WITHDRAWN: Stacked Models and the Explainability Tradeoff in Recommender Systems

• Activity Number: 319 - SLDS CSpeed 6
• Type: Contributed
• Date/Time: Wednesday, August 11, 2021 : 3:30 PM to 5:20 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): shaudi mahdavi hosseini
• Companies: m.i.t.
• Keywords: explainability; recommender systems

Abstract

Recommender systems often involve a series stacked models, complicating attempts at explainability that do not compromise accuracy metrics, particularly as systems increasingly grow complex. In this work we explore the applicability of the best-known explainability techniques to stylized recommender systems in the healthcare information search domain. We use a theoretical derivation to capture a explainability and accuracy tradeoff, and show conditions when this tradeoff attenuates in the proposed stylized setting.