Stanford Schema2QA Dataset

“Schema2QA 是第一个基于真实 Schema.org 数据的大型问答数据集。它涵盖 6 个常见领域：餐馆、酒店、人物、电影、书籍和音乐，基于从 6 个不同的爬取的 Schema.org 元数据网站（Yelp、Hyatt、LinkedIn、IMDb、Goodreads 和 last.fm。）。总共有超过 2,000,000 个用于训练的示例，包括增强的人类释义数据和 Genie 生成的高质量合成数据。所有问题都是使用可执行的虚拟助手编程语言 ThingTalk 进行注释。Schema2QA 包括从众包工人那里收集的具有挑战性的评估问题。工人只被提示领域是什么以及支持哪些属性。因此，句子自然而多样。它们还包含训练期间看不见的实体. 收集的句子由作者手动使用 ThingTalk 进行注释。总共有超过 5,000 个示例用于开发和测试。

Schema2QA is the first large-scale question answering dataset built on real Schema.org data. It covers six common domains: restaurants, hotels, people, movies, books, and music, sourced from six crawled Schema.org metadata websites including Yelp, Hyatt, LinkedIn, IMDb, Goodreads, and last.fm. There are over 2,000,000 training examples in total, comprising enhanced human paraphrase data and high-quality synthetic data generated by Genie. All questions are annotated using the executable virtual assistant programming language ThingTalk. Schema2QA includes challenging evaluation questions collected from crowdworkers, who were only prompted with the target domain and its supported attributes. Consequently, the sentences are natural and diverse, and they also contain entities unseen during training. The collected sentences were manually annotated by the authors using ThingTalk. In total, there are over 5,000 examples reserved for development and testing.

搜集汇总

背景与挑战

背景概述

Stanford Schema2QA Dataset是一个基于真实Schema.org元数据的大型问答数据集，覆盖餐馆、酒店等6个常见领域。它包含超过200万训练示例，使用ThingTalk进行注释，并提供了5000多个用于评估的多样化问题。