision makers and convey useful information to customers efficiently. As an effective predictive model, recommender systems use individual experience or preference to make personalized prediction and recommendation. In general, individual experiences and preferences change over time, and capturing such changes is essential for developing accurate recommender systems. We introduce a temporal recommender system applicable to product forecasting, which achievers more-accurate sales forecasting for stores and manufactures, and improves decision-making on new product introduction. Specifically, we develop a time-varying tensor function, implement tensor factorization under the nonparametric framework to provide the time-dependent predictions, and utilize joint information from subgroups to solve the ``cold-start'' problem in the absence of information from new customers, new products or new contexts. We develop the asymptotic consistency of the predictions and nonparametric parameter estimators. In addition, the proposed method is illustrated for the IRI marketing data. Our numerical studies indicate that the proposed method outperforms existing competitors in the literature.