TamperedNews & News400 Datasets (IJMIR'21 Update) BU

# Multimodal Analytics for Real-world News using Measures of Cross-modal Entity Consistency This repository contains the *TamperedNews* and *News400* datasets introduced in the paper: > Eric Müller-Budack, Jonas Theiner, Sebastian Diering, Maximilian Idahl, Sherzod Hakimov und Ralph Ewerth. „Multimodal news analytics using measures of cross-modal entity and context consistency“. In: _International Journal of Multimedia Information Retrieval_ 10.2 (2021), Springer, S. 111–125. DOI: https://doi.org/10.1007/s13735-021-00207-4 ## Content For both datasets *TamperedNews* and *News400*, we provide the: - ```*dataset*.tar.gz``` containing the ```*dataset*.jsonl``` with - Web links to the news texts - Web links to the news image - Outputs of the named entity recognition and disambiguation (NERD) approach - Untampered and tampered entities - ```*dataset*_vise_features.tar.gz```with visual features for events extracted from our event classification approach VisE presented at WACV'21 ([paper](https://openaccess.thecvf.com/content/WACV2021/html/Muller-Budack_Ontology-Driven_Event_Type_Classification_in_Images_WACV_2021_paper.html), [GitHub](https://github.com/TIBHannover/VisE)) Please note that the remaining visual features (```*dataset*_features.tar.gz```) and word embeddings (```*dataset*_wordembeddings.tar.gz```) have been already provided in the first version of both datasets ([News400](https://data.uni-hannover.de/dataset/news400), [TamperedNews](https://data.uni-hannover.de/dataset/tamperednews)). For all entities detected in both datasets, we provide: - ```entities.tar.gz``` containing an ```*entity_type*.jsonl``` for all entity types (events, locations, and persons) with: - Wikidata ID - Wikidata label - Meta information used for tampering - Web links to all reference images crawled from Google, Bing, and Wikidata - ```entities_features.tar.gz``` containing the visual features of the reference images for all entities ## Source Code The source code to reproduce our results as well as download scripts to crawl news texts and images can be found on our GitHub page: https://github.com/TIBHannover/cross-modal_entity_consistency# 基于跨模态实体一致性度量的真实世界新闻多模态分析 本仓库包含论文中提出的*TamperedNews*与*News400*数据集： > 作者：Eric Müller-Budack、Jonas Theiner、Sebastian Diering、Maximilian Idahl、Sherzod Hakimov 与 Ralph Ewerth。论文题为《基于跨模态实体与上下文一致性度量的多模态新闻分析（Multimodal news analytics using measures of cross-modal entity and context consistency）》，发表于《国际多媒体信息检索期刊（International Journal of Multimedia Information Retrieval）》2021年第10卷第2期，施普林格出版社（Springer）出版，页码111–125，DOI：https://doi.org/10.1007/s13735-021-00207-4 ## 数据集内容 针对*TamperedNews*与*News400*两个数据集，我们提供如下内容： - *dataset*.tar.gz：内含记录了以下信息的*dataset*.jsonl（JSON Lines格式）： - 新闻文本的网页链接 - 新闻配图的网页链接 - 命名实体识别与消歧（Named Entity Recognition and Disambiguation, NERD）模型的输出结果 - 未篡改实体与篡改实体 - *dataset*_vise_features.tar.gz：包含我们基于WACV'21会议发表的事件分类方法VisE提取的事件视觉特征。相关资料包括[论文](https://openaccess.thecvf.com/content/WACV2021/html/Muller-Budack_Ontology-Driven_Event_Type_Classification_in_Images_WACV_2021_paper.html)与[GitHub仓库](https://github.com/TIBHannover/VisE)。 请注意，两个数据集的首个版本已提供剩余视觉特征包（*dataset*_features.tar.gz）与词嵌入包（*dataset*_wordembeddings.tar.gz），相关数据集链接可访问：[News400](https://data.uni-hannover.de/dataset/news400)、[TamperedNews](https://data.uni-hannover.de/dataset/tamperednews)。 针对两个数据集中识别出的所有实体，我们提供如下内容： - entities.tar.gz：内含针对所有实体类型（事件、地点与人物）的*entity_type*.jsonl文件，其中包含： - Wikidata标识符 - Wikidata实体标签 - 用于实体篡改的元信息 - 从Google、Bing与Wikidata爬取的所有参考图片的网页链接 - entities_features.tar.gz：包含所有实体参考图片的视觉特征。 ## 源代码 可用于复现本文实验结果、以及爬取新闻文本与配图的源代码与下载脚本，均可在我们的GitHub仓库获取：https://github.com/TIBHannover/cross-modal_entity_consistency

搜集汇总

背景与挑战

背景概述

该数据集包括TamperedNews和News400两个部分，用于多模态新闻分析，重点关注跨模态实体一致性的度量，以检测新闻中的篡改行为。数据集提供新闻文本和图像的链接、实体识别输出、视觉特征以及实体参考图像，支持研究新闻真实性和多模态验证任务。