A Session-Based Song Recommendation Approach Involving User Characterization along the Play Power-Law Distribution

Table 1

Error rates of the methods involved in the study.

Rating	User attribute K-NN UPC
	RMSE	MAE	NMAE
Play count-based	0.701 ± 0.007	0.539 ± 0.004	0.135 ± 0.001
Session-based (α = 0.5)	0.693 ± 0.009	0.497 ± 0.006	0.124 ± 0.001
Session-based (α = 0.7)	0.691 ± 0.010	0.457 ± 0.006	0.114 ± 0.002
Session-based (α = 0.9)	0.707 ± 0.007	0.533 ± 0.005	0.133 ± 0.001
	User K-NN cosine distance
	RMSE	MAE	NMAE
Play count-based	0.771 ± 0.007	0.583 ± 0.004	0.146 ± 0.001
Session-based (α = 0.5)	0.760 ± 0.009	0.535 ± 0.006	0.134 ± 0.001
Session-based (α = 0.7)	0.743 ± 0.011	0.484 ± 0.006	0.121 ± 0.002
Session-based (α = 0.9)	0.778 ± 0.008	0.576 ± 0.005	0.144 ± 0.002
	User K-NN Pearson coefficient
	RMSE	MAE	NMAE
Play count-based	0.756 ± 0.007	0.572 ± 0.004	0.143 ± 0.001
Session-based (α = 0.5)	0.753 ± 0.009	0.528 ± 0.006	0.132 ± 0.001
Session-based (α = 0.7)	0.737 ± 0.012	0.479 ± 0.006	0.120 ± 0.002
Session-based (α = 0.9)	0.768 ± 0.007	0.567 ± 0.005	0.142 ± 0.002
	Matrix factorization
	RMSE	MAE	NMAE
Play count-based	0.955 ± 0.007	0.787 ± 0.006	0.197 ± 0.002
Session-based (α = 0.5)	0.784 ± 0.009	0.583 ± 0.006	0.146 ± 0.002
Session-based (α = 0.7)	0.723 ± 0.012	0.492 ± 0.006	0.123 ± 0.002
Session-based (α = 0.9)	0.854 ± 0.009	0.671 ± 0.006	0.168 ± 0.002
	Biased matrix factorization
	RMSE	MAE	NMAE
Play count-based	0.859 ± 0.004	0.704 ± 0.004	0.176 ± 0.001
Session-based (α = 0.5)	0.742 ± 0.009	0.564 ± 0.006	0.141 ± 0.001
Session-based (α = 0.7)	0.724 ± 0.011	0.533 ± 0.006	0.133 ± 0.002
Session-based (α = 0.9)	0.787 ± 0.009	0.619 ± 0.006	0.155 ± 0.001