StreamingQA
收藏资源简介:
StreamingQA是由DeepMind创建的大规模数据集,旨在研究问答模型如何适应不断变化的知识。该数据集包含10万个问题,这些问题是基于2007年至2020年间的时间戳新闻文章。数据集通过季度评估模型,以观察模型在阅读新文章时的适应性。StreamingQA特别关注于模型对频繁命名实体问题的处理能力,并展示了通过无监督微调或访问检索来提高问答性能的潜力。此外,数据集还强调了在动态世界中,模型需要不断更新以保持与最新信息的同步,从而提供更真实的评估环境。
StreamingQA is a large-scale dataset developed by DeepMind for researching how question answering (QA) models adapt to constantly evolving knowledge. The dataset includes 100,000 questions grounded in timestamped news articles spanning from 2007 to 2020. It enables quarterly evaluations of models to observe their adaptability when consuming newly published articles. StreamingQA specifically focuses on models' ability to handle questions involving frequently occurring named entities, and demonstrates the potential to improve QA performance through unsupervised fine-tuning or access to retrieval systems. Furthermore, the dataset highlights the necessity for continuous model updates to stay aligned with the latest information in a dynamic world, thereby creating a more realistic evaluation environment.

- 1StreamingQA: A Benchmark for Adaptation to New Knowledge over Time in Question Answering ModelsDeepMind, London, UK · 2022年



