Theoretical Modeling of the Iterative Properties of User Discovery in a Collaborative Filtering Recommender System

Abstract

The closed feedback loop in recommender systems is a common setting that can lead to different types of biases. Several studies have dealt with these biases by designing methods to mitigate their effect on the recommendations. However, most existing studies do not consider the iterative behavior of the system where the closed feedback loop plays a crucial role in incorporating different biases into several parts of the recommendation steps. We present a theoretical framework to model the asymptotic evolution of the different components of a recommender system operating within a feedback loop setting, and derive theoretical bounds and convergence properties on quantifiable measures of the user discovery and blind spots. We also validate our theoretical findings empirically using a real-life dataset and empirically test the efficiency of a basic exploration strategy within our theoretical framework. Our findings lay the theoretical basis for quantifying the effect of feedback loops and for designing Artificial Intelligence and machine learning algorithms that explicitly incorporate the iterative nature of feedback loops in the machine learning and recommendation process.

Debiased Explainable Pairwise Ranking from Implicit Feedback

“…Bias in recommendation can be categorized into seven types [6] that occur within the various stages of the recommendation feedback loop [23,24,32,42] between the user, the data, and the model. Among these categories, in the user-to-data phase, we find exposure bias, which is the focus of our work in this paper.…”

Fairness in Information Access Systems

“…ese di culties arise from a number of di erences, including the addition of multiple classes of stakeholders (Burke, 2017), the rivalrous nature of allocating retrieval opportunities (Biega et al, 2018;Diaz et al, 2020), and the immediacy of the interactive feedback loop (Chaney et al, 2018;Khenissi et al, 2020); classi cation-oriented fairness de nitions are not necessarily well-suited to assessing these situations.…”

Position Paper on Simulating Privacy Dynamics in Recommender Systems

“…In particular, [4,9] show that recommender systems tend to focus on a small set of documents when generating recommendations. This means that the evolution of the recommender system causes fewer documents to be covered.…”

“…Plus, [9] provides a theoretical framework to model the recommender system's iterative behavior to investigate the presence of different biases. In the same vein, [14] analyze the emergence of popularity bias via recommender system simulations.…”