KILT
收藏资源简介:
KILT数据集是由Facebook AI Research创建的,旨在为知识密集型语言任务提供一个统一的基准。该数据集包含约320万条实例,所有数据均基于2019年8月的Wikipedia快照,确保了数据的一致性和可比性。KILT涵盖了五个不同的任务领域,包括事实检查、开放领域问答、槽填充、实体链接和对话系统。数据集的设计允许研究者开发和评估能够访问特定知识源的模型,特别是在大型文本资源中。此外,KILT还提供了多种评估指标和工具,以支持对模型性能的全面评估,特别是在模型提供输出证明的能力方面。
The KILT dataset was created by Facebook AI Research to provide a unified benchmark for knowledge-intensive language tasks. It contains approximately 3.2 million instances, all based on the August 2019 Wikipedia dump to ensure data consistency and comparability. KILT covers five distinct task domains, including fact checking, open-domain question answering, slot filling, entity linking, and dialogue systems. The design of KILT enables researchers to develop and evaluate models that can access specific knowledge sources, particularly from large-scale textual resources. Additionally, KILT provides a range of evaluation metrics and tools to support comprehensive assessment of model performance, especially regarding the model's ability to furnish evidence for its outputs.

- 1KILT: a Benchmark for Knowledge Intensive Language TasksFacebook AI Research · 2021年



