CAP: Semi-supervised Fairness-Enhanced Knowledge Graph Construction on Social Media for AI-Enhanced Juvenile Justice

该项目是一项ExpandAI能力建设试点项目（CAP），重点是通过利用AI技术和社交媒体数据，在Prairie View A &； M大学建立和发展与AI相关的活动，以显著加强预防暴力和向青少年伸张正义。社交媒体是与帮派有关联、挑起肢体冲突的青年以及分享暴力和帮派冲突信息的社区成员的主要信息来源。通过Prairie View A&amp；M大学工程学院和青少年司法学院的合作，将使用从社交媒体中提取的数据构建公平增强知识图谱。这些知识图（KG）是智能系统用于解决复杂问题的事实表示的结构。调查结果将消除偏见，并用于揭示青少年暴力循环的机制、后果和当地知识。该项目预计将显著提高Prairie View A&amp；M大学和HBCU的人工智能研究和教学能力。该项目将促进公平，加强暴力预防，促进在自然语言处理、计算机视觉、机器学习和可信赖的人工智能领域对使用启发型人工智能的研究。在教育方面，计划的活动不仅将使非洲裔美国学生获得重要的前沿跨学科技能，而且还将大大扩展COE和COJJ学生的职业道路。公平增强知识图谱构建过程包括以下几个任务：(i)增强社交媒体平台上表示的司法相关数据的公平性；（ii）半监督式FKG施工；（iii）全面的FKG质量评估和工具开发。将使用基于人在环的群体学习方法将用户反馈整合到学习中，以促进假新闻检测。在构建知识图谱的过程中，将采用去偏技术来提高公平性。该项目将修改两所大学现有课程的元素，为学生提供指导，并为与司法应用相关的人工智能竞赛提供培训。与NVIDIA深度学习研究所等外部组织合作，将开发用于人工智能培训的教学工具包，此外，还将向社区、HBCU等进行推广。该项目由传统黑人学院和大学本科项目（HBCU-UP）共同资助，该项目为加强hbcu的STEM本科教育和研究提供奖励。ExpandAI计划支持少数族裔服务机构的人工智能教育和劳动力发展、基础设施和研究，以加强和多样化美国的研究和教育途径，并为历史上被边缘化的社区提供STEM职业的新机会。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。

This project is an ExpandAI Capacity building pilot (CAP), which focuses on establishing and growing AI related activities at Prairie View A & M University by utilizing AI technology and social media data for significant enhancement of violence prevention and delivery of justice to juveniles. Social media is a major source of information for gang-associated youth who instigate physical conflicts as well as community members who share information about violence and gang conflicts. Through a collaboration between Prairie View A&M University College of Engineering and College of Juvenile Justice, fairness-enhanced knowledge graphs will be constructed using data extracted from social media. These knowledge graphs (KG) are constructs for the representation of facts used by intelligent systems for the solution of complex problems. KGs will be de-biased and used to uncover mechanisms for, consequences of, and local knowledge about the cycle of youth violence. The project is expected to significantly enhance the AI research and instruction capacity of Prairie View A&M University – an HBCU. The project will promote fairness and enhance violence prevention, catalyzing research in use-inspired AI in the fields of natural language processing, computer vision, machine learning and trustworthy AI. On the educational side, the activities planned will not only enable African American students to acquire important cutting-edge cross-disciplinary skills but also significantly expand the career pathways for both COE and COJJ students. The fairness-enhanced knowledge graph construction process involves several tasks including (i) enhancement of the fairness of justice-related data represented on social media platforms; (ii) semi-supervised FKG construction; and (iii) comprehensive FKG quality assessment and tool development. A human-in-the-loop based swarm learning approach will be used to integrate user feedback into learning to facilitate fake news detection. De-biasing techniques will be employed to enhance fairness during Knowledge Graph construction. The project will revamp elements of existing curricula across both colleges, provide mentoring for students and training for AI competitions related to justice applications. In collaboration with external organizations like NVIDIA Deep Learning Institute, teaching kits for AI training will be developed, furthermore, there will be outreach to the community, HBCU and beyond. This project is co-funded by the Historically Black Colleges and Universities Undergraduate Program (HBCU-UP), which provides awards to strengthen STEM undergraduate education and research at HBCUs. The ExpandAI Program supports AI-powered education and workforce development, infrastructure and research at Minority Serving Institutions to strengthen and diversify U.S. research and education pathways and provide historically marginalized communities with new opportunities in STEM careers.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.