CC25k: A Citation Contexts Dataset for Reproducibility-Oriented Sentiment Analysis

The CC25k dataset is unique in its focus on reproducibility-oriented sentiments (ROS) within scientific literature. This introduces a novel approach to studying computational reproducibility by leveraging citation contexts, which are textual fragments in scientific papers that reference prior work. This dataset comprises 25,829 labeled citation contexts from scientific literature published at AI venues, each annotated with one of three ROS labels: `positive`, `negative`, or `neutral`. These labels reflect the cited work's perceived reproducibility. The dataset contains ROS labeled contexts along with metadata about the workers, reproducibility study, related original paper, and citing paper, and the final aggregated labels.