Scalable and Provably Fair Exposure Control for Large-Scale Recommender Systems (2024-02-22T00:00:00.000000Z)

TL;DR

This work develops a scalable, fast, and fair method called exposure-aware ADMM, based on implicit alternating least squares (iALS), a conventional scalable algorithm for collaborative filtering, but optimizes a regularized objective to achieve a flexible control of accuracy-fairness tradeoff.

Abstract

Typical recommendation and ranking methods aim to optimize the satisfaction of users, but they are often oblivious to their impact on the items (e.g., products, jobs, news, video) and their providers. However, there has been a growing understanding that the latter is crucial to consider for a wide range of applications, since it determines the utility of those being recommended. Prior approaches to fairness-aware recommendation optimize a regularized objective to balance user satisfaction and item fairness based on some notion such as exposure fairness. These existing methods have been shown to be effective in controlling fairness, however, most of them are computationally inefficient, limiting their applications to only unrealistically small-scale situations. This indeed implies that the literature does not yet provide a solution to enable a flexible control of exposure in the industry-scale recommender systems where millions of users and items exist. To enable a computationally efficient exposure control even for such large-scale systems, this work develops a scalable, fast, and fair method called exposure-aware ADMM (exADMM ). exADMM is based on implicit alternating least squares (iALS), a conventional scalable algorithm for collaborative filtering, but optimizes a regularized objective to achieve a flexible control of accuracy-fairness tradeoff. A particular technical challenge in developing exADMM is the fact that the fairness regularizer destroys the separability of optimization subproblems for users and items, which is an essential property to ensure the scalability of iALS. Therefore, we develop a set of optimization tools to enable yet scalable fairness control with provable convergence guarantees as a basis of our algorithm. Extensive experiments performed on three recommendation datasets demonstrate that exADMM enables a far more flexible fairness control than the vanilla version of iALS, while being much more computationally efficient than existing fairness-aware recommendation methods.

References63 items

Fast online ranking with fairness of exposure

Fair Ranking as Fair Division: Impact-Based Individual Fairness in Ranking

Optimizing Generalized Gini Indices for Fairness in Rankings

Revisiting Negative Sampling vs. Non-sampling in Implicit Recommendation

Fairness for Robust Learning to Rank

Two-sided fairness in rankings via Lorenz dominance

Revisiting the Performance of iALS on Item Recommendation Benchmarks

Negative Interactions for Improved Collaborative Filtering: Don’t go Deeper, go Higher

A Multi-Objective Optimization Framework for Multi-Stakeholder Fairness-Aware Recommendation

Computationally Efficient Optimization of Plackett-Luce Ranking Models for Relevance and Fairness

TFROM: A Two-sided Fairness-Aware Recommendation Model for Both Customers and Providers

Unbiased Pairwise Learning from Biased Implicit Feedback

Controlling Fairness and Bias in Dynamic Learning-to-Rank

Neural Collaborative Filtering vs. Matrix Factorization Revisited

ADMM SLIM: Sparse Recommendations for Many Users

Multistakeholder recommendation: Survey and research directions

Policy-Gradient Training of Fair and Unbiased Ranking Functions

Unbiased Recommender Learning from Missing-Not-At-Random Implicit Feedback

Policy Learning for Fairness in Ranking

Managing Popularity Bias in Recommender Systems with Personalized Re-ranking

Calibrated recommendations

Coupled Graphs and Tensor Factorization for Recommender Systems and Community Detection

Reducing Disparate Exposure in Ranking: A Learning To Rank Approach

Equity of Attention: Amortizing Individual Fairness in Rankings

Fairness of Exposure in Rankings

Variational Autoencoders for Collaborative Filtering

Exposure diversity as a design principle for recommender systems

Balanced Neighborhoods for Multi-sided Fairness in Recommendation

Constrained Tensor Factorization with Accelerated AO-ADMM

Controlling Popularity Bias in Learning-to-Rank Recommendation

Multisided Fairness for Recommendation

Beyond Parity: Fairness Objectives for Collaborative Filtering

Linearized ADMM for Nonconvex Nonsmooth Optimization With Convergence Analysis

Neural Collaborative Filtering

A Generic Coordinate Descent Framework for Learning from Implicit Feedback

The MovieLens Datasets: History and Context

LorSLIM: Low Rank Sparse Linear Methods for Top-N Recommendations

Distributed Stochastic ADMM for Matrix Factorization

The direct extension of ADMM for multi-block convex minimization problems is not necessarily convergent

Parallel Direction Method of Multipliers

Parallel Multi-Block ADMM with o(1 / k) Convergence

Parallel matrix factorization for recommender systems

Revisiting Frank-Wolfe: Projection-Free Sparse Convex Optimization

Bursting your (filter) bubble: strategies for promoting diverse exposure

Fairness-aware Learning through Regularization Approach

Distributed Optimization and Statistical Learning via the Alternating Direction Method of Multipliers

Goodbye Pareto Principle, Hello Long Tail: The Effect of Search Costs on the Concentration of Product Sales

Exploiting query reformulations for web search result diversification

Efficient Online and Batch Learning Using Forward Backward Splitting

Collaborative Filtering for Implicit Feedback Datasets

The long tail of recommender systems and how to leverage it

Large-Scale Parallel Collaborative Filtering for the Netflix Prize

Trust-aware recommender systems

Blockbuster Culture's Next Rise or Fall: The Impact of Recommender Systems on Sales Diversity

The Analysis of Permutations

On the measurement of inequality

An algorithm for quadratic programming

Adjustment of an Inverse Matrix Corresponding to a Change in One Element of a Given Matrix