Re$^3$Dial：用于长轮次开放域对话预训练的检索、重组和缩放对话语料库（英文版）

英文标题：Re$^3$Dial: Retrieve, Reorganize and Rescale Dialogue Corpus for Long-Turn Open-Domain Dialogue Pre-training中文摘要：本文提出了使用机器学习技术构建大规模多轮对话数据集及相应框架 Re$^3$Dial，可以将已有短轮对话数据转变成长轮对话数据，尤其适用于处理大规模开放域网络聊天数据，实验证明相比现有技术大幅提高了在不同预训练环境下的多轮对话建模能力。英文摘要：Large-scale open-domain dialogue data crawled from public social media hasgreatly improved the performance of dialogue models. However, long-turndialogues are still highly scarce. Specifically, most dialogue sessions inexisting corpora have less than three tu