Modeling and counteracting exposure bias in recommender systems.

Abstract

What we discover and see online, and consequently our opinions and decisions, are becoming increasingly affected by automated machine learned predictions. Similarly, the predictive accuracy of learning machines heavily depends on the feedback data that we provide them. This mutual influence can lead to closed-loop interactions that may cause unknown biases which can be exacerbated after several iterations of machine learning predictions and user feedback. Machine-caused biases risk leading to undesirable social effects ranging from polarization to unfairness and filter bubbles.In this paper, we study the bias inherent in widely used recommendation strategies such as matrix factorization. Then we model the exposure that is borne from the interaction between the user and the recommender system and propose new debiasing strategies for these systems.Finally, we try to mitigate the recommendation system bias by engineering solutions for several state of the art recommender system models.Our results show that recommender systems are biased and depend on the prior exposure of the user. We also show that the studied bias iteratively decreases diversity in the output recommendations. Our debiasing method demonstrates the need for alternative recommendation strategies that take into account the exposure process in order to reduce bias.Our research findings show the importance of understanding the nature of and dealing with bias in machine learning models such as recommender systems that interact directly with humans, and are thus causing an increasing influence on human discovery and decision making.

İşbirlikçi Filtreleme Algoritmalarının Çok-Beğenilen Ürünlere Yönelik Yanlılığı

“…Bu veri kümelerinin ayrıntılı özellikleri Tablo 1'de sunulmaktadır. Bununla birlikte, bu veri kümelerinin, öneri sistemleri literatüründeki potansiyel yanlılık sorunları üzerine yapılan çalışmalarda sıklıkla kullanılması nedeniyle bu çalışmanın amacına uygun olduğu söylenebilir [8,12,37…”

A Graph-based Approach for Mitigating Multi-sided Exposure Bias in Recommender Systems

“…An exposure for an item is the percentage of the times it has appeared in the recommendations [25,40]. Recommendation algorithms are often biased towards more popular items giving them more exposure than many other items.…”

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

“…Within this scope, previous studies have tried to estimate the probability that the user is exposed to an item before rating it, and then use this estimate to alter the training algorithm by designing better estimators of the performance of the algorithm [5,24,27,34] or using regularization techniques [19].…”

“…[15] and [25] also used Markov models to model the closed loop. Other studies [11,19] tried to empirically model the iterative bias by using simulations that study the effect of the algorithms on various diversity metrics. Furthermore, [28] tried to deconvolve the feedback loop to understand how it affects the final ratings Although these studies consider the temporal dependency in recommender systems, a simplified theoretical modeling of how the system evolves has been missing.…”

“…Input: Ratings S t , Genres, number of runs nruns, number of iterations, iterations Output: For this reason, we use a semi-synthetic data using the Movielens 1M dataset, as was done in [19,27]. The process to construct the dataset is done by running a matrix completion algorithm on the original dataset to generate a complete rating matrix.…”