TweetQA

随着社交媒体越来越受欢迎，其中报道了大量新闻和实时事件，开发自动问答系统对于许多依赖实时知识的应用程序的有效性至关重要。虽然以前的问答 (QA) 数据集集中在新闻和维基百科等正式文本上，但我们展示了第一个针对社交媒体数据的 QA 的大规模数据集。为了使推文有意义并包含有趣的信息，我们收集了记者用来撰写新闻文章的推文。然后，我们要求人工注释者在这些推文上写下问题和答案。与 SQuAD 等其他 QA 数据集的答案是抽取式的不同，我们允许答案是抽象的。该任务需要模型阅读一条简短的推文和一个问题，并输出一个文本短语（不需要在推文中）作为答案。As social media grows increasingly popular and serves as a major source of news and real-time event coverage, developing automatic question answering (QA) systems is critical to the effectiveness of numerous applications that rely on real-time knowledge. While prior QA datasets have focused on formal texts such as news articles and Wikipedia, we present the first large-scale QA dataset tailored for social media data. To ensure the collected tweets are meaningful and contain valuable information, we gathered tweets that journalists use to draft news articles. Subsequently, we recruited human annotators to generate questions and answers based on these tweets. Unlike other QA datasets such as SQuAD where answers are extractive, our dataset permits abstractive answers. This task requires the model to read a short tweet and a corresponding question, then output a textual phrase that does not necessarily appear within the original tweet as the answer.

搜集汇总

背景与挑战

背景概述

TweetQA是一个针对社交媒体数据的大规模问答数据集，专注于推文内容，旨在支持自动问答系统处理实时事件。该数据集基于记者使用的推文，通过人工注释生成抽象式答案，要求模型根据推文和问题输出文本短语，由加州大学和IBM Research于2019年发布。