SaTC: CORE: Small: Deconstructing and Neutralizing Search Rank Fraud

搜索排名欺诈，或为在线服务中托管的产品发布大量付费活动，旨在给人一种草根参与的错觉，并提高经济收益，推广恶意软件，甚至协助审查工作。经过服务提供商和学术界多年的投资，搜索排名欺诈仍然是一个严重的问题。缺乏进展的部分原因是对专业评分员群体的不完整和不准确的理解。这个项目建立在这样一个论点的基础上，即为了有效，欺诈检测和预防工作需要涉及那些对搜索排名欺诈做出贡献的组织和个人。为此，该项目正在设计和开发技术，以验证专业评分员，并研究他们的能力，行为和检测避免策略。该项目引入了新一代解决方案，以解决和消除搜索排名欺诈的影响。该项目将帮助保护数以百万计的在线服务用户免受误导信息、不合格产品甚至恶意软件和审查应用程序的影响。该项目引入了一种新的方法来研究、检测和预防在线服务中的欺诈行为，并评估开发的解决方案。该团队正在构建一个平台、技术和协议，以收集有关专业评分员和同行意见服务诚实用户的在线和移动终端行为的数据，并使用实时评分员评估开发的解决方案。此外，该小组正在设计欺诈生成器，以生成欺诈活动的合成但现实的时间表的基准数据集。该团队正在使用对抗学习方法来迭代训练神经网络，以将不同类型的欺诈与诚实行为区分开来。最后，该团队将利用从收集到的数据中提取的独特见解，沿着网络表示学习和图形划分技术，开发出可归因于欺诈的预测器，并识别由同一组织控制的用户帐户。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。

Search rank fraud, or posting large numbers of paid activities for products hosted in online services, intends to give the illusion of grassroots engagement and to boost financial gains, promote malware and even assist censorship efforts. Search rank fraud continues to be a significant problem after years of investment by service providers and the academic community. This lack of progress is partly due to an incomplete and inaccurate understanding of the diverse community of professional raters. This project builds on the thesis that to be effective, fraud detection and prevention efforts need to involve the organizations and individuals who contribute to search rank fraud. To this end, this project is designing and developing techniques to validate professional raters and to study their capabilities, behaviors and detection avoidance strategies. The project introduces a new generation of solutions to address and neutralize the effects of search rank fraud. This project will help protect millions of online service users from misleading information, substandard products and even malware and censorship-enabling apps.This project introduces a novel approach to study, detect and prevent fraud in online services, and to evaluate developed solutions. The team is building a platform, techniques, and protocols to collect data about the online and mobile device behaviors of professional raters and honest users of peer-opinion services, and to evaluate developed solutions using live raters. Also, the team is designing fraud generators to produce benchmark datasets of synthetic but realistic timelines of fraudulent activities. The team is using an adversarial learning approach to iteratively train neural networks to disentangle different types of fraud from honest behaviors. Finally, the team will leverage unique insights extracted from the collected data, along with network representation learning and graph partitioning techniques, to develop predictors that attribute fraud and identify user accounts controlled by the same organization.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.