MNRE-2

该数据集名为MNRE-2，包含了从推特上搜集的15,484个样本，其中有9,201组图像与标题的配对，涵盖了23种关系类型，包括无关系类型。该数据集被划分为训练集、验证集和测试集，分别包含12,247、1,624和1,614个样本。评估时，采用标准的微精度、召回率和F1分数进行报告。在规模上，该数据集共有15,484个样本，包含9,201组图像与标题的配对。其任务是进行多模态关系提取。

The dataset is named MNRE-2. It contains 15,484 samples collected from Twitter, among which there are 9,201 image-caption pairs, covering 23 relationship types including the no-relation category. It is divided into training, validation and test sets, with 12,247, 1,624 and 1,614 samples respectively. Standard micro-precision, recall and F1 score are adopted for evaluation reporting. In terms of scale, this dataset has a total of 15,484 samples and 9,201 image-caption pairs. Its task is multimodal relation extraction.

搜集汇总

背景与挑战

背景概述

MNRE-2是一个多模态关系提取数据集，包含9,201张图像和15,484个样本，涵盖23种关系类别。数据集分为训练集、开发集和测试集，分别包含12,247、1,624和1,614个样本。该数据集旨在通过视觉证据提升社交媒体帖子中实体关系的理解。