NSF Convergence Accelerator Track F: How Large-Scale Identification and Intervention Can Empower Professional Fact-Checkers to Improve Democracy and Public Health

美国的民主和公共卫生依赖于对机构的信任。对美国选举公正性的怀疑，以及与新冠肺炎疫苗相关的犹豫不决，是人们对基本政治进程和核心医疗机构信心下降的两个后果。社交媒体是有关选举和疫苗的非法信息的主要来源，用户网络积极散布对选举诚信和疫苗效力的怀疑，助长了错误信息的传播。该项目旨在支持和支持记者、开发人员和公民对此类错误信息进行事实核查的努力。他们迫切需要这样的工具：1)当事实核查报道在Twitter、Reddit和Facebook等社交媒体平台上传播时，能够测试选举和疫苗等主题的事实核查报道；2)在完全透明的情况下，实时提供纠正效果的反馈。因此，该项目将开发一个互动系统，使实况核查人员能够对实况核查信息进行快速循环测试，并监测其在有错误信息暴露风险的在线社区中的实时表现。为了透明，所有基础代码、调查和数据将可供社会科学和计算机科学界共享，所有基于证据的、对公共卫生专业人员和选举管理人员的即时效用信息将公开。该项目的动机是希望了解并帮助解决美国面临的两个民主和公共卫生危机：对美国选举诚信的怀疑和对新冠肺炎疫苗的犹豫不决。这两场危机都是由社交媒体上广泛传播的网络错误信息加剧的，用户网络积极地散布对选举诚信和疫苗效力的怀疑。该项目将提供一种创新的三步法来识别、测试和纠正这些形式的在线错误信息的真实情况。首先，使用计算手段，如自然语言处理、机器学习、社会网络分析和建模以及计算机视觉等技术，以识别流传的、容易受到错误信息影响的帖子和账户。其次，将使用推荐系统对最突出的误报声明形式进行实验室测试更正，以优化消息效率。第三，该项目将利用平台赞助的内容系统提供的各种可扩展干预技术，传播和评价循证矫正的有效性。更具体地说，对于该方法的第一步，该项目将使用多模式信号检测和知识图来从事知识驱动的信息提取，关于社交媒体上的选举怀疑和疫苗犹豫不决，整合用户属性、消息特征和在线网络结构属性，以预测未来可能接触错误信息并识别易受干预的在线社区。第二步将包括与专业事实核查组织合作，对两种类型的干预信息-暴露前接种和暴露后纠正-进行实验室测试，旨在缓解选民的怀疑和疫苗的犹豫不决，并使用推荐系统技术对其进行优化。对于第三步，将进行现场实验，部署实验室开发的干预措施，通过广告购买、自动化机器人和在线影响力的组合提供；并评估我们干预措施在卫生和民主相关领域的最佳决策方面的成功。最终，这种三步走的方法可以应用于政治和卫生领域的一系列主题。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Democracy and public health in the United States rely on trust in institutions. Skepticism regarding the integrity of U.S. elections and hesitancy related to COVID-19 vaccines are two consequences of a decline in confidence in basic political processes and core medical institutions. Social media serve as a major source of delegitimizing information about elections and vaccines, with networks of users actively sowing doubts about election integrity and vaccine efficacy, fueling the spread of misinformation. This project seeks to support and empower efforts by journalists, developers, and citizens to fact-check such misinformation. They urgently need tools that can 1) enable testing of fact-checking stories on topics like elections and vaccines as they move across social media platforms like Twitter, Reddit, and Facebook, and 2) deliver feedback on how well the corrections worked in real time and with full performance transparency. Accordingly, this project will develop an interactive system that enables fact-checkers to perform rapid-cycle testing of fact-checking messages and monitor their real-time performance among online communities at-risk of misinformation exposure. To be transparent, all of the underlying code, surveys, and data will be available to share with the social science and computer science communities, and all evidence-based messages of immediate utility to public health professionals and electoral administrators will be made publicly accessible.This project is motivated by a desire to understand and help address two democratic and public health crises facing the U.S.: skepticism regarding the integrity of U.S. elections and hesitancy related to COVID-19 vaccines. Both of these crises are fueled by online misinformation, widely circulating on social media, with networks of users actively sowing doubts about election integrity and vaccine efficacy. The project will deliver an innovative, three-step method to identify, test, and correct real-world instances of these forms of online misinformation. First, using computational means, such as techniques in natural language processing, machine learning, social network analysis and modeling, and computer vision to identify posts and accounts circulating and susceptible to misinformation. Second, lab-tested corrections to the most prominent forms of misinforming claims using recommender systems to optimize message efficacy will be produced. And third, the project will disseminate and evaluate the effectiveness of evidence-based corrections using various scalable intervention techniques available through the platforms sponsored content systems. More specifically, for the first step of the method, the project will use multimodal signal detection and knowledge graph to engage in knowledge driven information extraction about electoral skepticism and vaccine hesitancy on social media, integrating user attributes, message features, and online network structural properties to predict likely exposure to future misinformation and identify susceptible online communities for intervention. The second step will consist of working with professional fact-checking organizations to lab test two types of intervention messages—pre-exposure inoculation and post-exposure correction— aimed at mitigating electoral skepticism and vaccine hesitancy, optimizing them using recommender system techniques. For the third step, field experiments will be conducted that deploy the lab-developed interventions, delivered through a combination of ad-purchasing, automated bots, and online influencers; and assess the success of our interventions with respect to optimal decision-making in both health and democracy-related arenas. Ultimately, this three-step approach can be applied across a range of topics in politics and health.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.