BPR

BPR: Bayesian Personalized Ranking from Implicit Feedback

Steffen Rendle, Christoph Freudenthaler, Zeno Gantner and Lars Schmidt-Thieme · UAI 2009

Read paper · arXiv:1205.2618

Learn each user's ranking from observed-versus-unobserved item preferences.

What you will implement

Per-user negative candidates, BPR-MF pairwise objective/global prior and 120 actual synthetic training updates; no LearnBPR sampler or paper dataset benchmark.

• 68. BPR: Implicit Feedback and Personalized Pairwise Ranking

  Select unobserved items from complete user histories and train both factor tables with differentiable pairwise preferences.

  Per-user candidate masks, stable BPR-MF negative log likelihood and full-table L2 prior; triple construction and tiny training loop are provided.