GRS-QA
收藏资源简介:
GRS-QA是由加州大学圣克鲁兹分校的研究团队创建的图推理结构化问答数据集。该数据集包含113,000个基于维基百科的问答对,旨在通过引入显式的推理结构图来评估大型语言模型在多跳问答任务中的推理能力。数据集的创建过程包括从现有多跳问答数据集(如HotpotQA、MuSiQue和2WikiMultiHopQA)中提取问答对,并构建推理图。这些推理图不仅展示了从问题到答案的逻辑步骤,还通过引入负样本图来研究结构对问答性能的影响。GRS-QA的应用领域主要集中在评估和提升大型语言模型在复杂推理任务中的表现。
GRS-QA is a structured question answering dataset for graph reasoning created by a research team at the University of California, Santa Cruz. This dataset contains 113,000 Wikipedia-based question-answer pairs, aiming to evaluate the reasoning capabilities of large language models in multi-hop question answering tasks by introducing explicit reasoning graphs. The dataset construction process includes extracting question-answer pairs from existing multi-hop QA datasets such as HotpotQA, MuSiQue, and 2WikiMultiHopQA, and constructing reasoning graphs. These reasoning graphs not only demonstrate the logical steps from the question to the answer, but also introduce negative sample graphs to study the impact of structure on question answering performance. The main application areas of GRS-QA focus on evaluating and enhancing the performance of large language models in complex reasoning tasks.

- 1GRS-QA -- Graph Reasoning-Structured Question Answering Dataset加州大学圣克鲁兹分校 · 2024年



