Official-NV
收藏资源简介:
Official-NV数据集由苏州大学计算机科学与技术学院创建,包含10,000条官方发布的英文新闻视频,旨在支持多模态假新闻检测研究。数据集通过从新华网爬取视频并利用大型语言模型生成和人工修改扩展而来,包含5,000条真实视频和5,000条假视频。每条视频包含标题、视频帧和语音文本三种模态,其中假视频至少有一种模态与其他模态信息不一致。该数据集主要用于通过分析多模态信息来提高假新闻检测的准确性。
The Official-NV Dataset was developed by the School of Computer Science and Technology, Soochow University. It consists of 10,000 officially released English news videos, targeted at supporting research on multimodal fake news detection. The dataset is constructed by crawling videos from Xinhua News, followed by expansion via generation with large language models and manual revision. It contains 5,000 authentic videos and 5,000 fake videos. Each video includes three modalities: title, video frames, and speech transcript. For fake videos, at least one modality contains information inconsistent with that from the other modalities. This dataset is primarily utilized to enhance the accuracy of fake news detection through the analysis of multimodal information.

- 1Official-NV: A News Video Dataset for Multimodal Fake News Detection苏州大学计算机科学与技术学院 · 2024年



