aitetic/WikiDialog-OQ

--- configs: - config_name: default data_files: - split: train path: "WikiDialog-OQ.jsonl.gz" - split: validation path: "WikiDialog-OQ-validation.jsonl" --- # WikiDialog-OQ Dataset containing 11M information-seeking conversations from passages in English Wikipedia, publicly available. Each conversation was generated using the dialog inpainting method detailed in the paper using the Inpaint-OQ inpainter model, a T5-XXL model that was fine-tuned on OR-QuAC and QReCC using a dialog reconstruction loss. ### Abstract Many important questions (e.g. "How to eat healthier?") require conversation to establish context and explore in depth. However, conversational question answering (ConvQA) systems have long been stymied by scarce training data that is expensive to collect. To address this problem, we propose a new technique for synthetically generating diverse and high-quality dialog data: dialog inpainting. Our approach takes the text of any document and transforms it into a two-person dialog between the writer and an imagined reader: we treat sentences from the article as utterances spoken by the writer, and then use a dialog inpainter to predict what the imagined reader asked or said in between each of the writer's utterances. By applying this approach to passages from Wikipedia and the web, we produce WikiDialog and WebDialog, two datasets totalling 19 million diverse information-seeking dialogs---1,000x larger than the largest existing ConvQA dataset. Furthermore, human raters judge the answer adequacy and conversationality of WikiDialog to be as good or better than existing manually-collected datasets. Using our inpainted data to pre-train ConvQA retrieval systems, we significantly advance state-of-the-art across three benchmarks (QReCC, OR-QuAC, TREC CaST) yielding up to 40% relative gains on standard evaluation metrics. **Version**: * 1.0.0 (default): Initial release. **Examples**: * Train: 11,264,129 * Validation: 113,822 ``` @inproceedings{dai2022dialoginpainting, title={Dialog Inpainting: Turning Documents to Dialogs}, author={Dai, Zhuyun and Chaganty, Arun Tejasvi and Zhao, Vincent and Amini, Aida and Green, Mike and Rashid, Qazi and Guu, Kelvin}, booktitle={International Conference on Machine Learning (ICML)}, year={2022}, organization={PMLR} } ```--- 配置项： - 配置名称（config_name）: default 数据文件（data_files）: - 拆分集（split）: 训练集（train），路径: "WikiDialog-OQ.jsonl.gz" - 拆分集（split）: 验证集（validation），路径: "WikiDialog-OQ-validation.jsonl" --- # WikiDialog-OQ 本数据集包含源自英文维基百科段落的1100万条信息寻求型对话，已公开发布。所有对话均采用论文中详述的对话补全（dialog inpainting）方法生成，使用基于Inpaint-OQ补全器模型——该模型是在OR-QuAC与QReCC数据集上通过对话重构损失微调得到的T5-XXL大语言模型（Large Language Model）——完成构建。 ### 摘要 诸多重要问题（例如“如何健康饮食？”）均需借助对话搭建语境并开展深度探讨。然而，对话式问答（Conversational Question Answering, ConvQA）系统长期受制于训练数据稀缺且收集成本高昂的痛点。为解决该问题，我们提出了一种全新的合成式生成多样化高质量对话数据的技术：对话补全。我们的方法可将任意文档文本转换为作者与假想读者之间的双人对话：先将文章中的句子视作作者的发言内容，再利用对话补全模型预测作者每段发言之间假想读者的提问或发言内容。通过将该方法应用于维基百科与网络段落，我们构建了WikiDialog与WebDialog两个数据集，总计包含1900万条多样化的信息寻求型对话——规模为现有最大对话式问答数据集的1000倍。此外，人类评估者认为WikiDialog的回答充分性与对话流畅度与现有人工收集数据集相当，甚至更优。利用我们补全得到的数据对对话式问答检索系统进行预训练，可在QReCC、OR-QuAC与TREC CaST三个基准数据集上显著提升当前最优性能，在标准评估指标上实现最高40%的相对增益。 **版本**： * 1.0.0（默认版）：首次发布。 **示例数量**： * 训练集：11,264,129 * 验证集：113,822 @inproceedings{dai2022dialoginpainting, title={Dialog Inpainting: Turning Documents to Dialogs}, author={Dai, Zhuyun and Chaganty, Arun Tejasvi and Zhao, Vincent and Amini, Aida and Green, Mike and Rashid, Qazi and Guu, Kelvin}, booktitle={International Conference on Machine Learning (ICML)}, year={2022}, organization={PMLR} }