CPFair: Personalized Consumer and Producer Fairness Re-ranking for Recommender Systems

Abstract

Recently, there has been a rising awareness that when machine learning (ML) algorithms are used to automate choices, they may treat/affect individuals unfairly, with legal, ethical, or economic consequences. Recommender systems are prominent examples of such ML systems that assist users in making high-stakes judgments.A common trend in the previous literature research on fairness in recommender systems is that the majority of works treat user and item fairness concerns separately, ignoring the fact that recommender systems operate in a two-sided marketplace. In this work, we present an optimization-based re-ranking approach that seamlessly integrates fairness constraints from both the consumer and producer-side in a joint objective framework. We demonstrate through large-scale experiments on 8 datasets that our proposed method is capable of improving both consumer and producer fairness without reducing overall recommendation quality, demonstrating the role algorithms may play in minimizing data biases. CCS CONCEPTS• Information systems → Recommender systems.

PyCPFair: A framework for consumer and producer fairness in recommender systems

“…However, majority of prior research on fairness optimization algorithms in RS have adopted a unilateral consumer-or producer-centric approach, ignoring the reality that recommendations are made in a multi-stakeholder environment. By employing the consumer-producer fairness technique offered by the PyCPFair [23], recommendations can be evaluated and becomes fair from the viewpoints/interests of the various stakeholders. As an example, Rahmani et al [26] used PyCPFair to evaluate the generalizability of user-oriented fairness (C-fairness) in RS based on different user grouping methods on several domains of recommendation when deployed in a multi-stakeholder platform (e.g., E-commerce, POI, etc).…”

“…As an example, Rahmani et al [26] used PyCPFair to evaluate the generalizability of user-oriented fairness (C-fairness) in RS based on different user grouping methods on several domains of recommendation when deployed in a multi-stakeholder platform (e.g., E-commerce, POI, etc). Naghiaei et al [23] used PyCPFair to provide a fair recommendation concerning both consumers and producers. As a future development, it is possible to incorporate other notion of fairness (such as calibration based on item aspects) and recommendation novelty/diversity.…”

“…In recent years, there has been a surge of interest in research on fair recommendation [11,15]. This research covers a wide range of topics, including the impact of data characteristics [8,22] and biases [25,4,24] in recommendation performance (accuracy, fairness), developing evaluation frameworks to measure recommendation unfairness [6,7,21], mitigating existing data biases [22], and enhancing outcome fairness [23,20,32]. Fairness in recommender system (RS) is a multi-faceted concept depending on stakeholder, type of benefit, context, morality, and time.…”

“…In the current work accepted to SIGIR'22 [23], we introduce PyCPFair, a Python-based framework that enables scholars and practitioners to examine and enhance the fairness of a recommender system from the perspective of both key stakeholders (consumer and producer) and their interest benefits (accuracy and exposure). Different from existing open-source libraries (e.g.…”

“…In particular, PyCPFair will provide a model-agnostic re-ranking approach capable of jointly or unilaterally optimizing the marketplace objectives with respect to consumers (C), Providers (P), and both (CP) without compromising the recommendation overall accuracy. The main advantages of the PyCPFair are that it is fast and accurate, lever- aging Gurobi 1 solver and MIP 2 interface which makes it easier for future development, and its performance is evaluated through extensive experiments over eight real-world datasets from diverse domains [23,26].…”