How algorithmic confounding in recommendation systems increases homogeneity and decreases utility

By Allison June-Barlow Chaney, Brandon M Stewart, Barbara E. Engelhardt. 2017

397 citations recorded by the source corpus.

Recommendation systems are ubiquitous and impact many domains; they have the potential to influence product consumption, individuals' perceptions of the world, and life-altering decisions. These systems are often evaluated or trained with data from users already exposed to algorithmic recommendations; this creates a pernicious feedback loop. Using simulations, we demonstrate how using data confounded in this way homogenizes user behavior without increasing utility.

Subfields: Computer Science Applications, Information Systems, Management Science and Operations Research, Recommender Systems, AI Ethics, Fairness & Society

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Science Applications

In Information Systems

In Management Science and Operations Research

In Recommender Systems

In AI Ethics, Fairness & Society

Citations in the corpus

A Survey on Bias and Fairness in Machine Learning (2019, methodology): “y models this situation. This feedback loop is not only existent between the data and the algorithm, but also some work analyzed the existence of this loop between the algorithms and user interaction [28]. Getting inspiration from these papers, we also modeled categorization of bias definitions, as ”
Optimizing Long-term Social Welfare in Recommender Systems: A Constrained Matching Approach (2020, background): “Content recommender systems (RSs) , which match users to content (e.g., news, music, One aspect of “fairness” that has received little attention emerges when one considers the dynamics of the RS ecosystem .”
Explore, exploit, and explain: personalizing explainable recommendations with bandits (2018, background): “The problem of ignoring items or whole categories of items leads to filter bubble pathologies [3].”
Towards Intersectionality in Machine Learning: Including More Identities, Handling Underrepresentation, and Performing Evaluation (2022, background): “…for problems with intersectional concerns may be more akin to that of simulations as testbeds and tools for understanding sociotechical systems such as in modeling recommender systems [17, 21, 45, 70, 89] and online information diffusion [44, 48, 119], rather than as a source for more data.”
Disentangling User Interest and Conformity for Recommendation with Causal Embedding (2021, methodology): “Real-world recommender systems are often trained and updated continuously using real-time user interactions, which forms a feedback loop as illustrated in Figure 1, with training data and test data NOT independent and identically distributed (IID) [13].”
Causal Intervention for Leveraging Popularity Bias in Recommendation (2021, background): “The first type of work is on confounding effects [10, 13, 37, 44].”
Feedback Loop and Bias Amplification in Recommender Systems (2020, methodology): “] 25 Jul 2020 CIKM’20, October 19–23, 2020, Ireland Masoud Mansoury, et al. to their profiles as part of their interaction history. In this way, recommendations and user profiles form a feedback loop [4, 5]; the users and the system are in a process of mutual dynamic evolution where users profile ge”
Bias and Debias in Recommender System: A Survey and Future Directions (2020, background): “Many researchers also study the impact of feedback loop on the popularity bias [57], [58], [59].”
Data Feedback Loops: Model-driven Amplification of Dataset Biases (2022, background): “This issue arises due to the tension between improving ranking metrics and considerations of bias, fairness, or diversity (Steck, 2018; Chaney et al., 2018).”
Model-Agnostic Counterfactual Reasoning for Eliminating Popularity Bias in Recommender System (2020, background): “Real-world recommender systems are often trained and updated continuously using real-time user interactions with training data and test data NOT independent and identically distributed (nonIID) [11, 58].”