Evaluating Ranking Systems: From Offline Metrics to Online Interleaving
摘要
Evaluating improvements to ranking functions and recommendation algorithms requires combining the speed and safety of offline evaluation with the definitive, high-sensitivity results of online interleaving experiments.