Dataset Card for CLASum Dataset

Dataset Summary

The CLASum is built for abstractivity characterization and measurement in the summarization task. It contains 200 document-summary pairs extracted from CNN/DailyMail and XSUM datasets. Each summary has been annotated for 11 summary-related questions by 3 different annotators.

Data Fields

• id: Sample identifier
• source: Dataset of the document-summary pair
• source_id: Identifier of the pair in the original dataset
• document: The document to summarize
• summary: A summary for the document
• annotator: Identifier of the annotator
• A: Answer for a question related to relevance of the information in the summary
• B: Answer for a question related to amount of novel information within the summary
• C: Answer for a question related to the level of abstractivity
• D: Answer for a question related to the content exclusion action
• E: Answer for a question related to the information melting action
• F: Answer for a question related to the syntaxis alteration action
• G: Answer for a question related to the synonym action
• H: Answer for a question related to the generalization action
• I: Answer for a question related to the specification action
• J: Answer for a question related to the content reordering action

Additional Information

Licensing Information

The CLASum dataset is released under the Apache-2.0 License.

Citation Information

@Unpublished{Ahuir2025,
  author = {Ahuir, Vicent and Castro-Bleda, María José and Hurtado, Lluís-F},
  date   = {2025-10},
  title  = {Beyond Surface Metrics: Modeling Abstractivity in Summarization via Action-Based Spectrums.},
}