FW-HTF-RM: Collaborative Research: Augmenting Social Media Content Moderation

在世界各地，社交媒体平台的用户每天都会生成数百万条评论、视频和照片。在这个内容是危险的材料，如儿童色情，性交易，恐怖主义宣传。尽管平台利用算法系统来促进问题内容的检测和删除，但关于是否删除内容的决定，无论是像离题评论那样的善意，还是像自残或虐待视频那样的危险，通常都是由人类做出的。公司正在招聘成千上万的版主，成千上万的人志愿担任版主。这项工作涉及经济、情感和人身安全风险。本项目以社交媒体内容审核为工作重点，内容审核员为工作人员，通过设计新技术来提高审核员的绩效，促进人类与技术的合作。该项目将提高版主的生活质量，增强他们的能力，并帮助社会了解版主的决策是如何做出的，以及如何支持那些帮助保持互联网开放和愉快的工作人员。这些进步将使审核工作能够跟上用户生成内容的步伐，并确保有问题的内容不会压倒互联网用户。该项目包括与学术界、工业界、政策制定者和公众开展的外联和参与活动，以确保项目的发现和工具支持受用户生成内容及其审核影响的广泛利益相关者。具体而言，该项目涉及五个主要研究目标，将通过定性、历史、实验和计算研究方法来实现。首先，该项目将通过对不同内容领域的审核员进行访谈和观察，提高对“人在循环”决策实践和审核员心理模型的理解。其次，它将通过行业人员访谈评估技术增强适度的社会经济影响。第三，该项目将测试干预措施，以减少人类调节者的情绪损失，并通过一系列实验利用压力缓解理论优化他们的表现。第四，该项目将为直播主持人设计、开发和测试一套认知辅助工具。这些工具将专注于移除简单的决策，并帮助版主动态管理他们的情绪和认知能力。最后，该项目将采用历史视角来分析公司的内容审核政策，为法律和平台政策提供信息。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Around the world, users of social media platforms generate millions of comments, videos, and photos per day. Within this content is dangerous material such as child pornography, sex trafficking, and terrorist propaganda. Though platforms leverage algorithmic systems to facilitate detection and removal of problematic content, decisions about whether to remove content, whether it's as benign as an off-topic comment or as dangerous as self-harm or abuse videos, are often made by humans. Companies are hiring moderators by the thousands and tens of thousands work as volunteer moderators. This work involves economic, emotional, and often physical safety risks. With social media content moderation as the focus of work and the content moderators as the workers, this project facilitates the human-technology partnership by designing new technologies to augment moderator performance. The project will improve moderators' quality of life, augment their capabilities, and help society understand how moderation decisions are made and how to support the workers who help keep the internet open and enjoyable. These advances will enable moderation efforts to keep pace with user-generated content and ensure that problematic content does not overwhelm internet users. The project includes outreach and engagement activities with academic, industry, policy-makers, and the public that ensure the project's findings and tools support broad stakeholders impacted by user-generated content and its moderation.Specifically, the project involves five main research objectives that will be met through qualitative, historical, experimental, and computational research approaches. First, the project will improve understanding of human-in-the-loop decision making practices and mental models of moderation by conducting interviews and observations with moderators across different content domains. Second, it will assess the socioeconomic impact of technology-augmented moderation through industry personnel interviews. Third, the project will test interventions to decrease the emotional toll on human moderators and optimize their performance through a series of experiments utilizing theories of stress alleviation. Fourth, the project will design, develop, and test a suite of cognitive assistance tools for live streaming moderators. These tools will focus on removing easy decisions and helping moderators dynamically manage their emotional and cognitive capabilities. Finally, the project will employ a historical perspective to analyze companies' content moderation policies to inform legal and platform policies.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.