The dataset contains gold-standard summary labels for 39 "CSI: Crime Scene Investigation" episodes from seasons 1-5. Each episode contains the full-length screenplay and human annotations for its summary. The annotations include:
- scene-level binary labels denoting whether the scene belongs to the summary of the episode
- aspect-based labels for the scenes that belong to the summary, i.e., which aspect of the summary the scene addresses (e.g., information about the victim, the crime scene, the perpetrator etc.)
- sentence-level binary labels denoting the sentences of the screenplay that belong to the summary for 10 episodes of the dataset