Current browse context:
stat.ML
Change to browse by:
References & Citations
Statistics > Machine Learning
Title: Human Interaction with Recommendation Systems: On Bias and Exploration
(Submitted on 1 Mar 2017 (v1), revised 3 Mar 2017 (this version, v2), latest version 28 Mar 2018 (v3))
Abstract: Recommendation systems rely on historical user data to provide suggestions. We propose an explicit and simple model for the interaction between users and recommendations provided by a platform, and relate this model to the multi-armed bandit literature. First, we show that this interaction leads to a bias in naive estimators due to selection effects. This bias leads to suboptimal outcomes, which we quantify in terms of linear regret. We end the first part by discussing ways to obtain unbiased estimates. The second part of this work considers exploration of alternatives. We show that although agents are myopic, agents' heterogeneous preferences ensure that recommendation systems 'learn' about all alternatives without explicitly incentivizing this exploration. This work provides new and practical insights relevant to a wide range of systems designed to help users make better decisions.
Submission history
From: Sven Schmit [view email][v1] Wed, 1 Mar 2017 22:28:42 GMT (582kb,D)
[v2] Fri, 3 Mar 2017 01:25:51 GMT (563kb,D)
[v3] Wed, 28 Mar 2018 17:40:02 GMT (887kb,D)
Link back to: arXiv, form interface, contact.