Long-term Off-Policy Evaluation and Learning

arXiv:2404.15691v1 Announce Type: new
Abstract: Short- and long-term outcomes of an algorithm often differ, with damaging downstream effects. A known example is a click-bait algorithm, which may increase short-term clicks but damage long-term user engagement. A possible solution to estimate the long-term outcome is to run an online experiment or A/B test for the potential algorithms, but it takes months or even longer to observe the long-term outcomes of interest, making the algorithm selection process unacceptably slow. This work thus …

abstract algorithm arxiv b test click cs.lg effects engagement evaluation example experiment long-term policy solution stat.ml test type user engagement