Interpolating Item and User Fairness in Multi-Sided Recommendations

Authors: Qinyi Chen, Jason Cheuk Nam Liang, Negin Golrezaei, Djallel Bouneffouf

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Via both theoretical analysis and a numerical case study on real-world data, we demonstrate the efficacy of our framework and method in maintaining platform revenue while ensuring desired levels of fairness for both items and users.
Researcher Affiliation Collaboration Qinyi Chen1 Jason Cheuk Nam Liang1 Negin Golrezaei1 Djallel Bouneffouf2 1Massachusetts Institute of Technology 2IBM Research {qinyic,jcnliang,golrezae}@mit.edu djallel.bouneffouf@ibm.com
Pseudocode Yes Algorithm 1 Fair Online Recommendation Algorithm for Multi-Sided Platforms (FORM)
Open Source Code Yes We have provided code and data for our numerical experiments in the supplementary materials.
Open Datasets Yes We use an Amazon review dataset [71] from the Clothing, Shoes and Jewelry category.
Dataset Splits No The paper describes an online learning setting with a total simulation length (T = 2000 rounds) where data is learned sequentially. It does not specify conventional fixed training, validation, or test dataset splits in percentages or counts.
Hardware Specification Yes All algorithms were implemented in Python 3.7 and run on a Mac Book with a 1.4 GHz Quad-Core Intel Core i5 processor.
Software Dependencies No The paper mentions 'Python 3.7' and 'CBC solver accessed via the Pu LP library in Python'. While Python has a version, specific version numbers for the CBC solver or Pu LP library are not provided, which is required for a reproducible description of ancillary software.
Experiment Setup Yes Users are classified into M = 5 types using matrix factorization and k-means clustering on user feature vectors. ... Item revenues ri are uniformly drawn from [0.5, 1.5]. ... Upon arrival, each type-j user is shown an assortment S of up to K = 3 items.