IEEE Fourth International Conference on Big Data and Cloud Computing
We present two new models that take into account the information available in user-created "favorites" lists for enhancing the quality of item recommendation. The first model uses the popularity and ratings of items in the lists to predict ratings for new items to users that have rated some items on the lists. The second model is a matrix factorization model that incorporates lists as implicit feedback in ratings prediction. We compare our two approaches against another work for utilizing favorites lists, as well as the popular Singular Value Decomposition (SVD) ontwo large Amazon datasets and show that utilizing favorites lists gives significant improvements, especially in cold-start cases.