Obtaining Calibrated Probabilities with Personalized Ranking Models

For personalized ranking models, the well-calibrated probability of an item being preferred by a user has great practical value. While existing work shows promising results in image classification, calibration not been much explored for ranking. In this paper, we aim to estimate calibrated how likely will prefer item. We investigate various parametric distributions and propose two methods, namely Gaussian Gamma calibration. Each proposed method can be seen as post-processing function that maps scores pre-trained models preference probabilities, without affecting recommendation performance. also design unbiased empirical risk minimization framework guides methods learning true from biased user-item interaction dataset. Extensive evaluations with on real-world datasets show both significantly improve