Beyond-Accuracy Provider Fairness (Slides)
0 likes52 views
This study addresses the issue of popularity bias in recommendation systems, which leads to unfair treatment of providers offering niche content. It focuses on the trade-offs between enhancing fairness and maintaining accuracy in recommendations, proposing a method inspired by post-processing techniques. The research analyzes these trade-offs across various algorithms and datasets to improve provider fairness without sacrificing recommendation quality.
Beyond-Accuracy Provider Fairness (Slides)
- 1. Provider Fairness and Beyond-Accuracy Trade-offs in Recommender Systems Saeedeh Karimi1 , Hossein A. Rahmani2 , Mohammadmehdi Naghiaei3 , Leila Safari1 1 University of Zanjan, Iran 2 University College London, UK 3 University of Southern California, US The 6th FAccTRec Workshop: Responsible Recommendation The 17th ACM Conference on Recommender Systems Singapore, 18th-22nd September 2023
- 2. Introduction ● Recommendation systems often suffer from popularity bias, a phenomenon where popular items are disproportionately recommended at the expense of less popular or long-tail items ● Popularity bias leads to a skewed exposure of items and potentially unfair treatment of providers, particularly those offering niche or less popular content. ● Popularity bias may adversely affect the diversity, novelty, and serendipity of recommendations, thereby limiting users’ experiences and the discovery of new content.
- 3. Contribution ● Past research in recommendation systems mainly emphasized accuracy over fairness considerations. ● Our study explores the trade-offs involved in enhancing fairness in recommendation systems. ● We introduce a method for improving fairness while preserving accuracy, inspired by post-processing techniques. ● We extensively analyze the trade-offs between provider fairness and recommendation quality across four algorithms and datasets in various domains.
- 5. Datasets
- 6. Results
- 9. Thanks! Saeedeh Karimi (karimi.saeedeh@znu.ac.ir) Hossein A. Rahmani (hossein.rahmani.22@ucl.ac.uk) Mohammadmehdi Naghiaei (naghiaei@usc.edu) Leila Safari (lsafari@znu.ac.ir) Code: https://github.com/rahmanidashti/BeyondAccProvider Do you have any questions?