Title:Towards Sharing Task Environments to Support Reproducible Evaluations of Interactive Recommender Systems

Abstract:Beyond sharing datasets or simulations, we believe the Recommender Systems (RS) community should share Task Environments. In this work, we propose a high-level logical architecture that will help to reason about the core components of a RS Task Environment, identify the differences between Environments, datasets and simulations; and most importantly, understand what needs to be shared about Environments to achieve reproducible experiments. The work presents itself as valuable initial groundwork, open to discussion and extensions.