MMFakeBench
收藏资源简介:
MMFakeBench是由北京邮电大学等机构创建的混合源多模态虚假信息检测基准数据集,包含11,000对数据,涵盖文本真实性扭曲、视觉真实性扭曲和跨模态一致性扭曲三大类。数据集通过高级AI工具如扩散生成器和ChatGPT生成,包含12种伪造类型,旨在模拟真实世界中多源虚假信息的复杂性。该数据集应用于检测和分析多模态虚假信息,特别是在社交媒体上的应用,以解决政治、金融和公共卫生领域的虚假信息问题。
MMFakeBench is a mixed-source multimodal disinformation detection benchmark dataset developed by institutions including Beijing University of Posts and Telecommunications. It consists of 11,000 data pairs, covering three categories: textual authenticity distortion, visual authenticity distortion, and cross-modal consistency distortion. The dataset is generated via advanced AI tools such as diffusion generators and ChatGPT, and includes 12 types of forgeries, aiming to simulate the complexity of multi-source disinformation in the real world. This benchmark is applied to the detection and analysis of multimodal disinformation, particularly in social media scenarios, to address disinformation-related issues in the fields of politics, finance, and public health.

- 1MMFakeBench: A Mixed-Source Multimodal Misinformation Detection Benchmark for LVLMs北京邮电大学 · 2024年



