CRII: III: Learning networks from strategic decisions: enabling network intervention and revealing social privacy risks without structural information

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。数字平台和基于云的产品的兴起改变了企业运营和消费者决策。加强的连接提高了动态定价、营销活动和数字平台上大规模行为变化的有效性。在Yelp和Google Review等众包评论平台上，公开的评分和评论减少了消费者信息搜索摩擦;在亚马逊和阿里巴巴等电子商务零售商上，有关产品的信息加强了消费者对产品的信任。将机器学习方法和丰富的行为数据相结合，可以改善消费者体验，增加收入，并指导制造商和零售商的定价策略。尽管个人之间的联系在广泛的应用中是强大的，但由于各种原因，它并不总是可用的：（1）网络数据收集成本高;（2）网络数据可能过于敏感和机密，无法共享;（3）网络数据是动态的，静态数据的准确性可能会随着时间的推移而衰减。与此同时，社交互动和数据整合的加强带来了隐私风险，导致系统层面的数据安全问题。本项目为基于大规模行为数据的社交网络结构学习奠定了基础。两个反向相关的问题激发了本研究：1）如何设计网络干预措施，以利用社会外部性时，结构性数据不可用？2)公开的行为数据是否会在泄露社交网络信息时带来社交隐私风险？本项目研究基于观测决策学习网络结构和近似效用函数的一般和基本问题。本计画首先分析此网路学习问题的基本可辨识性与条件可辨识性，借由在策略决策者网路中建立一线性二次网路赛局结构。它将把这种线性二次博弈扩展到更广泛的网络博弈。该项目进一步开发了一个生成式深度学习框架，以近似社交网络上的人类决策过程。最后，研究人员将通过将理论和方法应用于网络干预（例如，信息扩散和影响最大化）和隐私风险评估（例如，影子剖析和社会攻击）使用三个大规模的数字行为数据和两个小规模的实验数据。这个跨学科的项目建立在博弈论、机器学习、网络科学和管理科学的基础上，并为之做出了贡献。该奖项反映了NSF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).The rise of digital platforms and cloud-based products changed business operations and consumer decision-making. The bolstered connections improve the effectiveness of dynamic pricing, marketing campaigns, and large-scale behavioral change on digital platforms. On crowdsourced review platforms such as Yelp and Google Review, publicly ratings and reviews reduce consumer information search friction; on E-commerce retailers such as Amazon and Alibaba, information about products strengthens consumers' trust in the products. Integrating machine learning methods and rich behavioral data improve consumers' experience, accrues revenues, and guides manufacturers' and retailers' pricing strategies. Even though connections among individuals are powerful in a wide range of applications: it is not always available for various reasons: (1) network data is costly to collect; (2) network data may be too sensitive and confidential to share; (3) network data is dynamic and static data's accuracy may decay over time. In the meantime, strengthened social interaction and data integration pose privacy risks, leading to data security issues at a systematic level.This project lays the groundwork for learning social network structures based on large-scale behavioral data. Two inversely-related issues motivate this research: 1) How to design network interventions to leverage social externality when structural data is unavailable? 2) Does publicly available behavioral data pose social privacy risk in leaking social network information? This project studies the general and fundamental problem of learning the network structures and approximating utility functions based on observed decisions. This project first analyzes this network learning problem's fundamental and conditional identifiability by imposing a linear-quadratic network game structure in a network of strategic decision-makers. It will extend this linear-quadratic game to a broader class of network games. This project further develops a generative deep learning framework to approximate human decision-making processes on social networks. Finally, the researchers will demonstrate the practical value of this research by applying the theory and approach to network intervention (e.g., information diffusion and influence maximization) and privacy risk evaluations (e.g., shadow profiling and social attack) using three large-scale digital behavioral data and two small-scale experimental data. This interdisciplinary project builds upon and contributes to game theory, machine learning, network science, and management science.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.