PFI-RP: A Multi-Disciplinary Approach to Detecting Adolescent Online Risks.

这个PFI项目的更广泛的影响/商业潜力是开发改进的社交媒体工具，使青少年在网上更安全。一个由研究人员、临床医生和行业合作伙伴组成的多学科团队将构建、评估最先进的技术，并将其商业化，这些技术可以主动检测青少年在线风险行为，包括心理健康问题、性招揽和在线骚扰，以降低这些风险并防止伤害。该团队计划发布两个独立的技术产品：第一，一个开放源码项目，与开发人员共享我们的风险检测算法，以便他们可以在这些解决方案的基础上构建一个致力于促进青少年在线安全的社区。第二，将这些风险检测算法结合到一项易于使用和可访问的服务中的商业产品，该服务为基于互联网的公司提供所需的支持和基础设施，这些公司最终为青少年在网上遇到的风险提供便利，以便他们可以分担保护使用他们平台的青少年的共同责任。作为一个社会，我们必须更加积极主动地保护青少年免受网络风险，特别是最脆弱的青少年，他们参与可能导致身体伤害甚至死亡的在线活动的风险最高。拟议的项目通过利用以人为中心的机器学习的多学科方法来准确检测青少年在线遇到的风险，解决了青少年网络安全这一关键和及时的问题。这项工作将分四个阶段完成：1)以用户为中心的阶段，审查对青少年及其家庭最重要的在线风险暴露类型。有了这一洞察力，该团队将建立基于证据的青少年在线风险模型，以确定风险的关键维度和模式。2)数据驱动阶段，它利用这些模型来确定基本事实并注释青少年社交媒体数据的丰富训练集；3)机器学习阶段，它改进现有的风险检测算法，并构建新的风险检测算法，这些算法针对青少年并能够随着时间的推移检测行为变化；4)技术开发阶段，它评估我们的解决方案，并将其在两个独立的产品线中进行商业化。最终的结果将是一套为青少年量身定做的算法，这些算法可以检测社交媒体中的不良内容，并识别出有问题的行为模式或变化，这些模式或变化预示着随着时间的推移即将到来的风险。这项工作将实现实时在线安全干预，以保护和授权青少年互联网用户。该奖项反映了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。

The broader impact/commercial potential of this PFI project is to develop improved social media tools to keep teens safer online. A multidisciplinary team of researchers, clinicians, and industry partners will build, evaluate, and commercialize state-of-the-art technologies that proactively detect adolescent online risk behaviors, including mental health issues, sexual solicitations, and online harassment, for the purpose of mitigating these risks and preventing harm. The team plans to release two separate technology offerings: First, an open source project that shares our risk detection algorithms with developers, so that they can build upon these solutions and form a community dedicated to promoting adolescent online safety. Second, a commercial product that combines these risk detection algorithms into an easy-to-use and accessible service that provides needed support and infrastructure to internet-based companies, who ultimately facilitate the risks teens encounter online, so that they can share in the joint responsibility of protecting the teens who use their platforms. It is imperative that as a society we become more proactive about protect teens from online risks, especially the most vulnerable teens who are at highest risk of engaging in online activities that can lead to physical harm or even death.The proposed project addresses the critical and timely problem of adolescent online safety by leveraging a multi-disciplinary approach of human-centered machine learning to accurately detect risks teen encounter online. The work will be accomplished in four stages: 1) a user-centered phase that examines the types of online risk exposure that matter most to teens and their families. With this insight, the team will build evidence-based adolescent online risk models that identify key dimensions and patterns of risk. 2) a data-driven phase that draws upon these models to establish ground truth and annotate a rich training set of teen social media data, 3) a machine-learning phase that improves existing and builds new risk detection algorithms that are contextualized to teens and able to detect behavioral changes over time, and 4) a technology development phase that evaluates our solution and commercializes it within two separate product lines. The end result will be a suite of algorithms tailored to adolescents that detect objectionable content within social media and identify problematic behavioral patterns or changes that are indicative of impending risks over time. This work will enable real-time online safety interventions that protect and empower teen internet users.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.