FETA

FETA数据集是由加利福尼亚大学圣巴巴拉分校的研究团队开发的，用于评估开放领域对话中少量样本任务迁移的性能。该数据集包含两个基础对话集，分别有10个和7个任务被标注，支持对数据集内任务迁移的研究。FETA数据集涵盖了多种对话属性（如双人对话与多人对话、匿名与重复说话者、不同对话长度）和任务类型（如话语级分类、对话级分类、跨度提取、多选题），并保持了数据量的多样性。此数据集不仅支持任务迁移的研究，还可用于研究预训练数据集和模型架构的效率与泛化能力，以及如持续学习和多任务学习等学习设置。

The FETA dataset was developed by a research team from the University of California, Santa Barbara, to evaluate the performance of few-shot task transfer in open-domain dialogues. This dataset contains two foundational dialogue corpora, with 10 and 7 annotated tasks respectively, supporting research on intra-dataset task transfer. The FETA dataset covers a variety of dialogue attributes (e.g., two-party and multi-party dialogues, anonymous and repeated speakers, varying dialogue lengths) and task types (e.g., utterance-level classification, dialogue-level classification, span extraction, multiple-choice questions), while maintaining diverse data scales. Beyond supporting research on task transfer, this dataset can also be used to investigate the efficiency and generalization capabilities of pre-trained datasets and model architectures, as well as learning settings such as continual learning and multi-task learning.