Collaborative Research: CISE-MSI: DP: SCH: Privacy Preserving Tutoring System for Health Education of Low Literacy Hispanic Populations

该奖项全部或部分由《2021年美国救援计划法案》（公法117-2）资助。该项目将为低识字率的西班牙裔乳腺癌幸存者提供计算机辅导。乳腺癌是西班牙裔美国人癌症相关死亡的主要原因，尽管研究表明，教育可以极大地减轻压力，提高生活质量，但针对这一人群的教育干预措施很少。这个项目产生的计算机导师将模仿人类导师，教授乳腺癌生存技能和乳腺癌的一般知识。因为辅导包括与幸存者的对话，他们有可能泄露敏感的个人信息；因此，在家教过程中对信息进行加密是很重要的，以防止任何隐私泄露。为了使辅导成分有效，必须特别注意自然语言处理模型的使用，以及在辅导和/或与技术互动时在目标人群中观察到的行为模型。为了使隐私组件有效，将探索能够高速加密和解密数据的技术，以使交互流畅。到目前为止，与学生交谈的计算机导师主要是在大学环境中受过高等教育的人群中进行的，因此他们对低文化水平的西班牙裔的影响尚不清楚。因此，该项目将促进我们对设计人工智能家教的影响的理解，以解决低文化个体子集获取信息的多样性和差异，以及我们对与复杂的自然语言处理模型实时工作的隐私保护算法的理解。更广泛地说，项目成果将促进少数民族人口获得信息，并将有助于在参与的以教学为导向的机构中建立研究能力和培训少数民族学生。该项目将在执行时考虑到两个目标。首先，开发一种新型的智能计算机辅导系统，该系统是定制的，因此它可以通过调整先前研究中为该人群创建的现有内容，有效地查询和与西班牙裔乳腺癌幸存者互动。因为研究表明，许多西班牙裔成年人的语言和目标人群与技术的互动，都比之前认为的更加微妙，我们的第一个目标还包括训练自然语言算法，并设计模拟西班牙裔乳腺癌幸存者的互动。第二个目标是开发隐私保护算法，利用健壮的端到端加密通信，可以以不妨碍与计算机导师交互的速度实时加密和解密分布式数据。这一发展过程的贡献将有三个方面：(1)了解文化和教育在低识字率西班牙裔乳腺癌幸存者与智能辅导系统之间的相互作用中的作用；(2)制定有利于少数民族群体智能辅导系统实施的框架；(3)为NLP模型训练和对话界面开发准确、低延迟的隐私保护机制。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 award is funded in whole or in part under the American Rescue Plan Act of 2021 (Public Law 117-2).This project will implement a computer tutor for low literacy Hispanic breast cancer survivors. Breast cancer is the leading cause of cancer-related deaths in Hispanics, and although research has shown that education can greatly mitigate stress and improve quality of life, few educational interventions for this population exist. The computer tutor resulting from this project will mimic a human tutor that teaches about breast cancer survivorship skills and about breast cancer in general. Because tutoring involves conversation with the survivor, it is possible that they reveal sensitive personal information; therefore, it is important to encrypt the information in the tutoring session to prevent any privacy breaches. And for the tutoring component to be effective, special attention must be paid to the utilization of natural language processing models as well as models of behavior that are observed in the target population when tutoring and/or interacting with technology. For the privacy component to be effective, techniques that can encrypt and decrypt data at high speeds will be explored to make the interaction fluid. To date, computer tutors that converse with their students have been tried mainly with a highly literate population in college settings, so their impact on low literacy Hispanics is unknown. Therefore, this project will advance our understanding of the impact of designing artificial intelligence powered tutors to address diversity and disparities in the access to information by a subset of low literacy individuals, as well as our understanding of privacy preserving algorithms that work in real-time with complex natural language processing models. More broadly, project outcomes will facilitate access to information for minority populations and will serve to build research capacity and train minority students in the participating teaching-oriented institutions.The project will be carried out with two objectives in mind. First, development of a novel intelligent computer tutoring system that is customized so that it can effectively query and interact with Hispanic breast cancer survivors by adapting existing content that was created for this population in prior research. Because it has been shown that both the language of many adult Hispanics, and target population interactions with technology, are more nuanced than previously thought, our first objective also involves training natural language algorithms and designing interactions that model those of Hispanic breast cancer survivors. The second objective is to develop privacy-preserving algorithms that utilize robust end-to-end encrypted communication and can encrypt and decrypt distributed data in real time at a speed that does not hinder the interactions with the computer tutor. The contributions of this development process will be threefold: (1) to understand the role of culture and education in the interaction between low literacy Hispanic breast cancer survivors and intelligent tutoring systems; (2) to develop a framework that facilitates the implementation of intelligent tutoring systems for minority populations; and (3) to develop accurate and low latency privacy preserving mechanisms for NLP model training and dialogue interfaces.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.