MHTS Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://www.gutenberg.org/ebooks/766
下载链接
链接失效反馈官方服务：
资源简介：
该数据集名为MHTS，采用多跳树结构合成了逻辑上相连的多块查询，以评估检索增强生成（RAG）系统。该数据集专门用于评估多跳推理的复杂性，其结构设计精细，能够进行难度级别的细致调整。数据集包含505个数据块，每个数据块的长度在540到1016个标记之间，针对的任务是多跳问答（Qa）。
The dataset is named MHTS. It synthesizes logically connected multi-block queries via a multi-hop tree structure to evaluate Retrieval-Augmented Generation (RAG) systems. This dataset is specifically designed to assess the complexity of multi-hop reasoning, with an elaborately structured design that enables fine-grained adjustment of difficulty levels. The dataset contains 505 data blocks, each with a length ranging from 540 to 1016 tokens, targeting the multi-hop Question Answering (QA) task.
