III:Small: A novel machine learning framework for combating misinformation in real life

社交媒体上持续存在广泛的虚假信息，给社会带来了重大挑战。错误信息涉及迅速演变的话题和不良行为者之间的复杂协调。现有的打击虚假信息的技术在一定程度上取得了成功。他们努力及时有效地消除虚假和误导性内容。另一个挑战是难以量化错误信息的影响。在这个项目中，将利用机器学习的最新发展来开发一个新的机器学习框架。它将促进对错误信息传播模式的理解，产生有效的错误信息检测算法，并有助于建立一个安全可信的网络空间。将开展一系列外联活动，以扩大妇女和其他任职人数不足的群体在计算机方面的参与。将提供教程和课程来传播研究成果。该项目将同时推进机器学习方法和社会科学见解，导致有效的解决方案，以及时、可扩展和有效的方式减少错误信息和操纵。本项目将推进三个研究主题。第一个推力开发了一个基于强化学习的解决方案，使用新闻来源可信度分析来最大限度地减少构建大规模错误信息数据集时的人工标记工作。第二个推力提出了一种基于知识知晓的机器学习模型的无监督协调检测方法，以识别不良行为者之间的协调行为。第三个推动力开发了新的反事实分析模型，以确定因果因素并评估错误信息的估计影响。关于各种主题的大规模错误信息数据集的集合将与研究社区共享。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。

Widespread misinformation has been persistently observed on social media, raising significant challenges to society. Misinformation involves rapidly evolving topics and sophisticated coordination among bad actors. Existing techniques to combat misinformation have achieved success to a limited extent. They struggle to eliminate false and misleading content in a timely and effective manner. Another challenge is the difficulty of quantifying the impacts of misinformation. In this project, a novel machine-learning framework will be developed by leveraging recent developments in machine learning. It will advance the understanding of misinformation propagation patterns, produce effective algorithms for misinformation detection, and help build a secure and trustworthy cyberspace. A range of outreach activities will be pursued to broaden participation in computing for women and other underrepresented groups. Tutorials and courses will be provided to broadcast the research outcomes.The project will advance both machine-learning methodologies and social science insights, leading to effective solutions to mitigate misinformation and manipulation in a timely, scalable, and effective manner.Three research thrusts will be pursued in this project. The first thrust develops a reinforcement-learning-based solution using news-source credibility analysis to minimize human labeling efforts in constructing large-scale misinformation datasets. The second thrust develops an unsupervised coordination detection method based on knowledge-informed machine-learning models to identify coordinated behaviors among bad actors. The third thrust develops novel counterfactual analysis models to identify causal factors and evaluate the estimated effects of misinformation. A collection of large-scale datasets of misinformation on a variety of topics will be shared with the research community.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.