PolicyQA

隐私政策文件冗长而冗长。问答 (QA) 系统可以
帮助用户查找信息
对他们相关且重要。先前的研究
在此域中，将 QA 任务框架为检索
给出问题的政策文件中最相关的文本片段或句子列表。相反，我们认为提供
策略文档中文本跨度较短的用户减少了从冗长文本段中搜索目标信息的负担。
在本文中，我们提出了 PolicyQA，一个数据集
包含 25,017 个阅读理解
从 115 个网站隐私政策的现有语料库中精选的样式示例。政策QA
为广泛的隐私实践提供 714 个人工注释问题。我们
评估两个现有的神经 QA 模型和
执行严格的分析以揭示 PolicyQA 提供的优势和挑战

Privacy policy documents are notoriously lengthy and cumbersome. Question Answering (QA) systems can assist users in locating information that is relevant and important to them. Prior research in this domain frames the QA task as retrieving a list of the most relevant text snippets or sentences from policy documents for a given query. In contrast, we argue that providing users with shorter text spans from policy documents reduces the burden of searching for target information within lengthy text passages. In this paper, we present PolicyQA, a dataset containing 25,017 reading comprehension-style examples curated from an existing corpus of privacy policies across 115 websites. PolicyQA includes 714 manually annotated questions covering a wide range of privacy practices. We evaluate two existing neural QA models and conduct rigorous analyses to uncover the advantages and challenges that PolicyQA presents.

搜集汇总

背景与挑战

背景概述

PolicyQA是一个专注于隐私政策文件的阅读理解数据集，包含25,017个示例，旨在通过问答系统帮助用户高效获取信息。该数据集由加州大学于2020年发布，提供了丰富的隐私政策问答资源。