Prosocial Conversations for Bridging Benchmark Dataset
收藏资源简介:
Prosocial Conversations for Bridging Benchmark Dataset是由佛罗里达大学、SIFT和谷歌Jigsaw合作创建的一个数据集,包含11,973条来自Civil Comments的评论。该数据集专注于标记与建立桥梁相关的社会属性,如疏离、同情、推理、好奇、道德愤怒和尊重。数据集的创建过程采用了创新的、迭代式的注释者参与方法,通过与七名美国注释者的深入合作,不断优化注释定义和过程。该数据集旨在提高机器学习模型对复杂社会概念的理解和处理能力,特别是在理解和促进建设性对话方面。
The Prosocial Conversations for Bridging Benchmark Dataset was collaboratively developed by the University of Florida, SIFT, and Google Jigsaw. It comprises 11,973 comments sourced from the Civil Comments corpus. This dataset centers on annotating socially significant attributes associated with bridge-building discourse, including alienation, sympathy, reasoning, curiosity, moral outrage, and respect. Its construction adopts an innovative, iterative annotator engagement methodology, which was refined through in-depth collaboration with seven U.S. annotators to optimize annotation definitions and procedures. This dataset aims to enhance the capacity of machine learning models to understand and handle complex social concepts, particularly in comprehending and facilitating constructive dialogues.




