SciTweets - A Dataset and Annotation Framework for Detecting Scientific Online Discourse

This repository contains an expert-annotated dataset of 1261 tweets and the corresponding annotation framework from the publication "SciTweets - A Dataset and Annotation Framework for Detecting Scientific Online Discourse" (arxiv.org/abs/2206.07360). The tweets are annotated with three different categories of science-relatedness: (1) Scientific knowledge (scientifically verifiable claims): Tweets that include a claim or a question that could be scientifically verified, (2) Reference to scientific knowledge: Tweets that include at least one reference to scientific knowledge (references can either be direct, e.g., DOI, title of a paper or indirect, e.g., a link to an article that includes a direct reference), and (3) Related to scientific research in general: Tweets that mention a scientific research context (e.g., mention a scientist, scientific research efforts, research findings).

Further, the annotations include the annotators' confidence scores as well as labels for compound claims and ironic tweets.