Optimizing Intervention Via Offline Policy Evaluation-Segal
#notesFromPaper Year : Tags : #cairnsRL citizen science - engagement Authors: Segal Gal Kamar Horvitz Miller
Their previous work here: Intervention_Strategies_For_Engagement-Segal_Gal
They used an ML model to predict disengagement, then threw a random message at players. It’s not fully clear yet what the difference is here, but this one appears to optimize across message types and its reward function optimizes for longer term effects.
So my shitty understanding of this paper is that their second model used data from random interventions to learn what kind of interventions were best
