Doc2Dial (Doc2Dial: Document-grounded Dialogue)

对于面向目标的以文档为基础的对话，它通常涉及识别最相关信息的复杂上下文，这需要更好地理解对话和文档之间的相互关系。同时，许多面向用户的在线文档同时使用半结构化和非结构化内容来引导用户访问不同上下文的信息。因此，我们创建了一个新的面向目标的基于文档的对话数据集，该数据集捕获了来自多个域（如 ssa.gov 和 studentaid.gov）的各种文档内容的更多样化的场景。对于数据收集，我们提出了一种用于对话数据构建的新型管道方法，该方法已针对多个领域进行了调整和评估。

Goal-oriented document-grounded conversations typically involve complex contexts for identifying highly relevant information, demanding a deeper comprehension of the interplays between the dialogue and its supporting documents. Meanwhile, most user-facing online documents utilize both semi-structured and unstructured content to guide users to access information across diverse contexts. Therefore, we construct a novel goal-oriented document-grounded conversation dataset that captures more diverse scenarios with varied document contents from multiple domains, such as ssa.gov and studentaid.gov. For data collection, we propose a novel pipeline approach for dialogue data construction, which has been adapted and evaluated across multiple domains.

搜集汇总

背景与挑战

背景概述

Doc2Dial是一个面向目标的文档对话数据集，旨在通过捕获多个领域的文档内容来促进对话与文档相互关系的理解。它采用新型管道方法构建，涵盖了如ssa.gov和studentaid.gov等多样化的场景。