EAGER: DCL: SaTC: EIC: Inclusive-ScamBuster: Inclusive Scam Detection Methods for Social Media to Design Assistive Tools for Protecting Individuals with Developmental Disabilities

防止基于社交媒体的诈骗是网络安全的一个关键挑战。存在在在线浏览期间保护个人的工具，然而，它们不是针对弱势亚群，如具有发育障碍的个人（例如，自闭症）。这些人在没有专门支持的情况下成为目标，以协助在潜在的诈骗帖子中识别威胁。该项目旨在了解自闭症和注意力缺陷/多动障碍（ADHD）患者所表现出的独特理解和注意力模式，以改进欺诈检测工具来帮助这些亚群。该项目的新颖之处包括一种多学科方法，结合了社会计算、认知心理学、特殊教育和计算语言学研究，以解决诈骗检测工具中使用的自然语言处理（NLP）人工智能方法中存在的偏见，该方法基于对弱势亚群所显示的浏览模式的行为研究。该项目更广泛的意义在于将对人类行为的洞察力整合到网络安全工具中，从而更好地保护弱势群体，并提高网络安全的包容性。 该项目追求两个目标。首先，它开发了一项眼动追踪研究，以发现在暴露于诈骗和合法社交媒体帖子时，有和没有发育障碍的人群中可观察到的注意力模式的变化。其次，它使用观察到的注意力模式的变化来突出基于NLP的骗局检测模型的标记数据集中的表示偏差。它还创建了一组新的语言属性，可用于训练为帮助弱势群体而量身定制的欺诈检测模型。项目成果包括更好地了解针对弱势群体的社交媒体诈骗，开发用于诈骗检测的包容性NLP模型，以及通过量身定制的诈骗警报帮助发育障碍人士的开源浏览器插件原型。该项目还创建了一个门户网站（Inclusive-ScamBuster），托管标记的诈骗数据集，以突出代表性偏见和开源教育资源，以支持特殊教育项目在教学和培训网络犯罪预防方面的工作。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

Preventing social media-based scams is a critical challenge for cybersecurity. There exist tools to protect individuals during online browsing, however, they are not tailored towards vulnerable subpopulations like individuals with developmental disabilities (e.g., Autism). Such individuals become targets without dedicated support to assist with threat identification in potential scam posts. This project aims to understand the distinctive comprehension and attention patterns displayed by individuals with Autism and Attention-Deficit/Hyperactivity Disorder (ADHD), to improve scam detection tools to assist these subpopulations. The project’s novelties include a multidisciplinary approach combining social computing, cognitive psychology, special education, and computational linguistics research to address existing biases in Artificial Intelligence methods of Natural Language Processing (NLP) used in scam detection tools, based on behavioral studies of browsing patterns displayed by vulnerable subpopulations. The project’s broader significance is in integrating insights of human behavior into cybersecurity tools, leading to better protection of vulnerable subpopulations and greater inclusiveness in cybersecurity. This project pursues two goals. First, it develops an eye-tracking study to discover variations in attention patterns observable across populations with and without developmental disabilities when exposed to scams and legitimate social media posts. Second, it uses observed variations in attention patterns to highlight representation biases in the labeled datasets of NLP-based scam detection models. It further creates a novel set of linguistic attributes that can be used to train scam detection models tailored to aid vulnerable subpopulations. Project outcomes include a better understanding of social media scams for vulnerable subpopulations, the development of an inclusive NLP model for scam detection, and an open-source browser plugin prototype to aid individuals with developmental disabilities via tailored scam alerts. The project also creates a web portal (Inclusive-ScamBuster) hosting labeled scam datasets to highlight representational biases and open-source educational resources to support Special Education programs in teaching and training cybercrime prevention.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.