CSR: Medium: Highly Scalable and Accurate System Support for Detecting Misbehaving Users and Mitigating Criminal Activities in Realtime Online Video-Based Services

点击翻译按钮获取中文摘要

The main goal of this proposal is to understand the safety problems introduced by realtime online video-based services, and to develop highly scalable and accurate solutions and systems that protect the safety of interactive video users. While realtime, online video-based services are newly emerging services, they have already gained tremendous popularity with millions of users. These services introduce new ways for people to interact and have the potential to significantly impact, even revolutionize, pervasive computing and online social networking services. Unfortunately, the presence of misbehaving users may jeopardize the entire utility of these services and bring harm to minors. This proposal investigates innovative ideas including fine-grained cascaded classification and a rule-based pre-classifier with logistic regression to improve scalability while preserving accuracy, incorporating contextual information from smartphone sensors and application scenarios to improve accuracy, and mechanisms to evaluate and circumvent cyber-bullying. The complete system will be deployed and evaluated using real-world data. With the growing popularity of realtime, online video-based services and the threat to minors of being exposed to unsuitable content or cyber-bullying attacks, developing effective techniques for detecting misbehaving users of these services will have significant and far-reaching beneficial impacts on society. All software developed from this project will be released as open source. The project will develop and integrate educational course modules on Internet safety topics into relevant middle and high school courses via an on-going NSF GK-12 project. Undergraduate and minority students will be recruited via the Colorado Diversity Institute and annual REU supplements.