INS-MMBench

INS-MMBench是由复旦大学创建的针对保险领域的大型视觉语言模型（LVLMs）的综合基准，包含2.2K个精心设计的多项选择视觉问题，覆盖12个元任务和22个基础任务，涉及汽车保险、财产保险、健康保险和农业保险四大类型。数据集通过多渠道公开数据源收集，经过筛选和处理，确保与保险场景高度相关。该数据集旨在深入评估LVLMs在保险领域的应用潜力，解决保险业务中的多模态数据处理和视觉推理问题，推动保险行业的技术创新和跨学科发展。INS-MMBench is a comprehensive benchmark for large vision-language models (LVLMs) in the insurance domain, developed by Fudan University. It includes 2.2K meticulously designed multiple-choice visual questions, covering 12 meta-tasks and 22 basic tasks across four insurance categories: auto insurance, property insurance, health insurance and agricultural insurance. The dataset is collected from multi-channel public data sources, and has been screened and processed to ensure high relevance to real-world insurance scenarios. This benchmark aims to thoroughly evaluate the application potential of LVLMs in the insurance field, address multimodal data processing and visual reasoning challenges in insurance operations, and promote technological innovation and interdisciplinary development within the insurance industry.

搜集汇总

背景与挑战

背景概述

INS-MMBench是由复旦大学创建的保险领域大型视觉语言模型综合基准，包含2.2K个多项选择视觉问题，覆盖12个元任务和22个基础任务，涉及汽车、财产、健康、农业四大保险类型。该数据集通过公开数据源收集并处理，旨在评估模型在保险场景中的多模态数据处理和视觉推理能力，推动保险行业技术创新。