AirsideLabs/SWIFTPortal_help

SWIFT Portal数据集包含了处理过的文档、问题和评估信息，适用于SWIFT Portal项目。数据集以Hugging Face数据集格式组织，分为多个配置，包括原始文档信息、带摘要的文档、分块文档、单跳问题、多跳问题以及问题回答的评估格式。可用于文档问答系统训练、多跳推理测试、文档摘要和分块方法基准测试以及检索系统开发。The SWIFT Portal Dataset contains processed documents, questions, and evaluations for the SWIFT Portal project. It is organized in the Hugging Face dataset format and consists of multiple configurations, including raw document information, documents with summaries, chunked documents, single-shot questions, multi-hop questions, and an evaluation format for question answering. It can be used for training QA systems, testing multi-hop reasoning, benchmarking summarization and chunking methods, and developing retrieval systems with citation tracking.

搜集汇总

背景与挑战

背景概述

该数据集包含SWIFT Portal项目的文档、问题及评估数据，共641行、741KB，以parquet格式存储。提供原始文档、摘要、分块、单跳和多跳问题以及评估集，适用于训练和评估文档问答系统及多跳推理能力。