用自动化工具识别社交媒体截图中的作者误标问题
Categorizing Social Media Screenshots for Identifying Author Misattribution
- 通过结构分类与元数据提取，自动整合截图中的多条推文
- 在75个手动收集样本上实现0.80的F1分数，准确率较高
- 适用于反虚假信息研究者和平台内容审核人员
虚假信息在社交媒体中广泛存在。作者误标是一种虚假信息形式，指用户错误地声称某人说过或发过某句话。本文区分了虚假信息与恶意信息，并分析了截图如何被用于传播作者误标。为验证截图是否正确归因，需找到原始帖子。为此，我们开发了多个工具：第一个是Python脚本，根据推文结构进行分类，提取截图中的元数据，并将截图内各条推文分组；在75个手动采集的含截图推文样本上测试，元数据提取与分组的F1得分为0.80。第二个是系列爬虫，用于收集可用于训练和测试模型的数据集，涵盖Facebook、Instagram、Truth Social和Twitter的网页版与移动端，包括亮色与暗色模式，共收集16,620张截图。
原文摘要 · Abstract (English)
Mis/disinformation is a common and dangerous occurrence on social media. Misattribution is a form of mis/disinformation that deals with a false claim of authorship, which means a user is claiming someone said (posted) something they never did. We discuss the difference between misinformation and disinformation and how screenshots are used to spread author misattribution on social media platforms. It is important to be able to find the original post of a screenshot to determine if the screenshot is being correctly attributed. To do this we have built several tools to aid in automating this search process. The first is a Python script that aims to categorize Twitter posts based on their structure, extract the metadata from a screenshot, and use this data to group all the posts within a screenshot together. We tested this process on 75 Twitter posts containing screenshots collected by hand to determine how well the script extracted metadata and grouped the individual posts, F1 = 0.80. The second is a series of scrapers being used to collect a dataset that can train and test a model to differentiate between various social media platforms. We collected 16,620 screenshots have been collected from Facebook, Instagram, Truth Social, and Twitter. Screenshots were taken by the scrapers of the web version and mobile version of each platform in both light and dark mode.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。