Collaborative Research: IMR: MM-1B: Automating Privacy-Preserving Data Sharing of Campus Network Traffic Logs

互联网测量数据在实现网络和网络安全监测等领域的创新方面发挥着关键作用。然而，这些数据通常包含敏感的用户信息，这些信息不能公开共享，只有少数与数据所有者建立了合作关系的受信任的研究人员才能访问。这一限制阻碍了研究结果的可重复性和研究界对收集数据的使用。该项目开发了一种系统的方法，以促进弗吉尼亚大学（UVA）和弗吉尼亚理工大学（VT）与更广泛的研究界共享校园网流量日志的数据。该项目的新颖之处在于：(i)数据匿名化：自动发现和匿名化网络流量日志中的敏感数据；（ii）数据共享：通过隐私保护框架模块化数据访问。该项目更广泛的意义和重要性在于(i)扩大了可以使用UVA和VT校园网流量日志的研究人员的数量，并作为其他校园和网络共享互联网测量数据的指针；（ii）课程开发，以吸引研究生和本科生对校园网日志进行研究活动。具体来说，该项目(i)开发了一个可扩展和可适应的模型，用于识别个人身份信息（PII），以便在离线设置中对结构化和半结构化网络日志进行匿名化，根据历史日志对模型进行微调，并将模型转换为能够支持实时数据源和离线存储库的可调服务；（ii）根据研究需要将不同级别的数据访问模块化，并构建具有不同隐私的隐私保护框架，为受限制级别的数据访问提供可证明的隐私保障；(3)将隐私保护框架集成回数据匿名化中，实现基于使用模式的动态匿名化。该项目在实际的现实环境中直接在UVA和VT流量收集管道上部署和评估所建议的方法。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。

Internet measurement data serves a critical role in enabling innovations in fields such as networking and cyber security monitoring. However, such data often contains sensitive user information, which cannot be shared publicly and is only accessible to a limited number of trusted researchers with established collaborations with data owners. This limitation hinders the reproducibility of research results and the use of collected data by the research community. This project develops a systematic approach to facilitate data sharing of campus network traffic logs from the University of Virginia (UVA) and Virginia Tech (VT) with the broader research community. The project’s novelties are (i) data anonymization: automating the discovery and anonymization of sensitive data in network traffic logs and (ii) data sharing: modularizing the data access through a privacy-preserving framework. The project's broader significance and importance are (i) expanding the number of researchers who can use the campus network traffic logs from UVA and VT, as well as serving as a pointer for sharing Internet measurement data by other campuses and networks; (ii) curriculum development to engage graduate and undergraduate students for research activities on the campus network logs. Specifically, the project (i) develops an extensible and adaptable model to identify Personal Identifiable Information (PII) to anonymize in structured and semi-structured network logs in an offline setting, fine tune the model based on historical logs, and transition the model to a tunable service capable of supporting live data feeds and offline repositories; (ii) modularizes varying levels of data access based on research needs and constructs a privacy-preserving framework with differential privacy to provide provable privacy guarantees to restricted levels of data access; and (iii) integrates the privacy-preserving framework back into data anonymization to achieve dynamic anonymization based on usage patterns. This project deploys and evaluates the proposed approach directly on the UVA and VT traffic collection pipeline in a practical real-world setting.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.