Consumer-side Fairness in Recommender Systems: A Systematic Survey of Methods and Evaluation

TL;DR

The paper addresses consumer-side fairness in recommender systems by proposing a high-level fairness interpretation taxonomy and a structured methodology for surveying the literature. It catalogs a wide range of approach families, including pre-, in-, and post-processing, as well as various model types and fairness incorporation strategies, with detailed mappings to metrics and datasets. The authors highlight the lack of consensus in fairness definitions and evaluation, advocate for consolidated concepts and standardized metrics, and discuss the need for more diverse benchmarks and online evaluations. The work advances practical guidance for designing fair recommendations and sets the stage for more rigorous comparative studies and reproducibility in the field.

Abstract

In the current landscape of ever-increasing levels of digitalization, we are facing major challenges pertaining to scalability. Recommender systems have become irreplaceable both for helping users navigate the increasing amounts of data and, conversely, aiding providers in marketing products to interested users. The growing awareness of discrimination in machine learning methods has recently motivated both academia and industry to research how fairness can be ensured in recommender systems. For recommender systems, such issues are well exemplified by occupation recommendation, where biases in historical data may lead to recommender systems relating one gender to lower wages or to the propagation of stereotypes. In particular, consumer-side fairness, which focuses on mitigating discrimination experienced by users of recommender systems, has seen a vast number of diverse approaches for addressing different types of discrimination. The nature of said discrimination depends on the setting and the applied fairness interpretation, of which there are many variations. This survey serves as a systematic overview and discussion of the current research on consumer-side fairness in recommender systems. To that end, a novel taxonomy based on high-level fairness interpretation is proposed and used to categorize the research and their proposed fairness evaluation metrics. Finally, we highlight some suggestions for the future direction of the field.

Figures (4)

• Figure 1: A PRISMA flow diagram illustrating the full selection process.
• Figure 2: The proposed taxonomy based on Fairness Interpretation.
• Figure 3: Diagram that illustrates the high-level differences between the three non-Custom Fairness Interpretations in a scenario where the sensitive groups $s_1$ and $s_2$ display different preferences and the base recommender perform better for $s_1$. The preferences and recommendations given to the groups are illustrated as probability distributions, while model representations are projected into two-dimensional scatterplots. The Recommendation Parity interpretation idealizes when the recommendation distributions overlap, while a Utility-Based interpretation requires that the respective recommendation distributions match and mismatch the "true" distributions equally. The Neutral Representation interpretation is optimized to move from the case where representations of different groups can be separated into distinct clusters to the case where the clusters overlap or are indistinguishable.
• Figure 4: Fairness Incorporation categories.