A collection of news articles and weak labels used for training fake news detection models via reinforcement learning as presented in the AAAI 2020 paper. It provides a collection of news content paired with noisy labels from multiple sources to facilitate research in weak supervision and label denoising.
Use Cases
- Train a text classifier using the news content and weak labels to identify misinformation
- Develop a reinforcement learning agent to filter noisy labels based on source reliability features
- Benchmark weak supervision algorithms against the provided ground truth labels for fake news classification
Strengths
- Includes news articles and metadata sourced from platforms like PolitiFact and GossipCop
- Features weak labels generated from source-level and user-level heuristics
- Designed specifically for reinforcement learning-based label weighting and selection tasks