Session-Based Collaborative Filtering for Predicting the Next Song

Most music recommender systems produce a set of recommendation based on user's previous preference. But the information is not always attainable. Focusing on the fact that music listening behavior is a repetitive action of playing one song at a time, we predict the next item based on user's currently selected items even when user's previous preference is not available. We propose a simple but effective recommendation method for this problem called Session-based Collaborative Filtering (SSCF), and we look into the different parameters that affect the recommendation accuracy. Our evaluation on real-world dataset indicated that SSCF improves recommendation accuracy.