CREST
收藏资源简介:
CREST数据集由新加坡国立大学和南洋理工大学的研究人员创建,旨在支持约束时间线摘要任务。该数据集包含235个时间线,涉及47个公众人物或机构,每个实体有5个约束条件。数据集的内容来源于CNN Fast Facts和《卫报》的文章,通过GPT-4生成约束条件并由人工标注事件是否符合约束。数据集的创建过程包括约束生成、事件标注和事件过滤,确保了数据集的高质量和多样性。该数据集主要应用于个性化新闻摘要生成,帮助用户根据特定兴趣获取相关事件的时间线。
The CREST dataset was developed by researchers from the National University of Singapore and Nanyang Technological University to support the constrained timeline summarization task. It contains 235 timelines spanning 47 public figures and organizations, with 5 constraints per entity. The dataset content is sourced from CNN Fast Facts and articles from The Guardian. Constraints were generated using GPT-4, and human annotators verified whether events comply with these constraints. The dataset construction process includes constraint generation, event annotation, and event filtering, which guarantees its high quality and diversity. This dataset is primarily utilized for personalized news summarization, assisting users in acquiring timelines of events aligned with their specific interests.

- 1Just What You Desire: Constrained Timeline Summarization with Self-Reflection for Enhanced Relevance新加坡国立大学 · 2024年



