Related Experiment Videos
Label Distribution Enhancement-Based Label Completion for High-Noise-Ratio Crowdsourcing
IEEE Transactions on Pattern Analysis and Machine Intelligence
|August 11, 2026
Summary
This study introduces a novel label completion algorithm (LDELC) to address sparse crowdsourced data with high noise. LDELC effectively enhances label distributions from low-quality workers, improving data preprocessing for better downstream integration.
Area of Science:
- Data Science
- Machine Learning
- Artificial Intelligence
Background:
- Label completion is crucial for sparse crowdsourced data.
- Existing methods struggle with high-noise environments.
Purpose of the Study:
- Propose a novel label completion algorithm (LDELC) for high-noise crowdsourcing.
- Improve label integration effectiveness in challenging data scenarios.
Main Methods:
- Developed a worker quality estimation method.
- Implemented class probability estimation for high-quality workers.
- Enhanced label distributions for low-quality workers using a novel enhancement method.
- Propagated label distributions to convergence for missing label completion.
Main Results:
- LDELC demonstrates effectiveness on real-world and simulated high-noise datasets.
- The proposed method successfully handles sparse crowdsourced label matrices.
- Improved downstream label integration compared to existing approaches.
Conclusions:
- LDELC is an effective solution for label completion in high-noise crowdsourcing.
- The algorithm enhances data quality for machine learning tasks.
- Addresses a significant limitation in current label completion techniques.
