BEGIN
收藏资源简介:
BEGIN数据集是由阿尔伯塔大学、谷歌研究院和纽约大学合作创建的,包含12000个对话回合,这些对话由神经对话系统生成,训练于三个知识驱动的对话语料库。该数据集用于评估对话系统生成响应的可归因性,即响应是否能够完全由提供的背景信息支持。BEGIN数据集的应用领域主要集中在提高知识驱动对话系统的评估标准,旨在解决现有评估指标在区分可归因和不可归因响应方面的不足,推动开发更复杂和鲁棒的评估指标。
The BEGIN Dataset was collaboratively created by the University of Alberta, Google Research, and New York University. It contains 12,000 dialogue turns generated by neural dialogue systems trained on three knowledge-driven dialogue corpora. This dataset is utilized to evaluate the attributability of responses generated by dialogue systems, i.e., whether a response can be fully supported by the provided background information. The primary application scenario of the BEGIN Dataset is to improve the evaluation benchmarks for knowledge-driven dialogue systems, aiming to address the shortcomings of existing evaluation metrics in distinguishing between attributable and non-attributable responses, and promote the development of more sophisticated and robust evaluation metrics.




