Deviation-Based Learning: Training Recommender Systems Using Informed User Choice

TL;DR

This paper introduces deviation-based learning for recommender systems, where the system learns from user actions and improves by selectively abstaining from recommendations to better capture user knowledge and enhance social welfare.

Contribution

It presents a novel deviation-based learning method that accounts for informed user choices and proposes abstention strategies to improve learning and social outcomes.

Findings

Learning stalls if recommendations are always made

Selective abstention enhances learning rate

Social welfare improves with abstention strategy

Abstract

This paper proposes a new approach to training recommender systems called deviation-based learning. The recommender and rational users have different knowledge. The recommender learns user knowledge by observing what action users take upon receiving recommendations. Learning eventually stalls if the recommender always suggests a choice: Before the recommender completes learning, users start following the recommendations blindly, and their choices do not reflect their knowledge. The learning rate and social welfare improve substantially if the recommender abstains from recommending a particular choice when she predicts that multiple alternatives will produce a similar payoff.