Year: 2025
Type: conference-paper
Abstract: Acquiring high-quality training data is a fundamental challenge in building accurate and robust machine learning models, particularly with the rise of foundational models. Existing online data marketplaces face limitations, including a lack of quality guarantees for purchased data and an overemphasis on sellers' benefits, resulting in unfair pricing and reduced buyer activity. To address these issues, we present FQora, a demonstration system for a fair and quality-based data marketplace. FQora introduces quality-constrained queries using quality-based pricing functions and effective quality assessment mechanisms. To ensure a balanced and sustainable market, FQora implements a mean-variance constraint and a novel Balanced Pareto Optimization to maximize utilities for both buyers and sellers. This system demonstrates advantages in usability, fairness, and market stability. The code of our prototype is open-sourced and available at https://github.com/Songyue-Guo/FQora_web_demo.
Authors: Songyue Guo, Maocheng Li, Zhao Chen, Caleb Chen Cao, Lei Chen
Language: English
Subfield: Computer Science Applications
Field: Computer Science
Domain: Physical Sciences
SDG: Reduced inequalities
Open Access status: gold
Awards: G1402809924, G380298712, G3952302843
OpenAlex ID: W4410636954