WebQA

WebQA 是多模式多跳推理的新基准，其中系统在搜索网络时呈现与人类相同的数据样式：片段和图像。然后，系统必须识别哪些信息是跨模式相关的，并将其与推理相结合以回答查询。系统将根据其答案的正确性和来源进行评估。

WebQA is a novel benchmark for multimodal multi-hop reasoning. In this benchmark, when conducting web searches, systems are provided with the same data formats as human users: web snippets and images. Subsequently, systems must identify which information is cross-modally relevant, and combine such information with reasoning processes to answer user queries. The performance of these systems will be evaluated based on the correctness of their answers and their supporting sources.

搜集汇总

背景与挑战

背景概述

WebQA是一个多模式多跳推理基准数据集，旨在评估系统在处理网络搜索中的片段和图像信息时进行跨模式相关识别和推理的能力。该数据集由卡内基梅隆大学和微软于2022年发布，包含约50.8GB的数据，适用于文本预训练、多模态查询和图像检索等任务，采用CC0 1.0许可证开放使用。