VPWiki
收藏资源简介:
VPWiki数据集是由天津大学智能与计算学院和新媒体与传播学院构建的,用于视觉提示引导的多模态实体链接任务。该数据集基于现有的MEL公共数据集,包含12720个样本,涵盖多种实体类型。数据集通过自动化和手动标注相结合的方式创建,旨在解决传统MEL方法对提及词的过度依赖问题。VPWiki数据集的应用领域主要集中在信息检索和多模态实体链接,旨在通过视觉提示提升图像与文本信息的有效结合,从而提高实体链接的准确性。
The VPWiki dataset was constructed by the School of Intelligence and Computing and the School of New Media and Communication, Tianjin University, for the task of visual-prompt-guided multimodal entity linking. Built upon the existing public MEL dataset, this dataset contains 12,720 samples covering multiple entity types. It was developed through a combination of automated and manual annotation, aiming to address the over-reliance of traditional MEL methods on entity mentions. The application domains of the VPWiki dataset mainly focus on information retrieval and multimodal entity linking, with the objective of enhancing the effective integration of image and text information via visual prompts, thereby improving the accuracy of entity linking.

- 1VP-MEL: Visual Prompts Guided Multimodal Entity Linking天津大学 · 2024年



